This is the AdaControl User Guide. It describes how to install and use AdaControl. Please refer to the AdaControl Programmer Manual to learn how to add new kinds of rules to AdaControl.
Last edited: 3 août 2010
AdaControl is Copyright © 2005-2010 Eurocontrol/Adalog, except for some specific modules that are © 2006 Belgocontrol/Adalog, © 2006 CSEE/Adalog, or © 2006 SAGEM/Adalog. AdaControl is free software; you can redistribute it and/or modify it under terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version. This unit is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License distributed with this program; see file COPYING. If not, write to the Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
As a special exception, if other files instantiate generics from this program, or if you link units from this program with other files to produce an executable, this does not by itself cause the resulting executable to be covered by the GNU General Public License. This exception does not however invalidate any other reasons why the executable file might be covered by the GNU Public License.
This document is Copyright © 2005-2010 Eurocontrol/Adalog. This document may be copied, in whole or in part, in any form or by any means, as is or with alterations, provided that (1) alterations are clearly marked as alterations and (2) this copyright notice is included unmodified in any copy.
AdaControl is an Ada rules controller. It is used to control that Ada software meets the requirements of a number of parameterizable rules. It is not intended to supplement checks made by the compiler, but rather to search for particular violations of good-practice rules, or to check that some rules are obeyed project-wide. AdaControl can also be handy to make statistics about certain usages of language features, or simply to search for the occurrences of particular constructs; its scope is therefore not limited to enforcing programming rules, although it is of course one of its main goals.
Commercial support is available for AdaControl, see file
doc/support.txt. If you plan to use AdaControl for industrial
projects, or if you want it to be customized or extended to match your
own needs, please contact Adalog at
info@adalog.fr.
AdaControl analyzes a set of Ada units, according to parameterizable controls. Controls can be given from the command line, from a file, or interactively. There is a wide range of controls available. Some are quite simple (although very useful):
AdaControl is very simple to use. It takes, as parameters, a list of units to process and a list of commands that define the controls to apply. The complete syntax of the commands is described in chapter Command language reference.
AdaControl produces messages to the standard output, unless redirected. Several levels of messages are defined (i.e. error or found), depending on the kind of the control (i.e. check or search).
Rules can be locally disabled for a part of the source code, and various options can be passed to the program.
Ex:
Given the following package:
package Pack is
pragma Pure (Pack);
...
end Pack;
The following command:
adactl -l "search pragmas (pure)" pack
produces the following result (displayed to standard output):
pack.ads:2:4: Found: PRAGMAS: use of pragma Pure
AdaControl integrates nicely in environments such as GPS (see Running AdaControl from GPS), AdaGide (see Running AdaControl from AdaGide), or emacs (see Control kinds and report messages). In those environments, you can run AdaControl from menus or by just clicking on a button!
The development of AdaControl was initially funded by Eurocontrol (http://www.eurocontrol.int), which needed a tool to help in verifying the million+ lines of code that does Air Traffic Flow Management over Europe. Because it was felt that such a tool would benefit the community at-large, and that further improvements made by the community would benefit Eurocontrol, it was decided to release AdaControl as free software. Later, Eurocontrol, Belgocontrol, Ansaldo (formerly CSEE-Transport), and SAGEM-DS sponsored the development of more rules.
The requirements for AdaControl were written by Philippe Waroquiers (Eurocontrol-Brussels), who also conducted extensive testing of AdaControl over the Eurocontrol software. The software was developped by Arnaud Lecanu and Jean-Pierre Rosen (Adalog). Rules, improvements, etc. were contributed by Pierre-Louis Escouflaire (Adalog), Alain Fontaine (ABF consulting), Richard Toy (Eurocontrol-Maastricht), and Isidro Ilasa Veloso (GMV). AdaGide support and improvement of icons were contributed by Gautier de Montmollin. Emmanuel Masker (Alstom) and Yannick Duchene contributed to GPS integration.
See file HISTORY for a description of the various versions of
AdaControl, including enhancements of the current version over the
previous ones. Users of a previous version are warned that the rules
are not 100% upward-compatible: this is necessary to make the rules
more consistent and easier to use. However, the incompatibilities are
straightforward to fix and should affect only a very limited number of
files. See Non upward-compatible changes for details.
Like any ASIS application, AdaControl can be run only if the compiler available on the system has exactly the same version as the one used to compile AdaControl itself. The executable distribution of AdaControl will work only with Gnat version GPL 2010, as distributed by ACT. If you are using any other version, please use the source distribution of AdaControl and compile it as indicated below.
Another reason for using the source distribution of AdaControl is that the user may not be interested in all provided rules. It is very easy to remove some rules from AdaControl to increase its speed. See Customizing AdaControl.
This section is only for the source distribution of AdaControl. If you downloaded an executable distribution (and are using Gnat GPL 2010), you may skip to the next section.
The following software must be installed in order to compile AdaControl from source:
Make sure to have the same version of GNAT and ASIS. The version used for running AdaControl must be the same as the one used to compile AdaControl itself.
Run the installer (adactl_src-setup.exe). This will
automatically build and install AdaControl, no other installation is
necessary.
Simply go to the src directory and type:
gnatmake -Pbuild.gpr
You're done!
Caveat: Due to a bug in recent versions, if you are using GnatPro 6.1.2 and above, you must set the variable GNAT_FIX to 1; i.e. invoke the command as:
gnatmake -Pbuild.gpr -XGNAT_FIX=1
The previous method may fail if Asis is not installed in an usual place. As an alternative method, it is possible to build AdaControl with a regular Makefile.
The file Makefile (in directory src) should be modified
to match the commands and paths of the target system. The following
variables are to be set:
Makefile.
See also the compilation options in this file; a change is needed if
you are using GnatPro 6.1.2 and above.
Then, run the make command:
$ cd src
$ make build
It is also possible to delete object files and do other actions with this “Makefile”, run the following command to get more information:
$ make help
NOTE: Building AdaControl needs the “make” command provide with GNAT; it works both with WIN32 shell and UNIX shell.
It should be possible to compile AdaControl with other compilers than
GNAT, although we didn't have an opportunity to try it. If you have
another compiler that supports ASIS, note that it may require some
easy changes in the package Implementation_Options to give
proper parameters to the Associate procedure of ASIS. Rules
that need string pattern matchings need the package
Gnat.Regpat. If you compile AdaControl with another compiler,
you can either port Gnat.Regpat to your system, or use a
(limited) portable implementation of a simple pattern matching
(package String_Matching_Portable). Edit the file
string_matching.ads and change it as indicated in the comments.
No other change should be necessary.
Alternatively, if you are using another compiler, you can try and compile your program with GNAT just to be able to run AdaControl. However, compilers often differ in their support of representation clauses, which can cause your program to be rejected by GNAT. In that case, we provide a sed script to comment-out all representation clauses; this can be sufficient to allow you to use AdaControl. See unrepr.sed.
Testing AdaControl needs a UNIX shell, so it works only with UNIX systems. However, it is possible to run the tests on a WIN32 system by using an UNIX-like shell for WIN32, such as those provided by CYGWIN or MSYS. To run the tests, enter the following commands:
$ cd test
$ ./run.sh
All tests must report PASSED. If they don't, it may be due to one of the following issues:
tfw_check reports “FAILED”
on some sytems, because it depends on the order in which the operating
system lists files. If this happens, try (from the test
directory):
diff res/tfw_check.txt ref/
If the only difference is that some lines are at different places, the test is OK.
If there are some rules that you are not interested in, it is very easy to remove them from AdaControl:
src directory, edit the file
framework-plugs.adb. There is a with clause for each
rule (children of package Rules). Comment out the ones you
don't want.
framework-plugs.adb. There will be error messages
about unknown procedure calls. Comment out the corresponding lines.
It is also possible to add new rules to AdaControl. If your favorite rules are not currently supported, you have several options:
All you need to run AdaControl is the executable named adactl
under Linux or adactl.exe under Windows. In addition,
pfni (or pfni.exe under Windows) is a convenient
utility, required by the GPS support. See pfni.
If you downloaded the Windows installer executable version of
AdaControl, simply run adactl_exe-setup.exe. This will install
all the files in the recommended locations (as has been done with the
Windows installer source version), including GPS support if you have
GPS installed and/or AdaGide support if you have AdaGide installed.
If you built AdaControl from source without an installer, the
executables are in the src directory of the distribution. If
you downloaded an executable distribution, they are in the root
directory of the distribution. Copy the executables to any convenient
directory on your path; a good place, for example, is in the
bin directory of your Gnat installation.
Integration of AdaControl into GPS with all functionalities requires GPS version 4.2 or above (delivered since GNAT/GPL2008).
To add AdaControl support to GPS, copy the file
GPS/adacontrol.xml into the <GNAT_dir>/share/gprconfig
directory; copy all other files from the GPS directory into the
<GPS_dir>/share/gps/plug-ins directory. Copy also HTML files
from the doc directory into the
<GPS_dir>/share/doc/gps/html to access AdaControl's guides from
the "Help" menu of GPS.
To add AdaControl support to AdaGide, copy the file
AdaControl.tdf from the AdaGide directory into AdaGide's
root directory. Note that AdaControl support requires AdaGide version
7.42 or above.
AdaControl is a command-line program, i.e. it is normally called directly from the system shell. Options are introduced by a “-” followed by a letter and can be grouped as usual. Some options take the following word on the command line as a value; such options must appear last in a group of options. Parameters are words on the command line that stand by themselves. Options and parameters can be given in any order.
The syntax for invoking AdaControl in regular mode is:
adactl [-deEirsTuvwx]
[-p <project file>] [-f <rules file>] [-l <rules list>]
[-o <output file>] [-t <trace file>] [-F <format>]
[-S <statistics level>] [-m <warning limit>] [-M <message limit>]
{<unit>[+|-<unit>]|[@]<file>} [-- <ASIS options>]
AdaControl can process only Ada-95, not Ada-2005, since there no ASIS
for Ada-2005 yet. If you are using a version of GNAT where Ada-2005
is the default (especially GNAT-GPL), and in the rare cases where your
program would not compile in Ada-2005 mode (notably if you have a
function that returns a task type), you must force Ada-95 mode by
having a “gnat.adc” file that contains a pragma Ada_95,
since the corresponding option cannot be passed to the compiler in
“compile on the fly” mode. Alternatively, you can generate the
tree files manually (see Generating tree files manually) with the
“-gnat95” option.
Units to be processed are given as parameters on the command
line. Note that they are Ada compilation unit names, not
file names: case is not significant, and there should be no
extension! Child units are allowed following normal Ada naming rules:
Parent.Child, but be aware that specifying a child unit will
automatically include its parent unit in the analysis. Subunits are
processed during the analysis of the including unit; there is
therefore no need to specify subunits explicitely. If you do specify a
subunit explicitly, it will result in the whole enclosing unit being
analyzed.
However, as a convenience to the user, units can be specified as file names, provided they follow the default GNAT naming convention. More precisely, if a parameter ends in “.ads” or “.adb”, the unit name is extracted from it (and all “-” in the name are substituted with “.”). File names can include a path; in this case, the path is automatically added to the list of directories searched (“-I” ASIS option). The file notation is convenient to process all units in a directory, as in the following example:
adactl -f my_rules.aru *.adb
In the unlikely case where you have a child unit called Ads or
Adb, use the “-u” option to force interpretation of all
parameters as unit names.
By default, both the specification and body of the unit are processed; however, it is possible to specify processing of the specification only by providing the “-s” option. If only file names are given, the “-s” option is assumed if all files are specifications (“.ads” files). It is not possible to specify processing of bodies only, since rules dealing with visibility would not work.
The “-r” option tells AdaControl to process (recursively) all user units that the specified units depend on (including parent units if the unit is a child unit or a subunit). Predefined Ada units and units belonging to the compiler's run-time library are never processed.
Ex:
adactl -r -f my_rules.aru my_main
will process my_main and all units that my_main depends
on. If my_main is the main procedure, this means that the whole
program will be processed.
It is possible to specify more than one unit (not file) to process in a parameter by separating the names with “+”. Conversely, it is possible to specify units that are not to be processed, separated by “-”. When a unit is subtracted from the unit list, it is never processed even if it is included via the recursive option, and all its child and separate units are also excluded. This is convenient to avoid processing reusable components, that are not part of a project. For example, if you want to run AdaControl on itself, you should use the following command:
adactl -f my_rules_file.aru -r adactl-asis-a4g
This applies the rules from the file my_rules_files.aru to
AdaControl itself, but not to units that are part of ASIS (units
Asis, A4G, and their children) that would be found by
the “-r” (recursive) option otherwise.
Alternatively, it is possible to provide units indirectly with a parameter consisting of an “@” followed by the name of a file. This file must contain a list of unit names (not files), one on each line. All units whose names are given in the file will be processed. If a name in the file starts with “@”, it will also be treated as an indirect file (i.e. the same process will be invoked recursively). If a line in the file starts with “#” or “--”, it is ignored. This can be useful to temporarily disable the processing of some files or to add comments.
Ex:
adactl -f my_rules.aru @unit_file.txt
Commands specify which processing AdaControl should apply to units. See Command language reference for a detailed description of all commands.
Commands can be given directly on the command line with the “-l” option. A commands list must be quoted with “"”.
Ex:
adactl pack.ads proc.adb -l "check instantiations (My_Generic);"
It is possible to pass several commands separated by “;”, but as a convenience to the user, the last “;” may be omitted.
Commands can also be read from a file, whose name is given after the
“-f” option (the “.aru” extension is taken by default). As
a special case, if the file name is “-”, commands are read from the
standard input. This is intended to allow AdaControl to be pipelined
behind something that generates commands; if you want to type commands
directly to AdaControl, the interactive mode is more
appropriate. See Interactive mode.
Ex:
adactl -f my_rules.aru proc.adb
Note that the “-l” and “-f” options are not exclusive: if both are specified, the commands to be performed include those in the file (first) and then those given on the command line.
Messages produced by controls are output to the output file; by default, it is the standard output, but it can be changed by specifying the “-o” option.
Ex:
adactl -f my_rules.aru -o my_output.txt proc.adb
If the output file exists, new messages are appended to it. This allows running AdaControl under several directories that make up the project, and gathering the results in a single file. However, if the “-w” option is given, AdaControl overwrites the output file if it exists.
All other messages, including syntax error messages, units processed (in verbose mode), and possible internal error mesages from AdaControl itself are output to the standard error file.
The “-F” option selects the output format. It must be followed by “Gnat”, “Gnat_Short”, “CSV”, “CSV_Short”, “Source”, “Source_Short”, or “None” (case insensitive). By default, the output is in “Gnat” format. See Control kinds and report messages for details.
The “-S” option selects which statistics are output after each run. It must be followed by a value in the range 0..3. See Control kinds and report messages for details on the various statistics levels.
The “-T” option prints a summary of timing at the end of the run. This indicates how long (in real-time seconds) was spent in processing each rule.
Ex:
adactl -F CSV -S 2 -f my_rules.aru -o my_output.csv proc.adb
The “-m” and “-M” options are used to limit the output of AdaControl. These options are followed by an integer value that specifies the maximum number of error messages (“-m”) or warning and error messages (“-M”). If the value is omitted, a previous limitation (comming for example from a command file) is cancelled.
If the indicated number of messages is exceeded during a run, AdaControl stops immediately.
An emacs project file (the file with a “.adp” extension used by the Ada mode of Emacs) can be specified with the “ -p” option. AdaControl will automatically consider all the directories mentioned in “src_dir” lines from the project file.
Ex:
adactl -f my_rules.aru -p proj.adp proc.adb
Note that AdaControl does not accept “.gpr” project files, because ASIS does not currently accept the “-P” option like other Gnat commands do. However, when run from GPS, the interface will automatically use the source directories from the current (root) project (unless you have explicitely set a “.adp” file in the switches AdaControl switches).
If you have a project that uses “.gpr” project files and you want to run AdaControl from the command line (not from GPS), you can generate a “.adp” project file from a “.gpr” project file from within GPS, by using the “Tools/AdaControl/Generate .adp project” menu. See Running AdaControl from GPS. Alternatively, it is also possible to use GPS project files by generating the tree files manually. see Generating tree files manually for details.
The “-i” option tells AdaControl to ignore disabling markers in Ada source code (see Disabling controls); i.e. all controls will be performed, regardless of the presence of disabling markers.
Note that if you you have many messages, setting this option can speed-up AdaControl considerably. It is therefore advisable to always set this option when you know that there is no disabling marker in your source code.
In the default mode, AdaControl displays only messages from triggered controls. It is possible to get more information with the verbose option (“-v”). In this mode, AdaControl displays a a progress indicator and unit names as they are processed, and its global execution time when it finishes. Note that the progress indicator includes an indication of the run number if there are more than one “go” command.
The “-d” option enables debug mode. This mode provides more information in case of an internal program error, and is of little interest for the casual user. In this mode, AdaControl may also, in rare occasions (and only with some versions of Gnat), display ASIS “bug boxes”; this does not mean that something went wrong with the program, but simply that an ASIS failure was properly recovered by AdaControl.
Output of the messages printed by the “-d” option can be directed to a “trace” file (instead of being printed to the standard error file). This is done by the “-t” option, which must be followed by the file name. If the trace file exists, new messages are appended to it.
The “-e” option tells AdaControl to treat warnings as errors, i.e. to report a return code of 1 even if only “search” controls were triggered. See Return codes. It does not change the messages however.
Conversely, the “-E” option tells AdaControl to not report warnings at all, i.e. only errors are reported. However, if you ask for statistics, the number of warning messages is still counted. See Control kinds and report messages.
If an internal error is encountered during the processing of a unit, AdaControl will continue to process other units. However, if the “-x” option is given, AdaControl will stop on the first error encountered. This option is mainly useful if you want to debug AdaControl itself (or your own rules). See In case of trouble.
Ex:
adactl -x -f my_rules.aru proc.adb
Everything that appears on the command line after “--” will be treated as an ASIS option, as described in the ASIS user manual.
Casual users don't need to care about ASIS options, except in one case: if you are running AdaControl from the command line (not from GPS), and if the units that you are processing reference other units whose source is not in the same directory, AdaControl needs to know how to access these units (as GNAT would). This can be done either by using an Emacs project file with the “-p” option (see Project files), by putting the appropriate directories into the ADA_INCLUDE_PATH environment variable, or by passing “-I” options to ASIS.
It is possible to pass one or several “-I” options to ASIS, to provide other directories where sources can be found. The syntax is the same as the “-I” option for GNAT.
Other ASIS options, like the “-Cx” and/or “-Fx” options, can be specified. Most users can ignore this feature; however, specifying these options can improve the processing time of big projects. See Optimizing Adacontrol.
In order to ease the automation of controlling programs with shell scripts, AdaControl returns various error codes depending on how successful it was. Values returned are:
If the environment variable “ADACTLINI” is set, its content is taken as a set of commands (separated by semi-colons) that are executed before any other command. Although any command can be specified, this is intended to allow changing default settings with “set” commands. See Set command.
For example, you can set ADACTLINI to “set format Gnat_Short” if you prefer having you messages in short format rather than the (default) long format.
The “-I” option tells AdaControl to operate interactively. In this mode, commands specified with “-l” or “-f” options are first processed, then AdaControl prompts for commands on the terminal. Note that the “quit” command (see Quit command) is used to terminate AdaControl.
The syntax of commands run interactively is exactly the same as the one used for files; especially, each command must be terminated with a “;”. Note that the prompt (“Command:”) becomes “.......:” when AdaControl requires more input because a command is not completely given, and especially if you forget the final “;”.
As with files, it is possible to give several commands on a single line in interactive mode. If a command contains syntax errors, all “go” commands (see Go command) on the same line are temporarily disabled. Other commands that do not have errors are normally processed however.
The interactive mode is useful when you want to do some analysis of your code, but don't know beforehand what you want to control. Since the ASIS context is open only once when the program is loaded, queries will be much faster than running AdaControl entirely with a new query given in a “-l” option each time. It is also useful to experiment with AdaControl, and to check interactively commands before putting them into a file.
In addition to normal usage, AdaControl features special options to ease its use; no Ada unit is analyzed when using these options.
The “-h” option provides help about Adacontrol usage. If the “-h” option is given, no other option is analyzed and no further processing happens.
Syntax:
adactl -h [<keyword> | <rule name>...]
<keyword> ::= all | commands | license | list | options | rules | version
The “-h” option without parameter displays a help message about usage of the AdaControl program, the various options, and the rule names.
Otherwise, the “-h” must be followed by one or several keywords or rule names (case irrelevant); its effect is:
Ex:
adactl -h pragmas Unnecessary_Use_Clause
adactl -h all
adactl -h version license
The “-C” option is used to check syntax of commands without executing any control.
Syntax:
adactl -C [-dv] [-f <rules file>] [-l <rules list>]
In this mode, AdaControl simply checks the syntax of the commands provided with the “-l” option, or of the commands provided in the file named by the “-f” option (at least one of these options must be provided). No other processing will happen.
AdaControl will exit with a return code of 0 if the syntax is correct, and 2 if any errors are found. A confirming message that no errors were found is output if the “-v” option is given.
This option is especially useful when you have modified a rules file, before trying it on many units. The way AdaControl works, it must open the ASIS context (a lengthy operation) before analyzing the rules. This option can therefore save a lot of time if the rules file contains errors.
The “-D” options produces a list of units that can be reused as an indirect file in later runs. Syntax:
adactl -D [-rsvw] [-o <output file>] [-p <project file>]
{<unit>[+|-<unit>]|[@]<file>} [-- <ASIS options>]
In this mode, AdaControl outputs the list of units that would be processed. It is especially useful when used with the “-r” option and given the main unit name, since it will then generate the whole list of dependent units (hence the name “D”).
This list can be directed to a file with the “-o” option (if the file exists, it won't be overwritten unless the “-w” option is specified). This file can then be used in an indirect list of units. See Input units. Note that it is more efficient to create the list of units once and then use the indirect file than to specify all applicable units or use the “-r” option each time AdaControl is run.
If you want to use AdaControl from GPS, make sure you have copied the necessary files into the required places. See Installing AdaControl.
AdaControl integrates nicely into GPS, making it even easier to use. It can be launched from menu commands, and parameters can be set like any other GPS project parameters. When run from within GPS, AdaControl will automatically retrieve all needed directories from the current GPS project.
After running AdaControl, the “locations” panel will open, and you can retrieve the locations of errors from there, just like with a regular compilation. Errors will be marked in red in the source, warning will be marked orange, and you will have corresponding marks showing the places of errors and warnings in the speedbar. Note that AdaControl errors appear under the “AdaControl” category, but if there were compilation errors, they will appear under the “Compilation” category. Final counts from “count” control kinds will appear under the “Counts summary” category, and statistics under the “Statistics” category.
GPS now features an “AdaControl” menu, with several submenus:
There are also two buttons representing Lady Ada in a magnifier glass in the toolbar, one with a red question mark in the background. These buttons launch AdaControl, by default on the file currently being edited; however, you can change this behaviour from the preferences to control either files from a list, or all files from the project. The button without the question mark uses rules from the current rules file, while the one with the question mark asks for the control to apply interactively.
Here are some tips about using the “interactive” menus (or the button with the question mark):
AdaControl adds two entries to the contextual menus (right click) of
Ada files. They call the pfni utility on the current
entity. See pfni. The entry “Print full name” displays the full
name of the entity in simple form, while the entry “Print full name
(with overloading)” ) prints it with overloading information. If the
name refers to an entity which is initialized (or to a parameter with
a default value), and the initial value is static, the name is followed
by this value in parenthesis.
This is convenient to find how to name entities in rule files. See Specifying an Ada entity name. It is also convenient to find where an entity is declared, and which of several overloaded entities is being referred to.
This is also convenient to find the actual value of a constant from anywhere in the program text, since the printed value is completely evaluated if it is a (static) expression.
The tab “switches” from the “Project/Edit Project Properties” menu includes a page for AdaControl, which allows you to set various parameters. Since the GPS interface analyzes the output of AdaControl, you should not set options directly in the bottom window of this page (the one that displays the actual options passed to AdaControl).
This section controls the definition of various files used by AdaControl.
This section offers options that control how units are processed.
This section controls the debugging options of AdaControl.
This section offers options that control where and how the output of AdaControl is displayed.
This section controls the ASIS parameters passed to AdaControl. The content of the input field “ASIS options” is used in place of the standard (“-CA -FM”) one.
Casual users don't need to change the default ASIS options. For more details, see ASIS options.
There is an entry for AdaControl in the “edit/preferences” menu:
If you check “AdaControl” in the “Languages” tab, GPS
will recognize files with extension .aru as AdaControl rules
files, and provide appropriate colorization.
The AdaControl User Manual (this manual) and the AdaControl Programmer Manual are available from the "Help/AdaControl" menu of GPS.
The "Help on rule" entry displays the list of all rules; if you click on one of them, you get help for the particular rule. Depending on the setting of the “Help on rule” preference (see above), it opens a pop-up that displays the rule(s) purpose and the syntax of its parameters, or opens the user guide at the appropriate location.
The “About” entry displays a popup with AdaControl's version number and license condition.
GPS may crash when the output of a command is too big (i.e. hundreds of messages with AdaControl). If this happens, use the “preferences” menu to limit the number of messages.
If you want to use AdaControl from AdaGide, make sure you have copied
the necessary file into the required place. See Installing AdaControl. Note that AdaGide does not have all the parameterization
facilities of sophisticated environments like GPS, but all AdaControl
options, like the name of the rules file or the output format, can
easily be changed by editing the tool description file
AdaControl.tdf.
AdaGide now features several AdaControl commands from the “tool” menu:
verif.aru.
This section describe utilities that are handy to use in conjunction with AdaControl.
The convention used to refer to entities (as described in Specifying an Ada entity name) is very powerful, but it may be difficult to spell out correctly the name of some entities, especially when using the overloaded syntax.
pfni (which stands for Print Full Name Image) can be used
to get the correct spelling for any Ada entity. The syntax of
pfni is:
pfni [-sofdq] [-p <project-file>] <unit>[:<span>]
[-- <ASIS options>]
<span> ::= <line_number>
| [<first_line>]-[<last_line>]
| <line_number>:<column_number>
or
pfni -h
If called with the “-h” option, pfni prints a help message
and exits.
Otherwise, pfni prints the full name image of all identifiers
declared in the indicated unit, unless there is a “-f” (full)
option, in which case it prints the full name image of all identifiers
(i.e. including those that are used, but not declared, in the
unit). The image is printed without overloading information, unless
the “-o” option is given.
The <unit> is given either as an Ada unit, or as a file name, provided the extension is “.ads” or “.adb” (as in AdaControl). If a span is given, only identifiers within the span are printed. In the first form, the span includes only the indicated line; in the second form, the span includes all lines from <first_line> to <last_line> (if omitted, they are taken as the first and last line of the file, respectively). In the third form, the span includes only the place at the specified <line_number> and <column_number>.
Normally, the source line corresponding to the names is printed above the names. The “-q” (quiet) option suppresses this.
If the “-s” option is given (or the unit is a file name with a “.ads” extension), the specification of the unit is processed, otherwise the body is processed. The “-p” option specifies the name of an Emacs project file, and the “-d” option is the debug mode, as for AdaControl itself. ASIS options can be passed, like for AdaControl, after a “--” (but -FS is the default). See ASIS options.
As a side usage of pfni, if you are calling a subprogram that
has several overloadings and you are not sure which one is called, use
pfni with the “-o” option on that line: the program will tell
you the full name and profile of the called subprogram.
This file (provided in the “src” directory) is a sed script that transforms a text file into a set of correponding regular expressions. It is useful to generate model header files. See Header_Comments.
This file (provided in the “src” directory) is a sed script that comments out all representation clauses. It is typically useful if you use a different compiler that accepts representation clauses not supported by GNAT.
Typically, you would copy all your sources in a different directory, copy “unrepr.sed” in that directory, then run:
sed -i -f unrepr.sed *.ads *.adb
You can now run AdaControl on the patched files. Of course, you won't be able to check rules related to representation clauses any more...
Note that the script adds “--UNREPR ” to all representation clauses. Its effect can thus easily be undone with the following commad:
sed -i -e "s/--UNREPR //" *.ads *.adb
There are many factors that may influence dramatically the speed of AdaControl when processing many units. For example, on our canonical test (same controls, same units), the extreme points for execution time were 111s. vs 13s.! Unfortunately, this seems to depend on a number of parameters that are beyond AdaControl's control, like the relative speed of the CPU to the speed of the hard-disk, or the caching strategy of the file system.
This section will give some hints that may help you increase the speed of AdaControl, but it will not change the output of the program; you don't really need to read it if you just use AdaControl occasionnally. This section is concerned only with the GNAT implementation of ASIS; other implementations work differently.
Bear in mind that the best strategy depends heavily on how your program is organized, and on the particular OS and hardware you are using. Therefore, no general rule can be given, you'll have to experiment yourself. Hint: if you specify the “-v” option to AdaControl, it will print in the end the elapsed time for running the tests; this is very helpful to make timing comparisons.
Note: all options described in this section are ASIS options, i.e. they must appear last on the command line, after a “--”.
Since AdaControl is an ASIS application, it is useful to explain here how ASIS works. ASIS (and therefore AdaControl) works on a set of units constituting a “context”. Any reference to an Ada entity which is not in the context (nor automatically added, see below) will be ignored; especially, if you specify to AdaControl the name of a unit which is not included in the current context, the unit will simply not be processed.
ASIS works by exploring tree files (same name as the corresponding Ada unit, with a “.adt” extension), which are “predigested” views of the corresponding Ada units. By default, the tree files are generated automatically when needed, and kept after each run, so that subsequent runs do not have to recreate them.
A context in ASIS-for-Gnat is a set of tree files. Which trees are part of the context is defined by the “-C” option:
The “-F” option specifies what to do if the program tries to access an Ada unit which is not part of the context:
The default combination used by AdaControl is “-CA -FM”.
It is also possible to generate the tree files manually before running AdaControl. Although this mode of operation is less practical, it is recommended by AdaCore for any ASIS tool that deals with many compilation units. Some reasons why you might want to generate the tree files manually are:
To generate tree files manually, simply recompile your project with
the “-gnatct” option. This option can be passed to gnatmake
normally. Of course, you will need all other options needed by your
project (like the “-P” option if you are using GNAT project files).
Tree files may be copied into a different directory if you don't want your current directory to be cluttered by them. In this case, use the “-T” ASIS option to indicate the directory where the tree files are located.
If you chose to generate the tree files manually, you may want to specify the “-FT” ASIS option (see above) to prevent from accidental automatic recompilation.
In order to optimize the use of AdaControl, it is important to remember that reading tree files is a time-consuming operation. On the other hand, a single tree file contains not only information for the corresponding unit, but also for the specifications of all units that the given unit depends on. Moreover, our measures showed that reading an existing tree file may be slower than compiling the corresponding unit on-the-fly (but once again, YMMV).
Here are some hints to help you find the most efficient combination of options.
adactl -f rules_file.aru example -- -FT -C1 example.adt
provided the tree file already exists.
If you are using an old version of GNAT and your project includes source files located in several directories, the ADA_INCLUDE_PATH environment variable may not be considered by ASIS, resulting in error messages that tell you that the bodies of some units have not been found (and hence have not been processed). This problem has been fixed in Gnat dated later than Sept. 1st, 2006. If this happens, either provide your source directories as “-I” options (see ASIS options), or generate the tree files manually (see Generating tree files manually). Note that this problem does not happen if you are using Emacs project files (see Project files), nor if you are running AdaControl from GPS.
Like any sophisticated piece of software, AdaControl may fail when encountering some special case of construct. ASIS may also fail occasionnally; actually, we discovered several ASIS bugs during the development of AdaControl. These were reported to ACT, and have been corrected in the wavefront version of GNAT - but you may be using an earlier version. In this case, try to upgrade to a newer version of ASIS. If an AdaControl or ASIS problem is not yet solved, AdaControl is designed in such a way that an occasionnal bug won't prevent you from using it.
If AdaControl detects an unexpected exception during the processing of a unit (an ASIS error or an internal error), it will abandon the unit, clean up everything, and go on processing the remaining units. This way, an error due to a special case in a unit will not affect the processing of other units. AdaControl will return a Status of 10 in this case.
However, if it is run with the “-x” option (eXit on error), it will stop immediately, and no further processing will happen.
If you don't want the garbage from a failing rule to pollute your report, you may chose to disable the rule for the unit that has a problem. See Inhibit command.
If you encounter a problem while using AdaControl, you are very welcome to report it to rosen@adalog.fr. Please include the exact control and the unit that caused the problem, as well as the captured output of the program (with “-d” option).
AdaControl is about controlling rules. Rules are built in AdaControl; each rule has a name, and may require parameters. For the complete description of each rule, see Rules reference.
To run AdaControl, you need to define which rules you want to apply to your Ada units, what are the parameters, etc. In addition, you may want to define various things, like the file where the results should go, the output format, etc.
AdaControl defines a small command language which is used to describe how you want to process your units. Commands can be specified either on the command line or in a file, that we call here a rules file. Commands can also be given interactively; See Interactive mode.
The command language is not case-sensitive, i.e. the case of the keywords, rule names, and parameters is not significant. The layout of commands is free (i.e. a command can extend over several lines, and spaces are freely allowed between syntactic elements).
Comments are allowed in and between commands. Comments begin with a “#” or a “--”, and extend to the end of the line.
Since wide characters are allowed in Ada programs, AdaControl accepts wide characters in commands as well. With GNAT, the encoding scheme is Hex ESC encoding (see the GNAT User-Guide/Reference-Manual). This is the prefered method, since few people require wide characters in programs anyway, and that keeping the default bracket encoding would not conveniently allow brackets for regular expressions, like those used by some rules. See Syntax of regular expressions.
If a syntax error is encountered in a command, an appropriate error message is output, and analysis of the rules file continues in order to output all errors, but no analysis of user code will be performed.
A control command is a command that declares one (or several) controls. A control defines how a rule is applied to Ada units. The syntax of a control command is as follows:
<control_command> ::= [<label> ":"] <control> {"," <control>} ";"
<control> ::= <ctrl_kind> <Rule_Name> [<parameters>]
<parameters ::= "(" [<modifiers>] <value> {"," [<modifiers>] <value>} ")"
<ctrl_kind> ::= "check"|"search"|"count"
If present, the label gives a name to the control(s); it will be
printed whenever each control is activated, and can be used to disable
the control(s). See Disabling controls. If no label is present,
the rule name is printed instead. The label must have the syntax of an
Ada identifier, or else the label must be included within double
quotes ("), in which case it can contain any character.
Each control consists of a <ctrl_kind> followed by a rule name, and (optionally) parameters. Some parameters may be preceded by modifiers (such as “not” or “case_sensitive”). The meaning of the rule parameters and modifiers depends on the rule.
Here are some examples of commands:
check unnecessary_use_clause;
All_Imports: search pragmas (Import);
"Why do you need that?": check entities (Unchecked_Conversion,
all 'Address);
Specifying several controls with the same label is a shorthand which is equivalent to specifying the same label for several controls. It is handy when the label is long, and/or to stress that several controls are part of the same programming rule. For example:
"Check why this obsolete stuff is still used":
check entities (obsolete_unit_1), -- Note comma here!
check instantiations (some_obsolete_generic);
There are three control kinds: “check”, “search”, and “count”.
“Check” is intended to search for rules that must be obeyed in your programs. Normally, if a “Check” control fails, you should fix the program. “Search” is intended to report some situations, but you should consider what to do on a case-by-case basis. Roughly, use “check” when you consider that the failure of the control is an error, and “search” when you consider it as a warning. AdaControl will exit with a status of 1 if any “Check” control is triggered, and a status of 0 if only “Search” controls were triggered (or no control was triggered at all).
“Count” works like “Search”, but instead of printing a message for each control which is triggered, it simply counts occurrences and prints a summary at the end of the run. There is a separate count for each control label (or if no label is given, the rule name is taken instead); if you give the same label to different controls, this allows you to accumulate the counts.
A report message (except for the final report of “count”) comprises the following elements:
The formatting of the report message depends on the format option, which can be selected with the “-F” command-line option or the “set format” command.
If the format is “Gnat” (the default) or “Gnat_Short”, items are separated by ':'; this is the same format as the one used by GNAT error messages. Editors (like Emacs or GPS) that recognize this format allow you to go directly to the place of the message by clicking on it. In order to avoid too long messages, only the label appears, unless there is none, in which case it is replaced with the rule name.
If the format is “CSV” or “CSV_Short”, items are separated by ',' and surrounded by double quotes. This is the “Comma Separated Values” format, which can be read by any known spreadsheet program, except Excel(tm) by default, which uses the semicolon and not the comma to separate fields. Therefore, the formats “CSVX” and “CSVX_Short” do the same thing, but using semi-colons (';') instead of commas. Both the label (replaced by an empty column if there is none) and the rule name appear. Note that when an output file is created in one of the “CSV” formats, a title line is issued as the first line, following normal CSV convention.
If the format is “Source” or “Source_Short”, the offending source line is output, and the message is output behind it, with a “!” pointing to the exact location of the problem.
If the format is “None”, no error message is output at all. This is useful when only the return code of running AdaControl is desired (just to check if a program is OK or not). Note that this does not prevent the output of statistics, since these are under control of the “-S” option or the “set statistics” command. In this case, statistics are output in CSVX format, since asking for statistics with a “none” format is mainly useful for analysing the statistics with a spreadsheet program.
With recent versions of Gnat, the file name includes the full path of the source file. If the “_Short” form of the format option is used, the file name is stripped from any path. This can make it easier to compare the results of controlling units from various directories. Note that with older versions of Gnat, the file name never includes the full path, and the “_Short” form of the format option has no effect.
After each run (see Go command), statistics may be output, depending on the statistics level which is set with the “-S” option or the “set statistics” command. The meaning of the various levels is as follows:
Most rules accept parameters. Parameters can be:
A numerical value is given with the syntax of an Ada integer or real literal (underscores and exponents are allowed as in Ada). Based literals are supported for integer values; if somebody can justify a need for supporting them for reals, we'll be happy to add this feature later...
A character string is given within double quotes “"”. As usual, quotes appearing within the string are doubled. The tilde character (“~”) can be used as a replacement delimiter, but the same character must be used at both ends of the string. The latter has been chosen as a character not used by the various shells, and can be useful to pass quoted strings from parameters on the command line (unfortunately, we could not use the percent (“%”) sign, because it plays a special role in DOS/Windows).
An Ada entity name is the full name (prefixed with the names of all units that include it) of something declared in a program. It can be followed by overloading information, in order to uniquely identify the Ada entity. If an Ada entity is overloaded and no overloading information is provided, the rule is applied to all (overloaded) Ada entities that match the name. Alternatively, it can be “all” followed by a simple name, in wich case it applies to all entities with that name. See Specifying an Ada entity name for the full description of the syntax. Here are some examples of entity names:
Ada.Text_IO.Put -- All Put defined in Ada.Text_IO
Ada.Text_IO.Put{Standard.Character} -- The Put on Character
all Put -- All Put
Standard.Integer'Image -- The 'Image function on Integer
all 'Image -- All 'Image functions
Most rules can be used in more than one control (with different parameters). There is no difference between a single or a multiple configuration rule use: outputs, efficiency, etc. are the same.
The following rules files produce an identical configuration:
Search Pragmas (Pure, Elaborate_All);
and
Search Pragmas (Pure);
Search Pragmas (Elaborate_All);
However, the second form can be used to give different labels. Consider:
Search Pragmas (Pure);
No_Elaborate: Search Pragmas (Elaborate_All);
The messages for pragma Pure will contain “PRAGMAS”, while
those for Elaborate_All will contain “No_Elaborate”. If a
disabling comment mentions pragmas, it will disable both controls,
but a disabling comment that mentions No_Elaborate will disable
only the second one.
It is possible to disable controls on parts of the source code by
placing markers in the source code. A marker is an Ada comment, where
the comment mark (--) is immediately followed by the special
tag “##” (by default).
There are two kinds of markers: block markers and line markers. Both kinds specify a list of controls to disable/re-enable. A list of controls is a list of rule names (to disable/re-enable all controls on the indicated rule(s)) or control labels (to disable/re-enable all controls with that label), separated by spaces. Alternatively, the list of controls can be the word “all” to disable/re-enable all controls.
In a “--##” line, everything appearing after another “##”
tag (by default) is ignored. This allows the insertion of a comment
explaining why the control is disabled at that point.
Both tags can be changed with the “set” command. See Set command.
A control is disabled from a “rule off” marker that applies to it until a “rule on” marker that applies to it. If there is no appropriate “rule on” marker, the control is disabled up to the end of file.
Syntax:
--## rule off <control_list>
Ada code block
--## rule on <control_list>
Ex:
--## rule off rule1 rule2 ## Authorized by QA ref 1234
I := I + 1;
Proc (I);
--## rule on rule2
A control is disabled only for the line where a marker that applies to it appears.
Syntax:
Ada code line --## rule line off <rule_list>
Ex:
I := I + 1; --## rule line off rule3 rule_label_1
Conversely, it is possible to re-enable a control for just the current line in a block where it is disabled:
Syntax:
Ada code line --## rule line on <rule_list>
Ex:
--## rule off rule1 rule2
...
I := I + 1; --## rule line on rule2
Since the disabling is based on special comments, there is a conflict with the rule “header_comments” which is based on the content of comments. Line disabling is not possible with this rule, and block disabling needs special care. See Header_Comments.
In addition to controls, AdaControl recognizes a number of commands. Although these commands are especially useful when using the interactive mode (see Interactive mode), they can be used in command files as well.
This command starts processing of the controls that have been specified.
Syntax:
go;
Controls are not reset after a “go” command; for example, the following program:
search entities (pack1);
go;
search entities (pack2);
go;
will first output all usages of Pack1, then all usages of both
Pack1 and Pack2. See Clear command to reset
controls.
If not in interactive mode, a “go” command is automatically added at the end, therefore it is not required in rules files.
This command terminates AdaControl.
Syntax:
quit;
If given in a file, all subsequent commands will be ignored. This command is really useful only in interactive mode. See Interactive mode.
This command prints a message on the output file.
Syntax:
message "<any string>" [pause];
The length of the message is limited to 250 characters. If the word “pause” (case irrelevant) is specified after the message, AdaControl will wait for the user to press the Return key before proceeding.
Note that the message is syntactically a string, and must therefore be quoted (double quotes).
This command prints various informations about the rules and AdaControl itself.
Syntax:
Help [<keyword> | <rule name> {,<keyword> | <rule name>}]
<keyword> ::= all | commands | license | list | options | rules | version
Without any argument, this command prints a summary of all commands and rule names. If given one or more keywords or rule names, it prints the corresponding help message. See Getting help for the details.
This command command clears (i.e. removes) controls that have been previously given.
Syntax:
Clear all | <rule name>{,<rule name>} ;
The command clears all controls given for the indicated rules, of for
all rules if the all keyword is given. For example, the
following program:
search entities (pack1);
go;
clear all;
search entities (pack2);
go;
will first output all usages of Pack1, then all usages of
Pack2. Without the “clear all” command, the second “go”
would output all usages of Pack1 together with all usages of
Pack2.
This command sets various parameters of AdaControl.
Syntax:
set Format Gnat|Gnat_Short|CSV|CSV_Short|Source|Source_short|None;
set Check_Key|Search_Key "<value>"
set Max_Errors [<value>];
set Max_Messages [<value>];
set Output|New_Output <output file>;
set Statistics <level>;
set Tag1|Tag2 "<value>";
set Trace <trace file>;
set Debug|Exit_On_Error|Ignore|Timing|Verbose|Warning|Warning_As_Error
On|Off;
The “set format” command selects the output format for the messages, like the “-F” option; see Control kinds and report messages for details.
The “set check_key” command defines a string which is used in place of “Error” in messages issued by a “check” control. Similarly, the “set search_key” command defines a string which is used in place of “Found” in messages issued by a “search” control. This can be useful when AdaControl is used, for example, to detect places where manual inspection is required; having the word “Error” in the message could be misleading to the persons in charge of the review. Note however that if you set these keys, the GPS interface will not be able to recognize properly the messages.
The “set max_errors” and “set max_messages” limit the output of AdaControl, like the “-m” and “-M” options; see Output limits for details. If no <value> is given after the command name, the corresponding limitation is removed.
The “set output” and “set new_output” commands redirect the output
of subsequent controls to the indicated file. If the string
console (case irrelevant) is given as the <output file>, output
is redirected to the console.
The “set new_output” always create a new file (or overwrites an existing file with the same name).
The “set output” command appends if the file exists, unless the “-w” option is given, in which case it is overwritten. However, the file is overwritten only the first time it is mentionned in an “output” command. This means that you can switch forth and back between two output files, all results from the same run will be kept. Note however that for this to work, you need to specify the output file exactly the same way: if you specify it once as “result.txt”, and then as “./result.txt”, the second one will overwrite the first one.
The “set statistics” command sets the statistics level, like the “-S” option; see Control kinds and report messages for details.
The “set Tag1|Tag2” command changes the tags used to disable (or
enable) rules. “Tag1” is the string that appears immediately after
the comment indicator (--), and “tag2” is the tag that
terminates the special comment. Note that these tags must be given as
strings (in quotes) and that case is relevant. See Disabling controls for details.
The “set trace” command redirects the trace messages of the
“-d” option to the indicated file. If the string console
(case irrelevant) is given as the <trace file>, trace messages are
redirected to the console. As with the “-t” option, if the file
exists, output is appended to it.
The “set Debug|Exit_On_Error|Ignore|Timing|Verbose|Warning|Warning_As_Error” command activates (“on”) or deactivates (“off”) options. “Debug” corresponds to the “-d” option, “Exit_On_Error” to the “-x” option, “Ignore” to the “-i” option, “Timing” to the “-T” option, “Verbose” to the “-v” option, “Warning” to the “-E” option, and “Warning_As_Error” to the “-e” option. See Verbose and debug mode, Exit on error, Treatment of warnings, Output format, and Local disabling control for details.
This command inputs commands from another file.
Syntax:
Source <input file>;
Commands are read and executed from the indicated file, then control is returned to the place after the “source” command. There is no restriction on the content of the sourced file; especially, it may itself include other “source” commands.
If <input file> is not found, AdaControl will retry the same name with
.aru appended. It is a syntax error if the file is not found
either.
If <input file> is a relative file path, it is taken relatively to the file where the “source” command is given. Especially, if no path is specified, the sourced file will be taken from the same directory as the sourcing file (irrespectively of where the command is being run from).
If the string console (case irrelevant) is given as the <input
file>, commands are read from the console until a “quit” command is
given. This command is of course useful only from files, and allows to
pass temporarily control to the user in interactive mode.
This command prevents execution of certain controls on particular units.
Syntax:
Inhibit <rule name>|all ([all] <unit> {,[all] <unit>});
Controls refering to the given rule (or all rules if “all” is specified in place of a rule name) for the indicated unit(s) are not performed. In addition, if “all” is specified in front of the unit name, the unit will not be accessed at all, even from rules that follow call graphs, and could thus access this unit while analyzing other units.
There are several reasons why you might want to inhibit a control of a rule for certain units:
The “all” option for a unit is intended for the last case, to prevent ASIS bugs from spoiling any unit that calls something from an offending unit.
Below is an example of a file with multiple commands:
message "Searching Unchecked_Conversion";
search entitities (ada.unchecked_conversion);
set output uc_usage.txt;
go;
clear all;
message "Searching 'Address";
search entities (all 'Address);
set output address_usage.txt;
go;
This file will output all usages of Ada.Unchecked_Conversion
into the file uc_usage.txt, then output all usages of the
'Address attribute into the file
address_usage.txt. Messages are output to tell the user about
what's happenning.
This chapter describes each rule currently provided by
AdaControl. Note that the rules directory of the distribution
contains a file named verif.aru that contains an example of a
set of rules appropriate to check on almost any software.
A general limitation applies to all rules. AdaControl is a static checking tool, and therefore cannot check usages that depend on run-time values. For example, it is not possible to check rules applying to an entity when this entity is aliased and accessed through an access value, or rules applying to subprogram calls when the call is a dispatching call.
This rule controls functions that may not terminate normally, i.e. where
Program_Error could be raised due to reaching the end of the
function without encountering a return statement.
<control_kind> abnormal_function_return;
The rule controls that the sequence of statements of each function body, as well as each of its exception handlers, ends with:
Ada.Exceptions.Raise_Exception or
Ada.Exceptions.Reraise_Occurrence).
This is a sufficient (but of course not necessary) condition to ensure
that no function raises Program_Error due to reaching the end
of its statements without encountering a return.
This rule can be specified only once.
Ex:
check abnormal_function_return;
This rule checks that a function always returns correctly, but does not
prevent multiple return statements in functions. If you want
to ensure that there is exactly one return statement in functions,
and that this statement is always the last one, use this rule together with
the rule statements(function_return).
See Statements.
This rule controls the use of allocators (i.e. dynamic memory allocation).
<control_kind> allocators
[(task|protected|<entity> {, task|protected|<entity>})];
If one or several <entity> are given, only allocators whose allocated type matches the <entity> are controlled; as usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name. If “task” or “protected” is given, allocators for task types or protected types (respectively) are controlled; otherwise all allocators are controlled. This rule is especially useful for finding memory leaks, since it tells all the places where dynamic allocation occurs.
Ex:
search allocators (standard.string);
check allocators (T'Class);
The type given in the rule must be a first named subtype, and the rule will also find allocators that use a subtype of this type. If the type is declared within a generic package, the rule will control all corresponding types from instantiations.
The type mentionned in the rule is the one following the new keyword, which is not necessarily the same as the expected type in presence of implicit conversions like this:
type T is tagged ...;
type Class_Access is access T'Class;
X : Class_Access;
begin
X := new T;
This allocator will be found for type T, not for type T'Class.
This rule controls properties of arrays, by enforcing a consistent value or range of values for the lower or upper bound, or by limiting the possible size. It can also control various aspects of the component type of the array.
<control_kind> array_declarations (first, <value> | <bounds>);
<control_kind> array_declarations (last, <value> | <bounds>);
<control_kind> array_declarations (dimensions, <value> | <bounds>);
<control_kind> array_declarations (length, <bounds>);
<control_kind> array_declarations (component, <compo_kind> {,<repr_cond>});
<bounds> ::= min|max <value> [, min|max <value> ]
<compo_kind> ::= <entity>|<category>
<category> ::= () | access | array | delta | digits | mod | private |
protected | range | record | tagged | task
<repr_cond> ::= [not] packed | sized | component_sized
The first parameter is a subrule keyword:
search array_declarations (first, 1);
check array_declarations (first, min -1, max 1);
will be silent if the lower bound of an array is 1, it will issue a warning if it is in the range -1 .. 1, and an error otherwise.
search array_declarations (Dimensions, 1);
check array_declarations (Dimensions, min 2, max 3);
will be silent for one-dimensional arrays, it will issue a warning for 2- and 3-dimensional arrays, and an error otherwise.
This rule can be specified several times for the “component” subrule. For other subrules, it can be specified at most once for each subrule and for each of “check”, “search” and “count”. It is thus possible for each subrule to have a value considered a warning, and a value considered an error.
Ex:
-- All arrays should start at 1:
check array_declarations (first, 1);
-- No arrray of more than 100 elements:
check array_declarations (length, max 100);
-- No empty array:
check array_declarations (length, min 1);
-- Arrays whose component type is private:
check array_declarations (component, private);
-- Packed arrays of Character
check array_declarations (component, Standard.Character, packed);
-- Packed arrays of record without size clause
check array_declarations (component, record, packed, not sized);
The subrule Max_Length ignores index constraints that are not
static. Non static index constraints can be controlled with the rule
Non_Static (Index_Constraint). See Non_Static.
Requiring the same upper bound for all arrays is not very useful, but:
check array_declarations (last, min 1);
can be used to check that no array has a negative or zero upper bound.
Although the language allows any expression as the barrier of a protected entry, it is generally better to use only “simple” expressions. This rule controls the kind of constructs allowed in barrier expressions.
<control_kind> Barrier_Expressions ([<allowable> {, <allowable>}]);
<allowable> ::= <entity> | <keyword>
<keyword> ::= allocation | any_component | any_variable |
arithmetic_operator | array_aggregate | comparison_operator |
conversion | dereference | indexing |
function_attribute | local_function | logical_operator |
record_aggregate | value_attribute
Without parameters, the only elements allowed in barriers are references to boolean components of the protected element and litterals (this corresponds to what is allowed for the Ravenscar profile). Parameters specify other constructs that are allowed:
Standard.Boolean.
"+", "**", etc.).
"=", ">", in, etc.).
'Pred, 'Image, etc.).
'First, 'Terminated, etc.).
This rule can be given only once for each of “check”, “search” and “count”.
Ex:
search barrier_expressions;
check barrier_expressions (logical_operator, comparison_operator,
any_component,
Pack.Global_State);
The goal of the “Simple_Barrier” restriction from the Ravenscar profile is to ensure that evaluation of barriers never raise exceptions. Even simple things like a qualified expression can raise exceptions, but in practice more than the restriction of the Ravenscar profile can be “reasonably” allowed.
Note that the various “operator” keywords allow only the use of predefined operators. If a user defined operator should be allowed, provide it explicitely as an <entity>. There is no way to allow any function call, since this would boil down to allowing pretty much anything, but you can of course specify explicitely functions that can be called.
You can provide this rule both for “check” and “search”, but of course it makes sense only if the set of allowed features for “search” is a superset of those allowed for “check”. This way, the use of certain features can be interpreted only as a warning.
This rule controls various metrics related to the case statement. It is intended for cases where it is desired to limit the complexity of case statements.
<control_kind> Case_Statement (<subrule>, <bound> [, <bound>]);
<subrule> ::= others_span | paths | range_span | values
<bound> ::= min | max <value>
The first parameter is a subrule keyword. The second (and optionnally third) parameter give the minimum and/or maximum allowed values (i.e. the rule will control values outside the indicated interval). If not specified, the minimum value is defaulted to 0 and the maximum value to infinity. The parameters controlled by each subrule are:
This rule can be specified at most once for each subrule and for each of “check”, “search” and “count”. It is thus possible for each subrule to have a value considered a warning, and a value considered an error.
Ex:
check Case_Statement (others_span, min 1);
search Case_Statement (others_span, min 5);
check Case_Statement (values, max 10);
check Case_Statement (paths, min 3, max 30);
To control that no range is used as a choice in a case statement:
check case_statement (range_span, max 0);
To control “when others” that cover no value at all:
check case_statement (others_span, min 1);
If some characteristic of the case statement depend on a generic formal type, it is not possible to control some of the features statically. Such cases are detected by the rule “uncheckable”. See Uncheckable.
This rule makes sure that the program text does not use “undesirable” characters.
<control_kind> characters [(<subrule> {, <subrule>})];
<subrule> ::= control | not_iso_646 | trailing_space | wide
The rule controls the occurrence in the source file of characters belonging to the classe(s) defined by the subrules. Without parameters, all classes are controlled. The classes are defined as follows:
Standard.Character.
This rule can be given only once for each class of characters.
Ex:
check characters (control, trailing_space);
search characters (not_iso_646);
With the “wide” subrule, the error message may seem to not always appear at the right place; this depends on the encoding scheme used. For example, if your source contains (using bracket encoding):
S : Wide_String := "["1041"]["1042"]";
it will appear to AdaControl as a string containing two characters, and therefore the error message for the second wide character will point at two characters after the opening quote of the string.
This rule controls only the characters in the source file; other means
of having characters in the corresponding classes (like using the
'Val attribute) are not controlled.
This rule controls comments that must, or must not, appear in certain cases.
<control_kind> comments (pattern, "<pattern>" {, "<pattern>"});
<control_kind> comments (position, <value> | <bounds>);
<control_kind> comments (terminating {, "<pattern>" | begin | end});
<control_kind> comments (unnamed_begin, <kind> {, <kind>});
<bounds> ::= min|max <value> [, min|max <value> ]
<kind> ::= [<condition>] <unit_kind>
<condition> ::= always | declaration | program_unit
<unit_kind> ::= all | procedure | function | entry | package | task
The first parameter is a subrule name which detemines what is being controlled.
--”
and spaces following it. Patterns are given using the full Regexp
syntax. see Syntax of regular expressions for details. Pattern
matching is always case insensitive.
This subrule is especially useful to find lines with comments like “TBSL” (To Be Supplied Later) or “fixme”, which are often used to mark places where something should be done before releasing the program.
search comments (position, 1);
check comments (first, min 1, max 6);
will be silent for comments that start in column 1, it will issue a warning for comments that start at columns 2 to 6, and an error otherwise.
The <condition> keyword determines circumstances where the comment is required:
The <unit_kind> keyword detemines the kind of program unit to which the rule applies (“all” stands for all kinds). The subrule can be given only once of each kind of program unit.
Ex:
check comments (pattern, "TBSL");
-- Report places where rules are disabled:
search comments (pattern, "##.* off");
-- End of line comments are not allowed, except for the
-- comment that repeats the name of a procedure on the "begin"
-- line, and special AdaControl comments
check comments (terminating, begin, "^ *##");
-- Named begin required for packages unless they have no
-- declaration, and subprograms if they have nested units
check comments (unnamed_begin, declaration package);
check comments (unnamed_begin, program_unit procedure);
check comments (unnamed_begin, program_unit function);
Remember that a Regexp matches if the pattern matches any part of the identifier. Use “^” and “$” to match the beginning (resp. end) of the comment, or both.
This rule does not support wide characters outside the basic Latin-1 set.
This rule controls usage of various kinds of declarations, possibly only those occurring at specified locations.
<control_kind> declarations (<subrule> {, <subrule>});
<subrule> ::= {<location_kw>} <declaration_kw>
<location_kw> ::= all | block | library | local | nested |
own | private | public | task_body
<declaration_kw> ::=
any_declaration |
abstract_function | abstract_procedure |
abstract_type | access_all_type |
access_constant_type | access_protected_type |
access_subprogram_type | access_task_type |
access_type | aliased_array_component |
aliased_constant | aliased_protected_component |
aliased_record_component | aliased_variable |
anonymous_subtype_allocator | anonymous_subtype_case |
anonymous_subtype_declaration | anonymous_subtype_for |
anonymous_subtype_indexing | array |
array_type | binary_modular_type |
character_literal | child_unit |
class_wide_constant | class_wide_variable |
constant | constrained_array_constant |
constrained_array_type | constrained_array_variable |
controlled_type | decimal_fixed_type |
defaulted_discriminant | defaulted_generic_parameter |
defaulted_parameter | deferred_constant |
derived_type | discriminant |
empty_private_part | empty_visible_part |
enumeration_type | entry |
exception | extension |
fixed_type | float_type |
formal_function | formal_package |
formal_procedure | formal_type |
function | function_call_renaming |
function_instantiation | generic |
generic_function | generic_package |
generic_procedure | handlers |
incomplete_type | in_out_generic_parameter |
in_out_parameter | initialized_protected_component |
initialized_record_component | initialized_variable |
instantiation | integer_type |
library_unit_renaming | limited_private_type |
modular_type | multiple_names |
multiple_protected_entries | named_number |
non_binary_modular_type | non_identical_operator_renaming |
non_identical_renaming | non_joint_CE_NE_handler |
non_limited_private_type | not_operator_renaming |
null_extension | null_ordinary_record_type |
null_procedure | null_tagged_type |
operator | operator_renaming |
ordinary_fixed_type | ordinary_fixed_type_no_small |
ordinary_fixed_type_with_small | ordinary_record_type |
out_parameter | package |
package_instantiation | package_statements |
predefined_operator | private_extension |
procedure | procedure_instantiation |
protected | protected_entry |
protected_type | protected_variable |
record_type | renaming |
renaming_as_body | renaming_as_declaration |
self_calling_function | self_calling_procedure |
separate | signed_type |
single_array | single_protected |
single_task | subtype |
tagged_type | task |
task_entry | task_type |
task_variable | type |
unconstrained_array_constant | unconstrained_array_type |
unconstrained_array_variable | unconstrained_subtype |
uninitialized_protected_component | uninitialized_record_component |
uninitialized_variable | variable |
variant_part
The <location_kw> restricts the places where the occurrence of the declaration is controlled. Several <location_kw> can be given, in which case the declaration is controlled at places where all the keywords apply. If there is no <location_kw>, it is assumed to be “all”.
all: puts no special restriction to the location. This keyword
can be specified for readability purposes, and if specified must
appear alone (not with other <location_kw>).
block: only declarations appearing in block statements are controlled.
library: only library level declarations are controlled.
local: only local declarations are controlled (i.e. only declarations
appearing in (generic) packages, possibly nested, are allowed).
nested: only declarations nested in another declaration are
controlled (i.e. only library level declarations are allowed).
own: only declarations that are local to a (generic) package body
are controlled.
public: only declarations appearing in the visible part of
(generic) packages are controlled.
private: only declarations appearing directly in a private
part are controlled.
task_body: only declarations appearing directly in a task body
are controlled. Note that it would not make sense to have a
<location_kw> for task specifications, since only entries can
appear there, and they cannot appear anywhere else.
The <declaration_kw> specifies what kind of declaration to control:
any_declaration controls all declarations. This is of course not
intended to forbid all declarations in a program (!), but
counting all declarations can be quite useful.
abstract_function and abstract_procedure control the
declarations of abstract functions and abstract procedures,
respectively.
abstract_type controls the declaration of non-formal abstract
types.
access_type controls all access type declarations, while
access_subprogram_type, access_protected_type, and
access_task_type control only access to procedures or functions,
access to protected types, or access to task types, respectively. Similarly,
access_all_type control generalized access to variables types (aka
"access all T", and access_constant_type control
generalized access to constants types (aka "access constant T").
aliased_variable and aliased_constant control the
declarations of aliased variables or constants, respectively.
aliased_array_component controls the declaration of arrays
(array types or single arrays) whose components are declared aliased.
aliased_record_component and aliased_protected_component
control the declarations of aliased record (respectively protected)
components.
anonymous_subtype_declaration controls the declarations of
anonymous subtypes and ranges that are part of some other
declaration. Similarly, anonymous_subtype_allocator,
anonymous_subtype_case, anonymous_subtype_for, and
anonymous_subtype_indexing control anonymous subtype
declarations and ranges that are part of allocators, case
statements (ranges in the when path), for loop
statements, and indexing of slices or array aggregates, respectively.
array controls all array definitions (array types and single
arrays), while array_type controls only array types and
single_array controls only single arrays (objects of an
anonymous array type). constrained_array_type controls only
constrained array types, while unconstrained_array_type
controls only unconstrained array
types. constrained_array_variable controls variable
declarations where the given (or anonymous) array type is constrained,
while unconstrained_array variable controls variable
declarations where the given (or anonymous) array type is
unconstrained (and the constraint is provided by the initial value).
constrained_array_constant and
unconstrained_array_constant do the same with constants instead
of variables.
character_literal controls the declaration of new character
literals, i.e. character literals defined as part of the values of an
enumeration type.
child_unit controls the declaration of all child units.
constant controls all constants, while
class_wide_constant control the declaration of constants of a
class-wide type, and deferred_constant controls the declaration
of deferred constants.
controlled_type controls the declaration of controlled types,
i.e. descendants of Ada.Finalization.Controlled or
Ada.Finalization.Limited_Controlled. Note that this includes
also private types that are not visibly controlled.
defaulted_parameter controls subprogram or entry (in)
parameters that provide a default value, while
defaulted_generic_parameter controls generic formal objects
that provide a default value.
derived_type controls regular derived types, but not type
extensions (derivations of tagged types). These are controlled by
extension and private_extension.
discriminant controls all declarations of types with
discriminants, while defaulted_discriminants controls only
those where defaults are provided for the discriminants.
empty_private_part controls package specification with an empty
private part, i.e. where the word private appears, but the
private part contains no declaration (even if it contains pragmas).
empty_visible_part controls package specifications that contain
no declaration in the visible part (before the word private
if any), even if it contains pragmas.
enumeration_type controls the declaration of enumeration types.
exception controls exception declarations.
fixed_type controls all declarations of fixed point types while
ordinary_fixed_type controls only ordinary (binary) fixed point
types, ordinary_fixed_type_no_small controls ordinary fixed
point type without a representation clause for 'SMALL,
ordinary_fixed_type_with_small controls ordinary fixed point
type with an explicit representation clause for 'SMALL, and
decimal_fixed_type controls only decimal fixed point types
(those can never have a representation clause for 'SMALL).
float_type controls declarations of floating point types.
formal_function, formal_package, formal_procedure, and
formal_type control generic formal functions, packages,
procedures, and types, respectively.
generic_function, generic_package,
generic_procedure control generic function (respectively
package, procedure) declarations.
handlers controls the presence of exception handlers in any
handled sequence of statements.
in_out_parameter and out_parameter control subprogram
and entry parameters of modes in out and
out (respectively), while in_out_generic_parameter
and out_generic_parameter do the same for generic formal
parameters
incomplete_type controls incomplete type declaration.
initialized_variable controls variable declarations that
include an initialization expression, unless they are of a class-wide
type since initialization is required in that case.
instantiation controls all instantiations, while
function_instantiation, package_instantiation,
procedure_instantiation control function (respectively package,
procedure) instantiations.
integer_type controls all declarations of integer types, while
signed_type controls only signed integer types, and
modular_type controls only modular types (both kinds);
binary_modular_type controls only modular types whose modulus
is a power of 2, and non_binary_modular_type controls only
modular types whose modulus is not a power of 2.
initialized_record_component and
initialized_protected_component control the declaration of
record (respectively protected) component that include a default
initialization, while uninitialized_record_component and
uninitialized_protected_component control the declaration of record
(respectively protected) component that do not include a default
initialization, unless they are of a limited type since initialization would
not be allowed in that case.
limited_private_type controls limited private type
declarations, while non_limited_private_type controls regular
(non limited) private type declarations.
multiple_names controls declarations where more than one
defining identifier is given in the same declaration.
multiple_protected_entries controls protected definitions (from
protected types or single protected objects) that have more than one
entry declaration. Note that a protected definition with a single
entry family declaration is counted as a single entry declaration.
named_number controls declarations of named numbers,
i.e. untyped constants.
non_joint_CE_NE_handler controls exception handlers whose
choices include Constraint_Error or Numeric_Error, but
not both. This is intended for legacy Ada 83 code that required to
always handle these exceptions together; it makes little sense for
Ada95 or Ada2005 code (and to be honnest, this subrule is provided
because Gnatcheck has it).
null_extension controls record extensions (derived tagged
types) that contain no new elements. Similarly,
null_ordinary_record_type and null_tagged_type control
ordinary records and tagged types that contain no elements. Note that
the record definitions may be plain “null
record” definitions, or full record definitions that
contain only null components. However, a definition is not considered
null if it contains a variant part.
null_procedure controls procedure declarations whose sequence
of statements contain only null statements (or blocks
without declarations and containing only null statements).
operator controls the definition of operators (things like
"+"); note that the message is given on the specification if
there is an explicit specification, on the body
otherwise. predefined_operator controls only operator
definitions that overload a predefined operator (like "+" on a
numeric type, for example).
package_statements controls the presence of elaboration
statements in the bodies of packages (or generic packages).
private_extension controls private extensions, i.e. derivations
from a tagged type with a with private extension part.
record_type controls all record type declarations (tagged or
not), while ordinary_record_type controls only non-tagged
record types, and tagged_type controls only tagged record types.
renaming controls all renaming declarations, while
renaming_as_body controls only those that are renamings as
bodies of subprograms, renaming_as_declaration controls only
those that are regular renamings of subprograms (i.e. not as bodies),
operator_renaming controls only those that are renamings of an
operator, not_operator_renaming controls only those that are
not renamings of an operator, function_call_renaming
controls renaming of the result of a function call, and
library_unit_renaming controls renaming of library units.
non_identical_renaming controls only renamings where the new
name and the old name are not the same, and
non_identical_operator_renaming does the same, but only for
renamings of operators.
self_calling_function controls functions whose body contains
only a single return statement, and the return expression
is a (recursive) call to the same function. Similarly,
self_calling_procedure controls procedures whose body contains
only a single statement which is a (recursive) call to the same
procedure. Note that this corresponds to bodies automatically
generated by gnatstub.
subtype controls all explicit subtype declarations (i.e. not
all anonymous subtypes that appear at various places in the
language), while unconstrained_subtype controls only the
subtype declarations that do not include a constraint.
task controls task type declarations as well as single tasks
declarations while single_task and task_type control
only single task declarations or task type declarations respectively
(and similarly for protected).
type controls all type (but not subtype) declarations.
variable controls all variable declarations, while
uninitialized_variable controls only variable declarations that
do not include an initialization expression, unless they are of a
limited type since initialization would not be allowed in that case.
class_wide_variable controls the declarations of variables of a
class-wide type. task_variable and protected_variable
control task and protected objects (respectively), whether given with
a named or anonymous type.
variant_part controls variant parts in record defintions.
Ex:
search declarations (task, exception);
check declarations (block procedure, block function, block package);
check declarations (public task);
Certain keywords are not exclusive, and it may be the case that several keywords apply to the same declaration; in this case, they are all reported. For example, if you specify:
check declarations (record_type, tagged_type);
tagged types will be reported both as “record_type” and “tagged_type”.
Some of the keyword do not seem very useful; it would be strange to have a programming rule that prevents all type declarations... But bear in mind that the <location_kw> can be used to restrict the check to certain locations; moreover, AdaControl can be used not only for checking, but also for searching; finding all type declarations in a set of units can make sense. As another example, “search declarations (own variable);” will find all variables declared directly in package bodies.
Some modifiers do not make sense with certain declarations; for example, a “private out_parameter” is impossible (a parameter occurs in a subprogram declaration, not directly in a private part). This is not a problem as far as the rule is concerned, but don't expect to find any...
In some rare cases, AdaControl may not be able to evaluate the modulus of a modular type definition, thus preventing correct operation of “binary_modular_type” and “non_binary_modular_type” subrules. Such cases are detected by the rule “uncheckable”. See Uncheckable.
This rule checks usage (or non-usage) of defaulted parameters.
<control_kind> default_parameter (<place>, <formal>, <usage>);
<place> ::= <entity> | calls | instantiations
<formal> ::= <formal name> | all
<usage> ::= used | positional | not_used
The rule controls subprogram calls or generic instantiations that use the default value for the indicated parameter, or conversely don't use it, either in positional notation or in any notation. If a subprogram is called, or a generic instantiated, whose name matches <entity>, and it has a formal whose name is <formal name>, then:
used (case irrelevant) is given as the third
parameter, the rule reports when there is no corresponding actual
parameter (i.e. the default value is used for the parameter).
positional (case irrelevant) is given as the
third parameter, the rule reports when there is an explicit
corresponding actual parameter (i.e. the default is not used for the
parameter), and the actual uses positional (not named) notation.
not_used (case irrelevant) is given as the third
parameter, the rule reports when there is an explicit corresponding
actual parameter (i.e. the default is not used for the parameter),
independently of whether it uses positional or named notation.
Alternatively, the <entity> can be specified as calls, to
control all calls or instantiations, to control all
instantiations. The <formal name> can be replaced by all, in
which case all formals are controlled.
Ex:
check default_parameter (P, X, used);
check default_parameter (P, Y, not used);
search default_parameter (all, all, positional);
If the <entity> is a generic subprogram, it is also possible to give a formal parameter (a parameter of the subprogram, not a generic parameter) as the <formal name>; in this case, all instantiations of the indicated generic subprogram will be controlled for the use of the indicated parameter.
This rule controls dependencies of units (i.e. with clauses, parents...), either according to a set of allowed units, or by count.
<control_kind> dependencies (others, <unit> {,<unit>});
<control_kind> dependencies (<counter>, <bound> [, <bound>]);
<counter> ::= raw | direct | parent
<bound> ::= min | max <value>
The kind of action depends on the specified subrule.
The “others” subrule controls semantic dependencies to units other than those indicated. This subrule can be specified only once, and at least one unit must be given.
Other subrules control that the number of various dependencies is whithin a specified range. The second (and optionnally third) parameter give the minimum and/or maximum allowed values (i.e. the rule will control values outside the indicated interval). If not specified, the minimum value is defaulted to 0 and the maximum value to infinity.
Ex:
check dependencies (others, Ada.Text_IO);
check dependencies (raw, max 15);
-- child units should not be nested more than 5 levels:
check dependencies (parent, max 5);
-- units that depend on nothing:
search dependencies (direct, min 1);
This rule checks that global variables in package bodies are accessed only through dedicated subprograms. Especially, it can be used to prevent race conditions in multi-tasking programs.
<control_kind> directly_accessed_globals [(<kind> {,<kind>})];
<kind> ::= plain | accept | protected
The rule controls global variables declared directly in (generic) package bodies that are accessed outside of dedicated callable entities (i.e. procedure or function, possibly protected, protected entries, and accept statements).
This rule can be specified only once. The parameters indicate which kinds of callable entity are allowed: “plain” for non-protected subprograms, “protected” for protected subprograms, and “accept” for accept statements). Without parameters, all forms are allowed.
More precisely, the rule ensures that the global variables are read from a single callable entity, and written by a single callable entity. Note that the same callable entity can read and write a variable, but in this case no other callable entity is allowed to read or write the variable.
Ex:
check directly_accessed_globals
Note that this rule controls global variables from package bodies, not those from the specification. This is intended, since it makes little sense to declare a variable in a specification, and then require it not to be accessed directly, but through provided subprograms. Obviously, in this case the variable should be moved to the body.
Note that AdaControl can check that no variable is declared in a package specification with the following rule:
check usage (variable, from_spec);
see Usage for details.
AdaControl cannot check entities accessed through dynamic names (dynamic renaming, access on aliased variables). Use of such constructs is detected by the rule “uncheckable”. See Uncheckable.
This rule checks that some procedures (notably initialization procedures) are not called several times in identical conditions.
<control_kind> duplicate_initialization_calls (<entity> {, <entity>});
This rule controls calls to initialization procedures that are duplicated. The <entity> parameters are the initialization procedures to be controlled. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
More precisely, the initialization procedures must follow one of these patterns:
The rule controls any violation of these patterns. If a procedure passed as parameter does not have a profile that corresponds to one of the above patterns, it is an error.
Ex:
check duplicate_initialization_calls (pack.init_proc);
If a variable passed as an out parameter is not statically determinable, it is not controlled by the rule. Such a case is detected by the rule “uncheckable”. See Uncheckable.
This rule is used to control usage of Ada entities, i.e. any declared element (type, variables, packages, etc).
<control_kind> entities (<entity> {, <entity>});
This rule controls all uses of the indicated entities. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name. This rule is not intended to replace cross-references, but can be quite handy to check, for example, that a program does not contain any more calls to debugging procedures before fielding it.
Note that this rules reports on the use of the entity, not the name: if an entity has been renamed, it will be found under its various names. Similarly, if the given entity is a generic unit or an entity declared inside a generic unit, all corresponding uses in all instances will be reported.
Ex:
search entities (Debug.Trace);
check entities (Ada.Text_IO.Float_IO.Put);
The second line will report on any use of a Put from any
instantiation of Float_IO.
This rule can also be used to check for all occurrences of certain
attributes with the “all <Attribute>” syntax. For example,
the following will report on any usage of 'Unchecked_Access:
check entities (all 'Unchecked_Access);
In certain contexts, only a limited set of the Ada predefined units is
allowed. For example, it can be useful to forbid entities from
Standard, System, or entities defined in special needs
annexes. The rules directory of Adacontrol contains files with
Entities rules that forbid the use of various predefined Ada
units. Comment out the lines for the units that you want to allow.
You can then simply “source” these files from your own rule file (or
copy the content) if you want to disallow these units. See Rules files provided with AdaControl.
Gnat defines Unchecked_Conversion and
Unchecked_Deallocation as separate entities, rather than
renamings of Ada.Unchecked_Conversion and
Ada.Unchecked_Deallocation. As a consequence, it is necessary
to specify explicitely both forms if you want to make sure that the
corresponding generics are not used.
This rule controls entities that appear within exception handlers.
<control_kind> entity_inside_exception (<spec> {, <spec>});
<spec> ::= [not] <entity> | calls | entry_calls
This rule controls exception handlers that contain references to one or several Ada entities specified as parameters. If the keyword “calls” is given, it stands for all subprogram and entry calls. If the keyword “entry_calls” is given, it stands for all entry calls (task or protected). If an <entity> (or “calls” or “entry_calls”) is preceded by the keyword “not”, it is not included in the list of controlled entities (i.e. the entity is allowed in the exception hhandler). This allows to make exceptions to a more general specification of an entity, or to allow calls to well-defined procedures if the keyword “calls” is given.
Ex:
-- No Put_Line in exception handlers:
check entity_inside_exception (ada.text_io.put_line);
-- No entry calls in exception handlers:
check entity_inside_exception (entry_calls);
-- No calls allowed, except to the Report_Exception procedure:
check entity_inside_exception (calls, not Reports.Report_Exception);
-- No Put allowed, except the one on Strings:
check entity_inside_exception (all Put,
not Ada.Text_IO.Put{Standard.String});
This rule controls that certain program units are guaranteed to never propagate exceptions, or that local exceptions cannot propagate out of their scope.
<control_kind> exception_propagation
(local_exception);
<control_kind> exception_propagation
([<level>,] interface, <convention> {, <convention> });
<control_kind> exception_propagation
([<level>,] parameter, <entity> {, <entity>});
<control_kind> exception_propagation
([<level>,] task);
<control_kind> exception_propagation
(<level>, declaration);
The “local_exception” subrule controls a design pattern that ensures
that a local exception cannot propagate outside the scope where it is
declared. If an exception is declared within a block, a subprogram
body, an entry body, or a task body, then this body must have either a
handler for this exception or for others; this handler must
not reraise the exception; and no handler is allowed to raise
explicitely the exception. The subrule controls explicit
raise statements and calls to Raise_Exception and
Reraise_Occurrence, but it does not control exceptions raised
as a consequence of calling other subprograms.
The other subrules control subprograms, tasks, or all declarations that can propagate exceptions, while being used in contexts where it is desirable to ensure that no exception can be propagated.
A subprogram or task is considered as not propagating if:
Ada.Exception.Raise_Exception or
Ada.Exception.Reraise_Occurrence.
A declaration is considered propagating if it includes elements that could propagate exceptions. This is impossible to assess fully using only static analysis, therefore the <level> parameter determines how pessimistic (or optimistic) AdaControl is in determining the possibility of exceptions. Possible values of the <level> parameter, and their effect, are:
These subrules serve several purposes:
Interface or Export pragma applies (with the given
convention(s)), and reports on those that can propagate
exceptions.
Since it is dangerous to call an Ada subprogram that can propagate exceptions from a language that has no exception (and especially C), any such subprogram should have a “catch-all” exception handler.
'Access or
'Address attribute that appears as part of an actual value for
the indicated formal. Similarly, the indicated formal can also be the
name of a formal procedure or function of a generic. In this case, the
rule will report on any subprogram that can propagate exceptions and
is used as an actual in an instantiation for the given formal.
Many systems (typically windowing systems) use call-back subprograms. Although the native interface is generally hidden behind an Ada binding, the call-back subprograms will eventually be called from another language, and like for the “interface” subrule, any such subprogram should have a “catch-all” exception handler.
Since tasks die silently if an exception is propagated out of their body, it is generally desirable to ensure that every task has an exception handler that (at least) reports that the task is being completed due to an exception.
It is sometimes desirable to make sure that no declaration raises an exception, ever.
Ex:
-- Make sure that C-compatible subprograms don't propagate exceptions:
check exception_propagation (interface, C);
-- Parameter CB of of procedure Pack.Register is used as a call-back
-- Make sure that not procedure passed to it can propagate exceptions.
check exception_propagation (parameter, Pack.Register.CB);
-- Make sure that tasks do not die silently due to unhandled exception:
check exception_propagation (task);
-- Make sure that no exception is raised by elaboration of declarations:
check exception_propagation (2, declaration);
The first example will report on any subprogram to which a
pragma Interface (C,...) applies that can propagate
exceptions.
If Proc is a procedure that can propagate exceptions, the
second example will report on every call like:
Pack.Register (CB => Proc'Access);
The third example will report on any task that can terminate silently due to an unhandled exception.
The fourth example will report on any declaration that makes use of function calls or variables.
Note that the registration procedure for a call-back can be designated by an access type, but in this case, use the name of the formal for the access type. For example, given:
package Pack is
type Acc_Proc is access procedure;
type Acc_Reg is access procedure (CB : Acc_Proc);
...
Ptr : Acc_Reg := ...;
You can give a rule such as:
check exception_propagation (parameter, Pack.Acc_Reg.CB);
All procedures registered by a call to Pack.Ptr.all will be considered.
An exception may be raised in a subprogram considered as not propagating by this rule, if an exception handler calls a subprogram that propagates an exception.
The rule will not consider subprograms whose body is missing, or that are not statically known (i.e. if a subprogram is registered through a dereference of a pointer to subprogram), like in the following example:
Pack.Register (CB => Pointer.all'Access);
Due to a weakness of the ASIS standard, references to subprograms that appear in dispatching calls are not considered. This limitation will be removed as soon as we find a way to work around this problem, but the issue is quite difficult!
These last two cases are detected by the rule “uncheckable”. See Uncheckable.
This rule controls usage of various kinds of expressions.
<control_kind> expressions (<subrule> {, <subrule>});
<subrule> ::= {<category>} <expression_kw>
<expression_kw> ::=
and | and_then |
array_aggregate | array_partial_others |
array_others | complex_parameter |
explicit_dereference | fixed_multiplying_op |
implicit_dereference | inconsistent_attribute_dimension |
inherited_function_call | mixed_operators |
or | or_else |
parameter_view_conversion | prefixed_operator |
real_equality | record_partial_others |
record_aggregate | record_others |
slice | type_conversion |
universal_range | unqualified_aggregate |
xor
<category> ::=
<> | () | range | mod | delta | digits | array |
record | tagged | access | new | private | task | protected
This rule controls usage of certain forms of expressions. The rule can be specified at most once for each subrule (i.e. subrules that accept categories can be specified once for each combination of categories and expression keyword).
Categories are used by certain subrules to further refine the control. They define categories of types to which they apply:
The subrule define the kind of expression being controlled:
and, or, xor, and_then, and or_else
control usage of the corresponding logical operator (or short circuit
form).
array_aggregate and record_aggregate control array and
record aggregates, respectively, while unqualified_aggregate
controls aggregates (both arrays and records) that do not appear
directly within a qualified expression
array_others and record_others control the occurrence of
a when others => association in array and record
aggregates, respectively.
array_partial_others and record_partial_others do the
same, but only if there are other associations in addition to the
when others => in the aggregate.
complex_parameter controls complex expressions used as actual
parameters in subprogram (or entry) calls. A complex expression is any
expression that includes a function call (including operators). This
rule is not applied to the parameters of operators, since otherwise
it would forbid any expression with more than a single operator.
explicit_dereference controls explicit dereferences of access
values (i.e. with an explicit .all).
fixed_multiplying_op controls calls to predefined fixed-point
multiplication and division (regular fixed-point or decimal-fixed
point).
implicit_dereference controls implicit dereferences of access
values (i.e. when the .all is omitted).
inconsistent_attribute_dimension controls when no dimension is
explicitely given for a 'First, 'Last, 'Range or
'Length attribute and the attribute applies to a
multi-dimensional array, or conversely, when an explicit dimension is
given, but the attribute applies to a one-dimensional array.
inherited_function_call controls calls to functions that have
been inherited by a derived type and not redefined. If a category is
specified, only calls whose result type belongs to the category are
controlled.
Derived types are followed, i.e. the “real” category from the original type is used for the matching; as a consequence, the “new” category cannot be specified for this subrule.
mixed_operators controls expressions that involve several
different operators, without parentheses. In a sense, it extends the
language rule that forbids mixing and and or in
logical expressions to all other operators.
prefixed_operator controls calls to operators that use prefixed
notation (i.e. "+"(A, B)). If a category is specified, only
calls whose result type belongs to the category are controlled.
Derived types are followed, i.e. the “real” category from the original type is used for the matching; as a consequence, the “new” category cannot be specified for this subrule.
real_equality controls usage of predefined exact equality or
inequality (“=” or “/=”) between real (floating point or fixed
point) values.
slice controls usage of array slices.
type_conversion controls all (sub)type conversions, while
parameter_view_conversion controls conversions that appear as
out or in out actual parameters. One or two
categories can be specified; if only one category is specified, only
conversions whose result type belong to that category are
controlled. If two categories are specified, only conversions whose
souce type belongs to the first category and whose target type belong
to the second category are controlled.
Derived types are followed, i.e. the “real” category from the original type is used for the matching; as a consequence, the “new” category cannot be specified for this subrule.
universal_range controls discrete ranges that are a part of an
index constraint, constrained array definition, or for-loop parameter
specification (but not type or subtype defintions), and whose bounds
are both of type universal_integer.
Ex:
search expressions (real_equality, slice);
check expressions (mixed_operators);
-- Find all conversions between integer and floating point types
search expression (range digits conversion);
-- Find all conversions from a fixed point type:
search expressions (delta <> conversion);
-- Find all view conversions between array types:
search expressions (array parameter_view_conversions);
The real_equality subrule does not control calls to an equality
operator that has been defined by the user; actually, it would make
little sense to write a function and then forbid its use! However, if
control of calls to such a function is desired, it can be easily
accomplished by using the entities rule. See Entities.
“inherited_function_call” controls only function calls. For procedure calls, see rule Statements.
This rule controls accesses to global elements that may be subject to race conditions, or otherwise shared.
<control_kind> global_references (<Ref_Kind> {, <Root>});
<Ref_Kind> ::= all | multiple | multiple_non_atomic
<Root> ::= <entity> | function | procedure | task | protected
This rule controls access to global variables from several entities
(the roots). The <entity> must be subprograms, task types,
single task objects, protected types, or single protected objects. As
usual, the whole syntax for entities is allowed for <entity>.
See Specifying an Ada entity name. The special keywords
function, procedure, task, and protected
are used to refer to all functions, procedures, tasks, and protected
entities, respectively.
If the first parameter (<Ref_Kind) is all, all references to
global elements from the indicated entities are reported. If the first
parameter is multiple, only global elements that are accessed
by more than one of the indicated entities (i.e. shared elements) are
reported. Note however that if a reference is found from a task type
or protected type, it is always reported, since there are potentially
several objects of the same type. If the first parameter is
multiple_non_atomic, references reported are the same as with
multiple, except that global variables that are atomic
or atomic_components and written from at most one of the
indicated entities are not reported. Note that this latter case
corresponds to a safe reader/writer use of atomic variables.
This rule follows the call graph, and therefore finds references from subprogram and protected calls made (directly or indirectly) from the indicated entities. However, calls to subprograms from the Ada standard library are not followed.
Ex:
-- Find global variables used by P1 or P2:
search global_references (all, P1, P2);
-- Find possible race conditions:
check global_references (multiple, task, protected);
This rule can be given several times, and conflicts (with
multiple) are reported on a per-rule basis, i.e. given:
check global_references (multiple, P1, P2);
check global_references (multiple, P1, P3);
the first rule will report on global variables shared between P1 and P2, and the second rule will report on global variables shared between P1 and P3.
The notion of “global” is relative, i.e. it designates every variable whose scope encloses (strictly) the indicated entities. This means that a same reference may or may not be global, depending on the indicated entity. Consider:
procedure Outer is
Inner_V : Integer;
procedure Inner_P is
begin
Inner_V := 1;
end Inner_P;
begin
Inner_P;
end Outer;
The rule
check global_references (all, outer);
will not report any global reference, while the rule
check global_references (all, outer.inner_p);
will report a reference to Inner_V. This is as it should be,
since there is no race condition if several tasks call Outer,
while there is a risk if several tasks (declared inside Outer)
call Inner_P.
Specifying:
check global_references (all, function);
will report on any function that access variables outside of their scope, i.e. all functions that have potential side effects. On the other hand, this check must follow the whole call graph for any function encountered, and can therefore be quite costly in execution time.
Calls through pointers to subprograms and dispatching calls are unknown statically; they are assumed to not access any global. Such calls are detected by the rule “uncheckable”. See Uncheckable.
This rule controls that every compilation unit starts with a standardized comment.
<control_kind> header_comments (minimum, <comment lines>);
<control_kind> header_comments (model, "<file name>");
If the keyword “minimum” is given as first parameter, this rule controls that every compilation unit starts with at least the number of comment lines indicated by the second parameter. If several forms of headers are possible, checking that the headers follow the project's standard requires manual inspection, but this rule is useful to control that unit headers have not been inadvertantly forgotten.
If the keyword “model” is given as first parameter, the second
parameter is a string, interpreted as a file name. If the file name is
not an absolute path, it is interpreted as relative to the directory
of the file that contains the rule, or to the current directory if the
rule is given on the command line. Each line of the indicated file is
a regular expression, and the rule controls that the corresponding
line of the source file matches the expression. See Syntax of regular expressions. In addition, it is possible to specify a repetition
for a line. If the first character of a line is a '{', the line
must have the following syntax:
{<min>,[<max>]}
where <min> and <max> specify the minimum and maximum number of occurrences of the pattern in the line that follows this one. <min> must be at least 0, and <max> must be at least 1, and be equal or greater than <min>. If <max> is omitted, it means that the line may occur any number of times.
As a convenience, if the first character of a line is a '*'
it means that the next line is a pattern that can occur any number of
times (same as {0,}). If the first character is a
'+', it means that the next line is a pattern that must occur
at least once (same as {1,}). If the first character is a
'?', it means that the next line is an optional pattern (same
as {0,1}).
Note that the repetition lines all start with a special character
which is not allowed at the start of a regular expression; there is
therefore no ambiguity. Everything after the special character (or the
closing '}') is ignored, and can be used to provide comments.
This rule can be given at most once with “minimum” for each of “check”, “search”, and “count”. The rule can be given only once with “model” (but it can be given together with one or more “minimum” rules).
Ex:
check header_comments (minimum, 10);
search header_comments (model, "header.pat");
count header_comments (minimum, 20);
This makes an error for every unit that starts with less than 10
comment lines, and a warning for units that do not follow the pattern
contained in the file header.pat. A count of units that start
with less than 20 comment lines is reported.
Example of a pattern file:
{1,3} 1 to 3 occurrences of next line
^--$
^-- Author: .+$
^-- Date: \d{2}/\d{2}/\\d{4}$
Remember that the lines of the file are regular expressions; every
character that is specially interpreted (like “+”, “*”, etc.) must
be quoted with “\” if it must appear textually. To ease the process
of generating the model file, the directory source contains a
script file for sed named makepat.sed; if you run this script
on a file that contains a standard header, it will produce a pattern
file where each line starts with “^”, ends with “$”, and every
special character is quoted with “\”.
When the model contains an indication of repeated lines (“*”), the repetition is not “greedy”, i.e. matching will stop as soon as what follows the repetition matches. This is very useful to check header comments that have sections, but where you don't want to impose a precise content to each section. Imagine for example that the structure is:
^-- HISTORY$
*
^--
^-- AUTHORS
*
^--
Since the “model” subrule analyzes the content of comments, there is a conflict with the disabling mechanism of AdaControl that uses special comments. See Disabling controls.
Specifically, line disabling is not possible at all. Block disabling is possible, provided the disabling line is allowed by the pattern. In short, if you want to be able to disable this rule, the first lines of the model file should be:
?
--##
i.e. allow an optional block disabling comment as the first line of the file. Note that there is no need to re-enable this rule, since it is checked only at the start of a compilation unit.
This rule enforces a coding pattern that ensures that variables and out parameters are properly initialized befor use.
<control_kind> improper_initialization [(<subrule> {,<subrule>})]
<subrule> ::= {<extra>} <target>
<extra> ::= access | limited | package
<target> ::= out_parameter | variable | initialized_variable
This rule controls variables and/or out parameters that are not “properly” initialized, i.e. those that are not “safely” initialized, those that have a useless initialization in their declaration, and those where the value is known to be used before having been assigned.
A variable (or out parameter) is considered safely initialized if there is an initialization expression in its declaration, or if it is given a value in the first statements of the corresponding body, before any “non-trivial” statement. The goal is not to perform a complete data-flow analysis, but rather to follow a design pattern where all variables are initialized before entering the “active” part of the algorithm. This makes it easier to ensure that variables are properly initialized.
“Trivial statements” are:
null statements;
out_parameter controls that out parameters are
safely initialized before the first non-trivial statement, and before
every (trivial) return statement.
variable controls that local variables are
safely initialized before the first non-trivial statement.
initialized_variable controls variables that are safely
initialized before the first non-trivial statement, but also have an
explicit (and therefore useless) explicit initialization in their
declaration.
A variable is considered initialized if it is the target of an assignment statement, or if it is used as an actual for an out (but not in out) parameter of a procedure call. Variables assigned in if or case statements must receive a value in all paths to be considered initialized after the statement. Note that the variable must be assigned to globally, i.e. assigning to some elements of an array, or some fields of a record, does not count as an initialization of the variable.
Variables declared immediately within a (generic) package specification or body are not checked, unless the modifier “package” is specified (quite often, package state variables are initialized through calls to dedicated procedures).
Normally, the rule does not control objects of an access types, or
arrays whose components are of an access type, since these are always
initialized by the compiler. Similarly, the rule does not control
objects of a limited type, since global assignment is not available
for them. The <extra> modifiers access and limited force
the controlling of these two kinds of objects, respectively.
This rule can be given only once for each value of <target>. Without parameters, it is equivalent to giving all, without any <extra>.
Ex:
check improper_initialization (out_parameter);
check improper_initialization (access limited variable);
search improper_initialization (initialized_variable);
Due to a weakness of the ASIS standard, dispatching calls and calls to
procedures that are attributes are not considered for the
initialization of variables. Note that for attributes, only
'Read and 'Input have an out parameter.
In the rare case where a variable is initialized by a dispatching call or an attribute call, this limitation will result in a false positive. Such a case is detected by the rule “uncheckable”. See Uncheckable. It is then easy to disable the rule for this variable. See Disabling controls.
The rule analyzes only initializations and uses that are directly in the unit, not those from nested units, since these are in the general case not statically checkable.
There are other cases where an object is automatically initialized by
the declaration, like controlled types that have redefined the
Initialize procedure, records where all components have a
default initialization, etc. The rule does not consider these as
automatically initialized, as it does for access types. Maybe later...
This rule controls all instantiations of a generic, or only instantiations that are made with specific values of the parameters. Control can be restricted to instantiations in specified places.
<control_kind> instantiations (<generic_spec>);
<generic_spec> ::= {<location_kw>} <entity> {, <formal_spec>}
<formal_spec> ::= <entity> | <category> | <> | =
<location_kw> ::= all | block | library | local | nested |
own | private | public | task_body
<category> ::= () | access | array | delta | digits | mod |
private | protected | range | record | tagged | task
The rule controls instantiations of the specified <entity>. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
The <location_kw> restricts the places where the occurrence of the instantiation is controlled. Several <location_kw> can be given, in which case the instantiation is controlled at places where all the keywords apply. If there is no <location_kw>, it is assumed to be “all”.
all: puts no special restriction to the location. This keyword
can be specified for readability purposes, and if specified must
appear alone (not with other <location_kw>).
block: only instantiations appearing in block statements are
controlled.
library: only library level instantiations are controlled.
local: only local instantiations are controlled (i.e. only
instantiations appearing in (generic) packages, possibly nested, are
allowed).
nested: only instantiations nested in another declaration are
controlled (i.e. only library level instantiations are allowed).
own: only instantiations that are local to a (generic) package
body are controlled.
public: only declarations appearing in the visible part of
(generic) packages are controlled.
private: only instantiations appearing directly in a private
part are controlled.
task_body: only instantiations appearing directly in a task
body are controlled. Note that it would not make sense to have a
<location_kw> for task specifications, since instantiations are
not allowed there.
An instantiation matches if it appears at a specified location (if any) and either:
In addition, two special signs can be given instead of an <entity> (or
<category>): a box (<>) matches any actual parameter
(i.e. it stands for any value), and an equal sign (=) matches
if there has been already an instantiation with the same value for
this parameter (i.e. it matches the second time it is encountered).
If an actual is an expression (which is possible only for a formal in object), it cannot be matched.
Ex:
-- Check all instantiations of Unchecked_Deallocation:
search instantiations (ada.unchecked_deallocation);
-- Check all instantiations of Unchecked_Conversion from or to String:
check instantiations (ada.unchecked_conversion, standard.string);
check instantiations (ada.unchecked_conversion, <>, standard.string);
-- Check all instantiations of Unchecked_Conversion from address
-- to an integer type:
check instantiations (ada.unchecked_conversion, system.address, range);
-- Check that Unchecked_Conversion is instantiated only once
-- for any pair of arguments:
check instantiations (ada.unchecked_conversion, =, =);
The various forms of <formal_spec> make the rule quite powerful. For example:
-- Not two instantiations of Gen with the same first parameter:
check instantations (Gen, =);
-- Not two instantiations of Gen with the same first and third parameter:
check instantiations (Gen, =, <>, =);
-- Not two instantiations of Gen with the same first parameter if the
-- second parameter is Pack.Proc:
check instantiations (Gen, =, Pack.Proc);
-- Not two instantiations of Gen with the same first parameter if the
-- second parameter is any procedure named Proc:
check instantiations (Gen, =, all Proc);
Note that a generic actual wich is a subtype matches all types (and subtypes) above it. Therefore,
check instantiations (ada.unchecked_deallocation (standard.natural));
will find only instantiations that use Natural, while:
check instantiations (ada.unchecked_deallocation (standard.integer));
will find instantiations that use either Integer,
Positive, or Natural.
Gnat defines Unchecked_Conversion and
Unchecked_Deallocation as separate entities, rather than
renamings of Ada.Unchecked_Conversion and
Ada.Unchecked_Deallocation. As a consequence, it is necessary
to specify explicitely both forms if you want to make sure that the
corresponding generics are not instantiated.
This rule controls calls to subprograms and entries where the values of parameters does not provide sufficient information to the reader to correctly identify the parameter's purpose.
<control_kind> insufficient_parameters (<max_allowed> {, <entity>});
<max_allowed> is the maximum number of allowed “insufficient” parameters (can be 0). The <entity> parameters designate enumeration types whose values should be included in the check. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
An actual parameter is deemed "insufficient" if it is given in
positional (as opposed to named) notation, it is an expression whose
primaries are all numeric literals, or enumeration literals belonging
to one of the types passed as parameters to the rule
(Standard.Boolean for example).
This rule can be given once for each of check, search, and count. This way, it is possible to have a level considered a warning (search), and one considered an error (check).
Ex:
search Insufficient_Parameters (1, Standard.Boolean);
check Insufficient_Parameters (2, Standard.Boolean);
This rule does not apply to operators that use infix notation, nor to calls to subprograms that are attributes, since named notation is not allowed for these.
This rule controls the use of positional parameters according to their
values; it is also possible to control the use of positional
parameters according to the number of parameters with the rule
style (positional_association). See Style.
Note also that this rules applies only to calls, while style
(positional_association) applies to all forms of associations.
This rule controls declarations that hide an outer declaration with the same name.
<control_kind> local_hiding [(<subrule> {,<subrule>})];
<subrule> ::= {<exception>} strict | overloading | overloading_short
<exception> ::= not_operator | not_enumeration
If “strict” is given (or if there is no subrule), the rule controls strict hiding (an inner subprogram that overloads an outer one is not considered hiding). If “overloading” or “overloading_short” is given, only subprograms that overload another subprogram in the same scope or in an outer scope are controlled. Note that following the normal Ada model, the declarations of enumeration literals are considered functions (and thus controlled).
If “not_operator” is mentionned, the subrule does not apply to the
declarations of operators (i.e. things like “"+"”). If
“not_enumeration” is mentionned, the subrule does not apply to the
hiding/overloading of enumeration literals by other enumeration
literals (the rule still applies to the hiding/overloading of
functions by enumeration litterals, for example).
“overloading” and “overloading_short” behave the same, except for the way occurrences are reported. When a construct that overloads several other constructs is encountered, “overloading” will issue a message for each overloaded construct, while “overloading_short” will issue a single message mentionning how many constructs are overloaded, and a pointer to the last one.
This rule can be given only once for “strict” and once for either “overloading” or “short_overloading”.
Ex:
Hiding: check local_hiding (strict);
Overloading: search local_hiding (not_operator overloading);
This rule controls excessive spacing in the program text.
<control_kind> max_blank_lines (<max allowed blank lines>);
This rule controls the occurrence of more than the indicated number of consecutive blank lines (empty lines, or lines that contain only spaces). This rule can be given once for each of check, search, and count. This way, it is possible to have a number of blank lines considered a warning (search), and one considered an error (check). Of course, this makes sense only if the number for search is less than the one for check.
Ex:
search max_blank_lines (2);
check max_blank_lines (5);
This rule controls the maximum depth of subprograms (or entry) calls.
<control_kind> max_call_depth (<allowed depth> | finite);
Roughly, the call depth is the number of frames that are stacked by a call: if you call a subprogram that calls another subprogram that calls nothing, then the call depth is 2. Note that a call to a task (not protected) entry has always a depth of 1, since the accept body that corresponds to the entry is executed on a different stack.
The value of the parameter is the maximum allowed depth, i.e. the rule will trigger if the call depth is strictly greater than the indicated value. A call to a (directly or indirectly) recursive procedure is considered of infinite depth, and will be therefore signaled (with an appropriate message) for any value of <allowed depth>. Alternatively, the keyword “finite” can be given in place of the <allowed depth>: in this case, only calls to recursive subprograms will be signalled.
This rule can be given once for each of check, search, and count. This way, it is possible to have a call depth considered a warning (search), and one considered an error (check). Of course, this makes sense only if the number for search is less than the one for check.
Ex:
search max_call_depth (9);
check max_call_depth (finite);
It is possible to give the value 0 for <allowed depth>. Of course, it would not make sense to forbid all subprogram calls in an Ada program, but this can be useful for inspection purposes, since every call will be reported, and the message indicates the depth of the call.
If the message says that the call depth “is N”, it is exactly N. If the message says that the call depth is “at least N”, it means that the call chain includes a call to a subprogram whose depth is unknown (see “Limitations” below); “N” is the call depth if this subprogram does not call anything else. Of course, the rule issues a message if this minimal value is greater than the maximum allowed value.
Calls to subprograms that are attributes are assumed to have a depth of 1. Calls to predefined operators are assumed to be in-lined (i.e. a depth of 0).
Calls through pointers to subprograms and dispatching calls are unknown statically; in addition, some subprograms may not have a body available for analysis, like imported subprograms, or possibly subprograms from the standard library; they are all assumed to have a depth of 1. Such calls are detected by the rule “uncheckable”. See Uncheckable.
This rule controls that no line exceeds a given length.
<control_kind> max_line_length (<max allowed length>);
This rule controls the maximum length of source lines. This rule can be given once for each of check, search, and count. This way, it is possible to have a length considered a warning (search), and one considered an error (check). Of course, this makes sense only if the length for search is less than the one for check.
Ex:
search max_line_length (80);
check max_line_length (120);
This rule controls excessive nesting of declarations.
<control_kind> max_nesting (<max allowed depth>);
This rule controls the nesting of declarative constructs (like subprograms, packages, generics, block statements...) that exceed a given depth. Nesting of statements (loop, case) is not considered. This rule can be given once for each of check, search, and count. This way, it is possible to have a level considered a warning (search), and one considered an error (check). Of course, this makes sense only if the level for search is less than the one for check.
Ex:
search max_nesting (5);
check max_nesting (7);
This rule controls the maximum size, in source lines of code, of various statements.
<control_kind> max_size (<subrule>, <max allowed lines>);
<subrule> ::= accept | block | case | case_branch |
if | if_branch |loop | simple_block |
unnamed_block | unnamed_loop
The first parameter is a subrule keyword that determines which statements are controlled:
For each kind of statement, the indicated value is the maximum allowed size of the full statement; however, for branches (“if_branch” and “case_branch”) it is the maximum size of the sequence of statements in the branch (i.e., the line that contains the elsif is not counted as part of an “if_branch”).
This rule can be given once for each of check, search, and count for each kind of statement. This way, it is possible to have a level considered a warning (search), and one considered an error (check). Of course, this makes sense only if the number of lines for search is less than the one for check.
Ex:
check Max_Size (if_branch, 30);
search Max_Size (if_branch, 50);
check Max_Size (unnamed_loop, 20);
This rule controls the nesting of compound statements.
<control_kind> max_statement_nesting (<subrule>, <max allowed depth>);
<subrule> ::= block | case | if | loop | all
If one of “block”, “case”, “if”, or “loop” is specified, it controls the nesting of statements of the same kind, i.e. an if within a loop within an if counts only 2 for the “if” keyword. If “all” is specified, all kinds of compound statements are counted together, i.e. an if within a loop within an if counts for 3. This rule can be given once for each of check, search, and count, and for each of the subrules. This way, it is possible to have a level considered a warning (search), and one considered an error(check). Of course, this makes sense only if the level for search is less than the one for check.
Ex:
check max_statement_nesting (loop, 3);
search max_statement_nesting (all, 5);
This rule controls statements that are inside accept statements and could safely be moved outside.
<control_kind> movable_accept_statements (certain|possible {, <entity>})
Since it is good practice to block a client for the shortest time possible, any action that does not depend on the accept parameters should not be part of an accept statement.
Statements that involve synchronisation (delay statements, accept or entry calls...) are not movable. Statements (including compound statements) that reference the parameters of the enclosing accept are not movable. In addition, statements that use one of the <entity> given as parameters are never considered movable. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name. Note that if a generic entity, or an entity declared in a generic package, is given, all statements that use the corresponding instantiated entity are considered not movable.
If the first parameter of the rule is certain, only statements
after the last non-movable statement are reported. If the first
parameter is possible, a simple data flow analysis is
performed, and every statement that does not reference a variable that
appears to depend (directly or indirectly) on a parameter is also
reported.
Ex:
check movable_accept_statements (possible, Log.Report_Rendezvous);
The list of <entity> given to the rule can be, for example, procedures whose execution must be part of the accept statement for logical reasons. They can also be global variables, when the rendezvous is intended to prevent concurrent access to these variables.
This rule controls groups of assignments that are either redundant of replaceable by aggregate assignment.
<control_kind> multiple_assignments (repeated);
<control_kind> multiple_assignments (groupable, <filter> {,<filter>});
<filter> ::= given <min_val> | missing <max_val> | ratio <min_val>
This rule controls properties of groups of assignment statements. A group is made of consecutive assignments, without any other intervening kind of statements (except null statements).
The first form (keyword “repeated”) controls when a same variable (or a same subcomponent of a structured variable) is assigned several times in the same group of assignments. This form of the rule can be given only once.
The second form (keyword “groupable”) controls assignments to different subcomponents of a same structured variable; such assignments are often replaceable by a global assignment of an aggregate to the variable. One or several <filter> parameters indicate under which conditions a group is reported:
If several filters are given, the rule is triggered if all conditions are met (“and” logic). Note however that this rule can be given several times, thus achieving “or” logic.
The rule is not triggered if the subcomponents belong to a stuctured variable of a limited type, since global assignment would not be allowed in that case.
Ex:
check Multiple_Assignments (repeated);
-- Warn if a at least 3 fields are given and at most
-- two fields are missing, or if 80% of the fields are given:
search multiple_assignments (groupable, given 3, missing 2);
search multiple_assignments (groupable, ratio 80);
Note that it is possible to give 1 for the “given” criterion; in this case, any assignment to parts of a structured variable will be reported, only global assignment is allowed.
As usual, AdaControl can control only static aspects of assignments. Therefore, it cannot control assignments whose target is not statically known (like dynamic indexing of arrays). Slices are always considered dynamic (the cases where it would be useful did not seem worth the additional complexity).
Similarly, if the number of subcomponents is not statically determinable (dynamic arrays, discriminated records), only the “given” criterion can be met.
This rule controls the form of identifiers to make sure that they follow the project's naming conventions. Different naming conventions can be specified, depending on the kind of Ada entity that the name is refering to.
<control_kind> naming_convention
([root] [others] {<location>} <filter_kind>,
[case_sensitive|case_insensitive] [not] "<pattern>"
{, ...});
<location> ::= global | local | unit
<filter_kind> ::= All |
Type |
Discrete_Type |
Enumeration_Type |
Integer_Type |
Signed_Integer_Type |
Modular_Integer_Type |
Floating_Point_Type |
Fixed_Point_Type |
Binary_Fixed_Point_Type |
Decimal_Fixed_Point_Type |
Array_Type |
Record_Type |
Regular_Record_Type |
Tagged_Type |
Class_Type |
Access_Type |
Access_To_Regular_Type |
Access_To_Tagged_Type |
Access_To_Class_Type |
Access_To_SP_Type |
Access_To_Task_Type |
Access_To_Protected_Type |
Private_Type |
Private_Extension |
Generic_Formal_Type |
Variable |
Regular_Variable |
Field |
Discriminant |
Record_Field |
Protected_Field |
Procedure_Formal_Out |
Procedure_Formal_In_Out |
Generic_Formal_In_Out |
Constant |
Regular_Constant |
Named_Number |
Integer_Number |
Real_Number |
Enumeration |
Sp_Formal_In |
Generic_Formal_In |
Loop_Control |
Occurrence_Name |
Entry_Index |
Label |
Stmt_Name |
Loop_Name |
Block_Name |
Subprogram |
Procedure |
Regular_Procedure |
Protected_Procedure |
Generic_Formal_Procedure |
Function |
Regular_Function |
Protected_Function |
Generic_Formal_Function |
Entry |
Task_Entry |
Protected_Entry |
Package |
Regular_Package |
Generic_Formal_Package |
Task |
Task_Type |
Task_Object |
Protected |
Protected_Type |
Protected_Object |
Exception |
Generic |
Generic_Package |
Generic_Sp |
Generic_Procedure |
Generic_Function |
Renaming |
Object_Renaming |
Exception_Renaming |
Package_Renaming |
Subprogram_Renaming |
Procedure_Renaming |
Function_Renaming |
Generic_Renaming |
Generic_Package_Renaming |
Generic_Sp_Renaming |
Generic_Procedure_Renaming |
Generic_Function_Renaming
The first parameter defines the kind of declaration to which the rule is applicable, and other parameters are strings, interpreted as regular expressions that define the patterns that must be matched. See Syntax of regular expressions.
If one or more <Location> keyword is specified, the pattern applies only to identifiers declared at the corresponding place. Otherwise, the pattern applies to all identifiers, irrespectively of where they are declared. The definition of locations is as follows:
If “case_sensitive” is specified, pattern matching considers casing. Otherwise (default or “case_insensitive”), casing is irrelevant. Note that the rule checks the name only at the place where it is declared; casing might be different when the name is used later.
If a pattern is preceded by “not”, then the pattern must not be matched (i.e. the rule reports when there is a match).
The rule will be activated if an identifier is declared that does not match any of the “positive” patterns (the ones without “not”), or if it matches any of the ”negative” patterns (the ones with a “not”). If only negative patterns are given, it is implicitely assumed that all other identifiers are OK. In other words, accepted identifiers must have the form of (at least) one of the “positive” patterns (if any), but not the form of one of the “negative” patterns.
The filter kinds are organized hierarchically, as reflected in the syntax above. To be valid, the name must match the patterns specified for its own filter, and for all filters above it in the hierarchy. For example, a modular type declaration must follow the rules (if specified) for “all”, “type”,”discrete_type”, “integer_type” and “modular_integer_type”. However, if a filter kind is preceded by “others”, the rule will apply only if there is no applicable positive pattern deeper in the hierarchy; similarly, if a filter kind is preceded by “root”, no rule above it in the hierarchy is considered (neither for itself nor its children). This is useful to make exceptions to a more general rule. For example:
-- All identifiers must have at least 3 characters:
check naming_convention (all, "...");
-- And start with an upper-case letter
-- (will not apply to types and access types, because of "others" and
-- other rules given below)
check naming_convention (others all, case_sensitive "^[A-Z]");
-- Exception to the rule for "all":
-- No minimum length for "for loop" identifiers, but must be
-- all uppercase
check naming_convention (root loop_control, case_sensitive "^[A-Z]+$");
-- Types must start with "t", then an upper-case letter:
-- (will not apply to access types, because of "others" and
-- other rule given below)
check naming_convention (others type, case_sensitive "^t[A-Z]");
-- Access types must start with "ta", then an upper-case letter:
check naming_convention (access_type, case_sensitive "^ta[A-Z]");
It is of course not necessary to specify all the filter kinds, nor to specify filters down to the deepest level; if you specify a rule for “type”, it will be applied to all type declarations, whether there is a more specific rule or not.
Subtypes and derived types must follow the rule for their respective original (full) type. Incomplete type declarations are not checked, since their corresponding full declaration is (normally) checked. Private types (including of course the full declaration of a private type) follow the rule for private types, not the rules for their full type view (otherwise it would be privacy breaking).
Renamings are treated specially: if there is no explicit rule for a given renaming, the applicable rule is the one for the renamed entity.
Ex:
-- Predefined name is forbidden:
check naming_convention (all, not "Integer");
-- Types must either start or end with T
check naming_convention (type, case_sensitive "^T_",
case_sensitive "_T$");
-- "Upper_Initials" naming convention:
check naming_convention
(all, case_sensitive "^[A-Z][a-z0-9]*(_[A-Z0-9][a-z0-9]*)*$");
-- All global variables must start with "G_"
check naming_convention (global variable, "G_");
The rule only checks the casing of identifiers at the place where they are declared. A useful companion rule is “style (casing_identifier, original)”, which ensures that every use of the identifier will use the same casing as in the declaration. See Style.
Remember that a Regexp matches if the pattern matches any part of the identifier. Use “^” and “$” to match the beginning (resp. end) of the name, or both.
“class_type” is applicable to subtypes that designate a class-wide type. Similarly, “access_to_class_type” is applicable to access types whose designated type is class-wide.
If you don't want any special rule for renamings (not even the one that applies to the renamed entity), specify:
check (renaming, "");
This imposes no constraint on renamings, but since it is specified explicitely, the implicit rule for the renamed entity won't apply.
The rules directory of Adacontrol contains two files named
no_standard_entity.aru and no_system_entity.aru. These
are files that contain a naming_convention rule that forbids the
declaration of names declared in packages Standard and System,
respectively. You can simply “source” these files from your own rule
file (or copy the content) if you want to disallow these identifiers.
Like usual, naming_convention rule can be given multiple times, and can be disabled. However, consider the following:
Rule1 : check naming_convention (constant, "^c_");
Rule2 : check naming_convention (constant, "^const_");
The rule will trigger if a constant is declared that does not start with either “c_” or “const_”. But here, we have two different rule labels. The message will refer to the first label encountered in the rule file; this is the label that must be mentionned in a disabling comment, unless you simply disable “naming_convention”.
This rule does not support wide characters outside the basic Latin-1 set.
This rule controls types whose usage profile indicates that they might be replaceable with enumerated types.
<control_kind> no_operator_usage [(none | logical)];
This rule controls integer types (both signed and modular) where no
operator of the type is used in the program (if the keyword none
is given), or only logical operators (if the keyword logical is
given). If no parameter is given, the rule will control both cases.
An integer type that uses no operator at all is a good candidate to be replaced by an enumerated type. A modular type where only logical operators are used is likely to be used as a bit field or a set, and is a good canditate for being replaced by an array of booleans.
This rule can be given only once for each value of the parameters.
Ex:
check no_operator_usage;
The rule does not make a distinction between predefined and user-defined operators. On the other hand, only call to operators are considered, operators used for example as actual generic parameters in instantiations are not considered.
The rule applies to private types whose full declaration is an integer type.
This rule controls that expressions used in certain contexts are static.
<control_kind> non_static [(<subrule> {, <subrule>})];
<subrule> ::= constant_initialization | variable_initialization |
index_constraint | discriminant_constraint |
instantiation | index_check
The <subrule> defines the elements that are required to be static:
If no keyword is given, all contexts are controlled.
Ex:
check non_static (index_constraint);
Currently, “constant_initialization” and “variable_initialization” do not control structured (record and array) variables. For access variables, the initial value is considered static only if it is a plain null. This may improve in future versions of AdaControl.
If all index and discriminant constraints are static, the space occupied by data structures is computable from the program text. This rule is useful to enforce this in contexts where the memory space must be statically determined.
This rule controls that certain subprograms (or allocators) are called only during program initialization.
<control_kind> not_elaboration_calls (<entity>|new {, <entity>|new});
The <entity> parameters are callable entities (procedure, function or entry calls). As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name. This rule controls calls to the indicated callable entities, or allocators if “new” is given, that are performed at any time except during the elaboration of library packages.
Ex:
search not_elaboration_calls (Data.Initialize, new);
Due to an (allowed by ASIS standard) limitation of ASIS-for-Gnat, the
rule will not detect calls to subprograms that are implicitely
defined, like calling a "+" on Integer. Fortunately,
it is very unlikely that the user would want to forbid that kind of
calls in non-elaboration code.
Note also that calls that cannot be statically determined, like calls to dispatching operations or calls through pointers to subprograms cannot be detected either.
This rule controls that certain entities are always refered to using selected notation, even in the presence of use clauses.
<control_kind> not_selected_name
(<exception places>, <entity> {, <entity>});
<exception places> ::= none | unit | compilation | family
A name is “selected” if it is prefixed by the name of the construct where it is declared. Only one level of prefix is required, unless the prefix itself is the target of a not_selected_name rule.
The first parameter specifies places where the rule is not enforced, i.e. where simple notation is allowed:
Other parameters indicate the <entity> to which the rule applies. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
Ex:
check not_selected_name (unit, all Instance);
search not_selected_name (none, Pack.T);
Note that, as usual, the entity can be given in the form “all
name”. This is especially useful for types that must always be
declared with a special name (like Instance, Object,
T) and are intended to be always used with the name of the
enclosing package.
This rule controls various aspects of object (constants and variables) declarations.
<control_kind> object_declarations (min_integer_span, <param> {, <param>});
<control_kind> object_declarations (volatile_no_address);
<control_kind> object_declarations (address_not_volatile);
<param> ::= [all | constant | variable] <value>
The action depends on the subrule.
This subrule can be given only once for each combination of check/search/count and constant/variable.
Since this subrule has no parameters, it can be given only once.
Since this subrule has no parameters, it can be given only once.
Ex:
check object_declarations (min_integer_span, variable 5, constant 10);
count object_declarations (min_integer_span, 8);
-- Same value for variables and constants
search object_declarations (volatile_no_address);
search object_declarations (address_not_volatile);
The “min_integer_span” rule can be useful for detecting variables that should use an enumerated type rather than an integer type.
Due to a shortcomming of the ASIS interface, the subrules “volatile_no_address” and “address_not_volatile” will not detect variables of a class-wide type that are volatile due to a pragma volatile applying to the class-wide type. If the pragma applies to the variable, the subrule will work correctly. A pragma volatile applied to a class-wide type is detected by the rule “uncheckable”. See Uncheckable.
Declaring a class-wide type as volatile seems very peculiar anyway...
This rule controls aliased use of variables in subprogram calls.
<control_kind> parameter_aliasing [([with_in] <level>)];
<level> ::= Certain | Possible | Unlikely
This rule identifies calls where the same variable is given as an actual to more than one out or in out parameter, like in the following example:
procedure Proc (X, Y : out Integer);
...
Proc (X => V, Y => V);
If the modifier “with_in” is given, aliasing between
out or in out parameters and in
parameters is also considered. Although this is generally considered
of less of an issue, it can lead to unexpected results when the
in parameter is passed by reference.
There are many cases where aliasing cannot be determined statically. The optional parameter specifies how aggressively the rule will check for possible aliasings. Possible values are (case irrelevant):
Swap (Tab (I), Tab (J));
there is no aliasing, unless I equals J.
If all expressions used for indexing in both variables are static, the rule will be able to eliminate the diagnosis of aliasing (if the values are different). This avoids unnecessary messages in cases like:
Swap (Tab (1), Tab (2));
type R is
record
X : aliased Integer;
end record;
X : R;
Y : Access_All_Integer := R.X'access;
...
P (X, Y.all);
The rule may be specified at most once for each value of the parameter. This allows for example to “check” for “Certain” and “search” for “Possible”.
Ex:
check parameter_aliasing (with_in certain);
search parameter_aliasing (Possible);
Note that the rule is quite clever: it will consider partial aliasing (like a record variable as one parameter, and one of its components as another parameter), and will not be fooled by renamings.
Due to a weakness of the ASIS standard, dispatching calls are not analyzed. Some calls cannot obviously have aliasing (if there is only one parameter, or if there are no variables in the parameters f.e.); other calls are detected by the rule “uncheckable”. See Uncheckable.
This rule controls various characteristics of the declaration of parameters for all callable entities (i.e. functions, procedures and entries).
<control_kind> parameter_declarations (<subrule>, [<value>,] {,<callable>});
<subrule> ::= max_parameters | max_defaulted_parameters |
min_parameters | single_out_parameter
<callable> ::= function | procedure | protected_entry |
protected_function | protected_procedure | task_entry
The action depends on the first parameter:
If one or more <callable_kind> is specified after the <value>, the rule applies only to the corresponding declaration(s), otherwise it applies to all callable entities.
This rule can be given once for each of check, search, and count for each subrule and each kind of entity. This way, it is possible to have a level considered a warning (search), and one considered an error (check).
Ex:
-- Callable entities should preferably not have more than 5
-- parameters, and in any case not have more that 10 parameters,
check parameter_declarations (max_parameters, 10);
search parameter_declarations (max_parameters, 5);
-- All functions must have parameters:
check parameter_declarations (min_parameters, 1, function);
-- A regular (not protected) procedure with one out parameter
-- should be replaced by a function
check parameter_declarations (single_out_parameter, procedure);
This rule applies to generic subprograms as well as to regular ones. On the other hand, it does not apply to generic formal subprograms, since instantiations would only be possible with subprograms which are supposed to have been already controlled.
Instantiations are also controlled; the number of parameters is taken from the corresponding generic.
Note that this rule controls only “regular” parameters, not generic formal parameters.
This rule controls usage of potentially blocking operations (as defined in LRM 9.5.1 (8..16)) from within protected operations.
<control_kind> potentially_blocking_operations;
The rule follows the call graph, starting from every protected operation, and identifies all (direct and indirect) potentially blocking operations encountered. All protected types in the program are controlled.
Of course, calls to standard subprograms (notably IOs) that are defined to be potentially blocking are recognized.
Ex:
check potentially_blocking_operation;
This rule is very clever at finding potentially blocking operations resulting from external calls (or requeues) to the current protected object, even if this happens through a long chain of subprogram calls. Typically, this happens when a protected operation calls a subprogram, which in turn makes a call to an operation of the same protected object. Such calls generally result in dead-locks.
Therefore, it is advisable to run this rule on any program that exhibits mysterious (and hard to find) deadlocks that seem to involve protected objects.
When a single protected object is being analyzed, the rule will diagnose a circularity if there is a call to an operation of the same object in the call chain; however, if a protected type is being analyzed, the rule will diagnose a circularity if there is a call to any object of the same type in the call chain. Although it is possible to construct examples of this latter case where there is no risk of deadlock, it is so contrieved that it certainly deserves being looked at. But since the call is not 100% certain to be potentially blocking, the message will tell “possible external call” instead of “external call” in this case.
There is one case defined in LRM E.4(17) which is not recognized: remote subprograms calls.
Calls through pointers to subprograms and dispatching calls are unknown statically; they are assumed to be non potentially blocking. Such calls are detected by the rule “uncheckable”. See Uncheckable.
This rule controls usage of one or several specific pragmas.
<control_kind> pragmas
(all|nonstandard|<pragma name> {, <pragma name>});
If the special name “nonstandard” is given, then all implementation-defined and unrecognized pragmas will be controlled. If the special name “all” is given, then all pragmas will be controlled. Otherwise, the parameters are the names of pragmas to be controlled. Note that <pragma name> must be the simple name of the pragma, since pragma names are predefined and do not follow the rules for regular Ada entities.
Ex:
check pragmas (elaborate_all, elaborate_body);
If “all” and/or “nonstandard” is given together with a specific pragma name in a “search” or “check” rule, a message is issued only for the most specific occurrence. However, for “count”, all appropriate occurrences are counted, i.e. given the following rules:
C1 : count pragmas (annotate);
C2 : count pragmas (nonstandard);
C3 : count pragmas (all);
Counter C1 will report the number of occurrences of pragma
Annotate (a non-standard Gnat pragma), counter C2 will report the
number of non-standard pragmas (including occurrences of
Annotate), and counter C3 will report the total number of
pragmas (including occurrences of Annotate).
This rule controls various aspects of the components of records.
<control_kind> record_declarations (component, <compo_kind> {,<repr_cond>});
<compo_kind> ::= <entity>|<category>
<category> ::= () | access | array | delta | digits | mod | private |
protected | range | record | tagged | task
<repr_cond> ::= [not] in_variant | aligned | initialized | packed | sized
The first parameter is a subrule keyword:
Storage_Unit.
Storage_Unit.
This rule can be specified several times for the “component” subrule.
Ex:
-- All record components of a discrete type should be initialized:
check record_declarations (component, (), not initialized);
-- The size of all components of type Hardware_Types.Squeezed must
-- have a component clause:
check record_declarations (component, Hardware_Types.Squeezed, not sized);
-- Find unaligned components of a packed array type:
check record_declarations (component, array, packed, not aligned);
It may seem strange to have a rule with only one subrule, but we expect to add more in the near future. Stay tuned...
If “[not] aligned” is specified, there are some rare cases where AdaControl cannot evaluate whether a component is aligned or not; in this case, it will “assume the worse” (i.e. report as if the component had the specified alignment), thus creating possible false positives. Such cases are detected by the rule “uncheckable”. See Uncheckable.
This rule controls declarations that could be moved to some inner scope.
<control_kind> reduceable_scope [(<subrule> {, <subrule>})];
<subrule> ::= {<restriction>} all | variable | constant |
subprogram | type | package |
exception | generic | use
<restriction> ::= no_blocks | to_body
The rule reports on any declaration that is referenced only from a single, inner scope, or in the case of use clauses, it will report on packages named in a use clause whose elements are used only in a single, inner scope. For entitities declared in package specifications, the rule reports if they are used only from the corresponding package body.
The initialization of an object is considered a usage of the object at the place where it is declared, thus preventing it from being moved. Therefore, constants and initialized variables are never reported as being movable to inner scopes; they are reported as being movable to package bodies however. Entities that are used as prefixes of a 'Access or 'Address attribute are never reported, since moving them would change their accessibility level. Similarly, task objects are not reported since moving them would change their master. Finally, dispatching operations (primitive operations of tagged types) are not reported either, since they can be the target of an “invisible” (dispatching) call.
If no <subrule> is given, or the <subrule> is “all”, all declarations
are controlled. If no_blocks is specified in front of a
<subrule>, the rule will not consider blocks as possible targets for a
reduced scope for the corresponding category. If to_body is
specified in front of a <subrule>, the rule will report only elements
declared in a package specification that could be moved into the body.
Specifying “all” explicitely is only useful in the case where there
is a <restriction>.
As a side effect, the rule will report about entities that are declared but not used (i.e. whose scope reduces to nothing).
Ex:
-- Types and variables shall be declared in the innermost scope
-- where they are useful:
check reduceable_scope (variable, type);
-- Packages and subprograms shall be declared in the innermost
-- scope where they are useful, but they are not allowed in blocks:
check reduceable_scope (no_blocks subprogram, no_blocks package);
-- Use clause should be as restricted as possible:
search reduceable_scope (use);
If you think that use clauses are acceptable, but should be limited to the smallest possible scope, you would generally specify:
check unnecessary_use_clause;
check reduceable_scope (use);
Currently, the rule does not report use clauses declared in a package specification that could be moved to the body. Such clauses appear as “unused” (but of course, the compiler will complain on the body if the clause is removed).
This rule controls usage of representation clause.
<control_kind> representation_clauses [(<subrule> {, <subrule>})];
<subrule> ::= {<category>} <repr_kw> | [object] <attribute>
<repr_kw> ::=
at | at_mod | enumeration |
fractional_size | incomplete_layout | non_aligned_component |
non_contiguous_layout | overlay | layout
<category> ::=
() | range | mod | delta | digits | array | record |
tagged | extension | access | new | private | task | protected
Without parameter, the rule controls all representation clauses, otherwise it will control the representation clauses given as parameter.
If a representation keyword or attribute is preceded by one or several
categories, the rule controls only the representation items that apply
to types belonging to the categories (the type of the component for
the non_aligned_component subrule):
The meaning of the representation keywords is:
System.Storage_Unit.
'Address of some other element.
In addition to these keyword, any specifiable attribute can be given
(including the initial “'”); the rule will control specifications of
this attribute. If the modifier “object” is given before the
attribute, only attribute specifications for objects are controlled
(as opposed to types for example). Note that double attributes (like
“'CLASS'INPUT”) can be given, and are considered different from the
simple attribute (“'INPUT”). It is of course possible to specify
both.
Ex:
All_Addresses: check representation_clauses (at, 'address);
All_Input: check representation_clauses ('input, 'class'input);
Sized_Objects: check representation_clauses (object 'size);
count representation_clauses ('SIZE);
-- check layout clauses for derived types:
check representation_clauses (new layout);
-- check layout clauses for root tagged types and type extensions:
check representation_clauses (tagged extension layout);
For the “fractional_size” and “non_contiguous_layout” subrules, there are some rare cases where AdaControl cannot evaluate the given size or elements of the record representation clause, and thus not detect the corresponding situation. Such cases are detected by the rule “uncheckable”. See Uncheckable.
The specifiable attributes (the ones that can be given as parameters
to this rule) are 'Address, 'Size,
'Component_Size, 'Alignment, 'External_Tag,
'Small, 'Bit_Order, 'Storage_Pool,
'Storage_Size, 'Write, 'Output, 'Read,
'Input, and 'Machine_Radix. See Ada Reference Manual
13.3(77).
Ada allows partial record representation clauses, i.e. it does not require all fields to be specified. This means that if you add a field to a record and forget to update the associated representation clause, there will be no compilation error. The “incomplete_record” subrule is handy for making sure that this does not happen.
Derived types with a representation clause may incur efficiently penalty, since calling an inherited subrograms requires a change of representation. Representation clauses for tagged types are dubious, since these types have hidden fields added by the compiler.
This rule controls that certain form of types are not used for function results.
<control_kind> return_type [(<subrule> {, <subrule>})];
<subrule> ::= class_wide | constrained_array |
protected | task |
unconstrained_array | unconstrained_discriminated
This rule controls functions whose return type belongs to one of the indicated type kinds:
class_wide controls class-wide types
constrained_array controls constrained array types
unconstrained_discriminated controls types with discriminants
(but not constrained subtypes of such types)
unconstrained_array controls unconstrained array types
task controls task types, or composite types that include tasks
as subcomponents.
protected controls protected types, or composite types that include protected
objects as subcomponents.
If no subrule is specified, all type kinds are controlled. Note that more than one kind may apply to a type: for example, a function can return a class-wide type with discriminants that includes tasks and protected objects as subcomponents. In this case, several messages are issued for the same type.
Ex:
check return_type (unconstrained_discriminated, unconstrained_array);
There is a (very rare) case where AdaControl does not properly recognize that a function returns a class-wide type. This is due to an ASIS bug fixed in version 5.05, and therefore appears only with earlier versions of the compiler. This happens when a generic unit contains functions whose return type is a generic indefinite formal type, and this generic is instantiated with a class-wide type.
This rule controls calls that may depend on the order of evaluation of parameters.
<control_kind> side_effect_parameters (<entity> {, <entity>});
This rule controls subprogram calls or generic instantiations where different actual parameters call functions known to have side effects. This is dangerous practice, since correct behaviour may depend on a certain evaluation order of parameters, which is not specified by the language.
All <entity> are functions that are assumed to interfere, i.e. the rule will signal if any of these functions is called more than once in the parameters of a call. As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
It is allowed to give the name of a generic function, or of a function declared in a generic package; in this case, all functions resulting from instantiations of these generics will be considered.
In the case of renamings, you must give the name of the original function; the rule will work correctly if the call is made through a renaming of this function.
Ex:
check side_effect_parameters (F1);
check side_effect_parameters (G1, G2);
Here, F1 has a side effect, and the rule will signal if it is called more than once. G1 and G2 are assumed to interfere, and therefore the rule will signal if either is called more than once, or if both are called. However, having a call that mentions F1 and G2 is OK.
Due to the size of internal structures, this rule may not be given more than 100 times.
Due to an unimplemented feature of ASIS-for-Gnat, this rule will not process defaulted parameters, and hence not detect interferences due to calling a side-effect function through the default value.
This rule controls exception handlers that can cause exceptions to silently disappear.
<control_kind> silent_exceptions (<element> {, <element>});
element ::= <control-item> | <report-item>
control-item ::= not | with <entity> | others
report-item ::= raise | explicit_raise | reraise | return |
requeue | <entity>
The rule controls handlers that do not call one of the given subprograms (for example a reporting procedure) nor perform other required operations, like returning, requeuing, or re-raising an exception.
A parameter that starts with “not” or “with” is a <control-item> and defines wich exceptions are controlled; the <entity> should be either an exception, or the name of a library unit (in which case, it applies to all exceptions declared in the library unit). As usual, the whole syntax for entities is allowed here. See Specifying an Ada entity name. If the <entity> is (part of) a generic, then it applies to all exceptions from all corresponding instantiations. If there is no <control-item>, then all exceptions are controlled.
If several <control-item> are given, the ones with “with” add exceptions to the set of controlled exceptions, and the ones with “not” remove exceptions, in order, starting from the empty set if the first <control-item> is a “with”, or starting from the set of all exceptions if the first <control-item> is a “not”. See examples below.
“when others” handlers are always controlled, unless there is an explicit “not others” <control-item>. A “with others” <control-item> can be specified to check only “when others” handlers.
The other parameters are <report-item> and define the constructs considered “reporting”. <entity> should correspond to an Ada callable entity or generic package; as usual, the whole syntax for entities is allowed here. See Specifying an Ada entity name. If a generic procedure or function is given, then all corresponding instances are considered reporting subprograms. If a generic package is given, any instantiation (in an inner block of the handler) is considered reporting. In addition, the special names “explicit_raise”, “reraise”, “return” and “requeue” mark raise statements with an explicit exception name, raise statements without an exception name, return statements, and requeue statements (respectively) as reporting. “raise” is a shorthand for both “explicit_raise” and “reraise”.
If “explicit_raise” is given as a parameter, the procedure
Ada.Exceptions.Raise_Exception is automatically added to the
list of procedures for both Check and Search, unless it is
explicitely specified as a parameter in a rule; and similarly
Ada.Exceptions.Reraise_Occurrence is added for “reraise”.
This way, it is possible to consider them as reporting procedures for
Check (for example) and not for Search.
A handler where all exceptions are uncontrolled is not controlled at all (i.e. it is allowed to be non reporting). Otherwise, the rule reports if the handler does not contain at least one of the <report-item> in each possible path of the handler. If the <report-item> appear only in if or case statements, but not in all possible paths, or if they appear only in the body of loop statements, the rule will issue a message asking for a manual verification, since it cannot be statically determined whether the proper treatment happens in every case.
Note that the purpose of this rule is to require the reporting calls to be “eye-visible”, i.e. textually written in the exception handler. For example, the rule will accept a call to a procedure inside the sequence of statements of a package body declared in some inner block; however, it will not accept the same call if it is in the sequence of statements of a package instantiation (unless the generic package is itself mentionned as reporting), because the call is not “eye-visible”. For the same reason, a call to a reporting function which happens as the default value of an omitted parameter in some other call will not be accepted.
This rule can be given once for each of check, search and count. This way, it is possible to have a level considered a warning (search), and one considered an error (check).
Ex:
-- Make an error if exception is not reraised and does not call
-- Reports.Trace, but make it only a warning if the exception is an
-- IO exception or Constraint_Error:
check silent_exceptions (not ada.io_exceptions,
not standard.constraint_error,
raise,
reports.trace);
search silent_exceptions (raise, reports.trace);
-- check handlers that do not reraise the exception, except for
-- IO exceptions:
check silent_exceptions (not Ada.IO_Exceptions, reraise);
-- Same for predefined exceptions, except Constraint_Error:
check silent_exceptions (not Standard, with Standard.Constraint_Error,
reraise);
-- Same for all exceptions named User_Error, wherever they are declared,
-- and no others
check silent_exceptions (with all User_Error, reraise);
-- Same for "when others" handlers
check silent_exceptions (with others, reraise);
Currently, “return” includes all return statements. It would be nice to separate function returns from procedure or accept returns. This is expected to be done in some future version of AdaControl.
There are two cases that are not statically checkable, and thus may not be identified by this rule: if an exception is raised in an inner block statement and handled locally, and if the exception handler aborts the current task.
If a reporting function is given, there are a few cases where the calls will not be recognized:
This rule controls expressions that can be simplified in various ways.
<control_kind> simplifiable_expressions [(<subrule> {, <subrule>})];
<subrule> ::= conversion | logical | logical_false | logical_not |
logical_true | parentheses | range
Without parameters, all kinds of simplifiable expressions are controlled; otherwise, the controlled expressions depend on the subrule:
<expr> = True (or /=), and “logical_false” does the
same for comparisons with false.
T'First .. T'Last
that should be T'range (or even simply T).
This rule can be given at most once for each subrule.
Ex:
search simplifiable_expressions (parentheses);
check simplifiable_expressions (range, logical);
There are cases where parentheses may seem unnecessary, but are (purposedly) not reported by this rule. Consider for example:
X := A + (B + C);
Removing the parentheses would change the expression to mean:
X := (A + B) + C;
If the "+" operator has be redefined and is no more
associative, this would actually change the meaning of the program. In
a less contrieved example, note that:
X mod (A*B)
is not the same as:
X mod A * B
For these reasons, and to make the rule easier to understand for the user, the rule does not report unnecessary parentheses between operators of identical priority levels.
Conversion of universal value is never necessary, however there are cases where overloading resolution may require the conversion to be replaced by a qualification, rather than being simply removed.
This rule controls statements that can be removed or simplified in various ways without changing the meaning of the program.
<control_kind> simplifiable_statements [(<subrule> {, <subrule>})];
<subrule> ::= block | dead | handler | if | if_for_case |
if_not | loop | loop_for_while | nested_path | null
Without parameter, all kinds of simplifiable statements are controlled; otherwise, the controlled statements depend on the subrule:
block controls block statements that have no labels, no
declarations, and no exception handlers.
dead controls dead code, i.e. statements that are statically
known to be never executed. This includes statements that follow a
return, requeue, or goto statement,
or an exit statement that is either unconditional or whose
condition is statically known to be true. It includes also
while statements and if statements (including
elsif paths) whose condition is statically false, and
for loops whose range is statically empty.
handler controls “trivial” exception handlers, i.e. handlers
whose sequence of statements includes only a single raise
statement without an exception name. However, a handler is not
reported if there is also a non trivial handler for others.
These examples show the situation:
exception
when Constraint_Error => -- Reported (no when others)
raise;
end;
exception
when Constraint_Error => -- Reported (trivial when others)
raise;
when others => -- Reported
raise;
end;
exception
when Constraint_Error => -- Not reported (non trivial when others)
raise;
when others =>
Put_Line ("Error");
end;
if controls if statements with an else
path that contains only null statements (and can thus be
removed).
if_for_case controls usage of if statements that
could be replaced by case statements. An if
statement is assumed to be replaceable if it has at least one
elsif and all conditions are comparisons (or membership
tests, possibly connected by logical operators) of the same discrete
variable with static values. Typically, this subrule will spot
constructs like:
if X = 1 then
...
elsif X = 2 or X = 3 or X = 4 then
...
elsif X >= 5 and X <= 10 then
...
elsif X in 11 .. 20 then
...
else
...
end if;
if_not controls if statements with an
else path and no elsif path, and where the
condition is given in negative form (i.e. it is a not, or a
"/=" comparison). Such statements could be made positive (and
thus less error-prone) by interverting the if and
else paths.
nested_path controls paths from if statements that
can be moved outside. This happens if the if has only
then and else paths, and either of them ends
with a “breaking” statement (raise, return,
exit or goto); in this case, the other path
needs not be nested inside the if statement. However, if
both paths end with the same “breaking” statement, no error is
reported. In short, the rule signals the following examples:
if Cond then
return;
else
I := 1;
end if;
if Cond then
I := 1;
else
return;
end if;
because they can be changed to:
if Cond then
return;
end if;
I := 1;
if not Cond then
return;
end if;
I := 1;
The rule will not signal the following example, where both paths end with the same “breaking” statement (return), because it would break the symetry of the statement:
if Cond then
return 1;
else
return 2;
end if;
null controls null statements that serve no purpose
and can be removed. Note that if a null statement carries
a label, it is not considered simplifiable.
loop controls while loop statements where the condition
is a plain True, and can thus be changed to simple loops.
loop_for_while controls simple loop statements whose first
statement is an exit (for the same loop), and which can
therefore be turned into a while loop.
This rule can be given at most once for each subrule.
Ex:
check simplifiable_statements (block, null);
search simplifiable_statements (if);
loop may seem a strange thing to check, since no Ada
programmer is supposed to write this. However, experience shows that
it is a good indicator of code written by people who did not get
proper Ada training. Such code is certainly worth a peer review...
This rule controls usage of certain Ada statements.
<control_kind> statements (<statement_kw> {, <statement_kw>};
<statement_kw> ::=
any_statement |
abort | accept_return |
assignment | asynchronous_select |
block | case |
case_others | case_others_null |
code | conditional_entry_call |
declare_block | delay |
delay_until | dispatching_call |
effective_declare_block | entry_call |
entry_return | exception_others |
exception_others_null | exit |
exit_expanded_name | exit_for_loop |
exit_outer_loop | exit_while_loop |
for_loop | function_return |
goto | if |
if_elsif | inherited_procedure_call |
labelled | loop_return |
multiple_exits | no_else |
null | procedure_return |
raise | raise_locally_handled |
raise_nonpublic | raise_standard |
requeue | reraise |
selective_accept | simple_loop |
terminate | timed_entry_call |
unconditional_exit | unnamed_block |
unnamed_exit | unnamed_for_loop |
unnamed_loop_exited | unnamed_multiple_loop |
unnamed_simple_loop | unnamed_while_loop |
untyped_for | while_loop
Statement keywords that are Ada keywords control the corresponding Ada statements. The meaning of other keywords is as follows:
any_statement controls all statements. This is of course not
intended to forbid all statements in a program (!), but
counting all statements can be quite useful.
accept_return controls return statements that return from an
accept statement, entry_return controls return
statements that return from a (protected) entry body, and
procedure_return controls return statements that return from a
procedure. loop_return controls return statements that appear
inside a loop statement.
assignment controls all assignment statements.
asynchronous_select controls the select
... then abort statement. conditional_entry_call
controls the select ... else
statement. timed_entry_call controls the select
... or delay statement. selective_accept controls
the regular select statement.
block controls all block statements, while unnamed_block
controls blocks without a name, declare_block controls blocks
with an explicit declare (even if the declarative part is
empty), and effective_declare_block controls blocks with a
declarative part that includes anything else than use
clauses and pragmas.
case controls all case statements.
case_others controls any when others path in a
case statement, while case_others_null controls only
when others paths in a case statement that
contain only null statements.
code controls code statements.
delay controls only relative delay statements, while
delay_until controls absolute delay until
statements.
dispatching_call controls all dispatching calls. Note that this
subrule controls dispatching procedure calls as well as dispatching
function calls, although the latter is technically an expression and
not a statement.
entry_call controls all entry call statements, including those
that are part of a conditional or timed entry call statement.
exit controls all exit statements, while exit_for_loop
and exit_while_loop control exit statements that
terminate for and while loops,
respectively. unconditional_exit controls exit
statements without a when condition. multiple_exits
controls loop that have more than one exit statement.
unnamed_loop_exited controls exit statements that terminate an
unnamed loop. exit_outer_loop controls exit statements
that exit from an outer loop (i.e. not the innermost one). exit_expanded_name controls named exit statements
where the name is given as an expanded name.
exception_others controls any when others exception
handler, while exception_others_null controls only
when others exception handlers that contain only
null statements.
if_elsif controls if statements that have at least
one elsif.
for_loop controls all for loops.
function_return controls return statements from
functions. Obviously, return statements cannot be forbidden in
functions; this keyword controls that there is only one return
statement in the body of functions, and at most one return statement
in each exception handler of the exception part of functions.
if controls all if statements.
inherited_procedure_call controls calls to procedures that have
been inherited by a derived type and not redefined.
labelled controls statements with a label (true statement
labels, not block and loop names).
no_else controls if statements that have no else
path.
null controls all null statements.
raise controls all raise statements.
reraise controls raise statements in exception
handlers that reraise the same exception, and calls to the
Ada.Exceptions.Reraise_Occurrence procedure.
raise_standard controls raise statements that raise one
of the predefined exceptions (those declared in package
Standard). raise_nonpublic controls statements that
raise exceptions that are neither predefined nor defined in the
visible part of a package which is the enclosing library unit of the
statement. raise_locally_handled controls statements that
raise an exception which is handled by a handler in the same
subprogram body as the statement.
Note that for these subrules, the exception can be raised either by a
raise statement, or by a call to
Ada.Exceptions.Raise_Exception where the raised exception is
statically determinable.
simple_loop controls simple loops, i.e. those that are neither
while nor for loops.
unnamed_exit controls exit statements without a
loop name that exits from a named loop.
unnamed_for_loop, unnamed_simple_loop, and
unnamed_while_loop control loops of the given kind that are not
named.
unnamed_multiple_loop controls nested loops that are not named
(i.e. under this rule, only loops that contain no inner loop, and are
not nested in another loop, are allowed not to be named). The kind of
loop (plain, for, while) is not considered.
untyped_for controls for loops that use a range
without an explicitely named type (i.e. for I in 1..10
loop). Using a 'Range attribute is OK.
while_loop controls all while loops.
Ex:
search statements (delay);
check statements (goto, abort);
check statements (case_others_null, exception_others_null);
It may seem strange to control things like if or case statements, since no coding standard would prohibit their use. However, this may be useful, especially with “count”, for statistical purposes, like measuring the ratio of if to case statements.
The plain “raise” subrule controls the raise statement, and only this one. If you want to check all places where exceptions can be raised, use also the “entities” rule like this:
"all raise": check statements (raise);
"all raise": check entities (Ada.Exceptions.Raise_Exception,
Ada.Exceptions.Reraise_Occurrence);
Other subrules of the “raise” family are more about which kind of
exception is being raised, and therefore control also exceptions
raised by calling the procedures from Ada.Exceptions.
“inherited_procedure_call” controls only procedure calls. For function calls, see rule Expressions.
This rules controls usage of various “general” Ada coding style.
<control_kind> style;
<control_kind> style (casing_attribute, <casing_kw>);
<control_kind> style (casing_identifier, <casing_kw>);
<control_kind> style (casing_keyword, <casing_kw>);
<control_kind> style (casing_pragma, <casing_kw>);
<control_kind> style (compound_statement);
<control_kind> style (default_in);
<control_kind> style (exposed_literal, <type_kw>, {, <value_place>});
<control_kind> style (formal_parameter_order {, <mode list>});
<control_kind> style (multiple_elements {,<element_kw>});
<control_kind> style (negative_condition);
<control_kind> style (no_closing_name [, <max_lines>]);
<control_kind> style (numeric_literal, [not] <base> [, <block_size>]);
<control_kind> style (parameter_order {, <mode list>});
<control_kind> style (positional_association
{,<context_kw> [,<max_allowed>]{,<entity>}}
| [, <max_allowed>]);
<control_kind> style (renamed_entity);
<casing_kw> ::= uppercase | lowercase | titlecase | original
<context_kw> ::= [not_operator] call | pragma | discriminant |
instantiation | array_aggregate | record_aggregate
<element_kw> ::= [flexible] clause | declaration | statement
<mode_list> ::= <mode> {<mode>}
<mode> ::= in | defaulted_in | access | in_out | out |
type | procedure | function | package
<type_kw> ::= integer | real | character | string
<value_place> ::= <value> | <place>
<value> ::= <integer number> | <real number> | "<pattern>"
<place> ::= constant | exponent | index | number | pragma |
repr_clause | var_init | type
The first parameter specifies which style aspect is to be checked:
in mode for a parameter. Access
parameters are not reported, since an an explicit in is not
allowed in that case.
"**") function call, “index” for array
indexing, “number” for named number declarations, “pragma” for
pragma arguments, “repr_clause” for representation clauses, “type”
for type (and subtype) declarations, and “var_init” for the
initialization expression of variable declarations. If no <place> is
given, it is taken as number, constant, i.e. any literal is
allowed in named numbers and constant declarations.
If no parameter is given, the order for regular parameters is “in” or “access” first, then “in_out”, then “out”, then “defaulted_in”. The order for formal_parameters is “type” first, then “in” “defaulted_in” and “access”, then “in_out”, then “procedure” and “function”, then “package”.
For calls, positional association is not reported for operators
that use infix notation (since named notation is not possible); in
addition, if the “not_operator” modifier is specified before the
“call” keyword (not allowed elsewhere), positional association is
never reported for operators, even if they are called with the syntax
of a normal function call (i.e. Pack."+" (A,B)). Calls to
subprograms that are attributes are not reported either, since named
notation is not allowed for them.
Ex:
search style (no_closing_name);
search style (no_closing_name, 5);
check style (casing_identifier, original);
check style (default_in);
check style (numeric_literal, 10, 3);
check style (exposed_literal, integer, 0, 1);
check style (exposed_literal, real, 0.0, 1.0);
-- in parameters (with or without default) and access
-- parameters must be first, then in out parameters, then
-- out parameters. In parameters are allowed last if they
-- have defaults.
check style (parameter_order,
in defaulted_in access,
in_out,
out
defaulted_in);
-- For generics, formal objects must come first, then formal
-- types, then formal subprograms, then formal package:
check style (formal_parameter_order,
in in_out,
type,
procedure function,
package);
-- All positional associations:
check style (positional_association);
-- All positional associations in calls and aggregates:
check style (positional_association, array_aggregate,
record_aggregate, call);
-- All positional associations with more than 3 elements:
search style (positional_association, 3);
-- Positional associations in calls with more than 3 elements,
-- and positional associations in aggregates with more than 4 elements:
search style (positional_association, call, 3,
array_aggregate, 4,
record_aggregate, 4);
-- Positional associations in calls with more than 2 elements (except calls
-- to any subprogram called Put), and in instantiations (except for
-- instantiations of Text_IO subpackages):
search style (positional_association,
call, 2, all put,
instantiation, Ada.Text_IO.Integer_IO,
Ada.Text_IO.Fixed_IO,
Ada.Text_IO.Decimal_IO,
Ada.Text_IO.Float_IO,
Ada.Text_IO.Enumeration_IO);
Without parameter, the rule will control all style aspects with parameter values that correspond to the most commonly used cases, i.e. it is equivalent to the following:
style (no_closing_name);
style (casing_attribute, titlecase);
style (casing_keyword, lowercase);
style (casing_identifier, original);
style (casing_pragma, titlecase);
style (positional_association);
style (default_in);
style (negative_condition)
style (multiple_elements)
style (literal, 10, 3);
style (exposed_literal, integer, 0, 1)
style (exposed_literal, real, 0.0, 1.0);
Note that operators always follow the casing rule for keywords, even
for calls that use the infix notation (i.e. in "and"(A, B)).
There are two kinds of calls where the rule does not complain about usage of positional association: infix operator calls (since requiring named notation would not allow infix notation any more), and calls to subprograms that are attributes (since named notation is not allowed for these).
For calls, another rule controls positional associations according to the value of parameters rather than their number: See Insufficient_Parameters.
In many cases, badly laid-out compound statements will trigger both the “multiple_elements, statement” and the “compound_statement” subrules. For example:
if C then I := 1; end if;
will complain that the assignment is on the same line as the if, and that the if statement spans less than 3 lines. However, the subrules are not equivalent. For example,
if C then I := 1;
end
if;
will only find that the assignment is on the same line as the if, while
if C then
I := 1; end if;
will only find that the if statement spans less than 3 lines. In most cases, you'll want to specify both subrules to ensure proper lay-out.
If a predefined operator or an attribute is renamed, the “renamed_entity” subrule cannot check that the original entity is not used in the scope of the renaming. Such cases are detected by the rule “uncheckable”. See Uncheckable.
This rule controls tasks that can terminate.
<control_kind> terminating_tasks
A task is considered a terminating task if its last statement is not an unconditional loop, or this if this loop is exited. It is also considered terminating if it contains a selective accept with a terminate alternative.
Since this rule has no parameters, it can be given only once.
Ex:
check terminating_tasks
There is still one case where a task terminates, which is not reported by this rule: when a task is aborted. This is intended, since there are cases (like mode changes) where a logically non-terminating task is aborted.
If aborts are also to be reported, use the rule “statements (abort)”. See Statements.
This rule controls that a special constant is declared together with each type, for example to serve as a default initial value.
<control_kind> type_initial_values [("<pattern>")];
This rule controls types that do not feature an initialization constant declared in the same declarative part as the type. If no <pattern> is given, any constant is considered an initialization constant for its type; otherwise, only constants whose name matches the given pattern are considered initialization constants.
Ex:
check type_initial_values ("^C_Init_");
The above example will ensure that every declared type features a constant of the type whose name starts with “C_Init_”.
This rules controls cases where it is not possible to guarantee the accuracy of checks performed by AdaControl, and where manual inspection may be required.
<control_kind> uncheckable [(<subrule> [,<subrule>])];
<subrule> ::= false_positive | false_negative | missing_unit
If the keyword “missing_unit” is given, this rule controls missing units, i.e. units given on the command line that are not found (and therefore not controlled) will result in an usual error message.
Otherwise, this rule controls constructs that are not static and prevent other rules from being fully reliable. This rule is special, since it really affects the way other rules behave when they encounter a statically uncheckable construct. Therefore, if a label is given, the message will include the label as usual, with an indication of the rule that triggered the message; if no label is given, the message will include the name of the rule that detected the uncheckable construct, not “uncheckable” itself.
If the keyword “false_negative” is given, the rule will control constructs that could result in false negatives, i.e. possible violations that would go undected, while if the keyword “false_positive” is given, it will control constructs that could result in false positives, i.e. error messages when the rule is not really violated. If no keyword is given, both occurrences are controlled.
As far as statistics are concerned (see Control kinds and report messages), “uncheckable” messages from rules are counted under the corresponding rule's statistics (like other messages), but there will be also a count of all “uncheckable” messages under the rule “UNCHECKABLE”, and also subtotals corresponding to the number of “uncheckables” for each rule.
This rule can be given only once for each of value of the parameters.
Ex:
check uncheckable (false_negative);
search uncheckable (false_positive);
check uncheckable (missing_unit);
This rule is especially important when AdaControl is used in safety critical software, since it will detect constructs that could escape verification. Such constructs should be either disallowed, or require manual inspection. On the other hand, in casual software, it may lead to many messages, since for example dispatching calls are uncheckable with many rules.
With “missing_unit”, the message does not include a reference to a source location, since there is no place in the source which can be considered as the origin of the error. If you run AdaControl from GPS, there will always be a separate category (“Uncheckable”) in the locations window, under which the message will appear, with a file name of “none”. Don't try to click on the error message, since GPS will find no file named “none”!
This rule controls that all necessary units, and only those, are processed by AdaControl.
<control_kind> units [(<subrule> [,<subrule>])];
<subrule> ::= unreferenced | unchecked
If the keyword unreferenced is given, the rule controls
compilation units that are part of the set of analyzed units, but
withed by no other unit. If the keyword unchecked is given, the
rule controls compilation units that are withed by other unit(s), but
not part of the set of controlled units (except standard units).
This rule can only be given once for each of the subrules.
Ex:
check units (unchecked);
search units (unreferenced);
The main program will appear as unreferenced, since it is normally part of the controlled units, and not withed by any other unit. As usual, the corresponding message can be disabled by putting the comment:
--## rule line off units
on the main program.
This rule controls use clauses that do not serve any purpose.
<control_kind> unnecessary_use_clause [(<subrule> {,<subrule>})];
<subrule> ::= unused | qualified | operator | nested | movable
The rule controls use clauses that can safely be removed, moved, or changed to a use type clause. This happens in the following cases:
In this case, the use clause can safely be removed.
In this case, the use clause can safely be removed, but you may want to keep it for documentation purposes, since the package is actually used within this scope.
In this case, and except for some pathological cases (definition of operators that are not primitive operations of the corresponding type), the use clause can be replaced by one or several use type clause(s).
If you also have a message that the outer use clause is unnecessary, this means that all references to the package appear inside the inner use clauses, and that the outer one can be removed. If not, you can either remove the inner use clauses, or remove the outer one and add more local use clauses where necessary.
In this case, the use clause can safely be moved to the body, unless it appears in a library package, and there are unqualified references to its elements from child units.
If no parameter is given, all cases are controlled, otherwise only cases corresponding to the specified keyword(s) are controlled. This rule can be given only once for each value of the parameters.
Ex:
remove: search unnecessary_use_clause (unused);
use_type: check unnecessary_use_clause (operator);
This rule checks only usage of use clauses. The rule “reduceable_scope” can be used to check that use clauses do not span unnecessarily to wide a scope. See Reduceable_Scope.
There are some rare cases where the rule may signal that a use clause is not necessary, where it actually is. There is no risk associated to this since if you remove the use clause, the program will not compile.
The first one comes from a limitation of the ASIS standard: if the only use of the use clause is for making the “root” definition of a dispatching call visible.
The second one comes from a limitation in ASIS-for-Gnat. This happens when the only use of the use clause is for making an implicitely declared operation (an operation which is declared by the compiler as part of a type derivation) visible, and when:
Since these problems come from intrinsic limitations of ASIS, there is nothing we can do about it. When this happens, you can disable the unnecessary_use_clause rule using the line (or block) disabling feature. See Disabling controls. Note that for the third alternative of the second case, you can also qualify one of the parameters, so it is not universal any more.
This rule controls usage of calls to operations that are normally paired (like P/V operations) and do not follow a "safe" coding pattern.
<control_kind> unsafe_paired_calls
(<opening procedure>, <closing procedure> [, <lock type>]);
<opening procedure> ::= <entity>
<closing procedure> ::= <entity>
<lock type> ::= <entity>
The following explanations are given in terms of “locks” since this is the primary use of this rule, however the rule can be used for any calls that need to be properly paired.
The rule can deal with three different kinds of locks:
As usual, the whole syntax for entities is allowed for <entity>. See Specifying an Ada entity name.
The "safe" coding pattern is defined as follows:
Typically, the “safe” pattern corresponds to the following structures:
-- Abstract state machine
begin
P;
-- Do something
V;
exception
when others =>
V;
-- handle exception
end;
-- Object abstract data type
declare
My_Lock : Lock_Type;
begin
P (My_Lock);
-- Do something
V (My_Lock);
exception
when others =>
V (My_Lock);
-- handle exception
end;
-- Reference abstract data type
declare
Lock_Ptr : constant Lock_Access := Get_Lock;
begin
P (Lock_Ptr);
-- Do something
V (Lock_Ptr);
exception
when others =>
V (Lock_Ptr);
-- handle exception
end;
Ex:
check unsafe_paired_calls (Semaphore.P, Semaphore.V, Semaphore.Lock_Access);
If the <Lock type> parameter is provided, both procedures must have a single parameter of the given type, it must not correspond to an “out” parameter, and if it corresponds to an “in” parameter, the type must be discrete or access.
This rule can be specified several times, and it is possible to have
the same procedure belonging to several rules. For example, if you
have a Mask_Interrupt procedure that should be matched by
either Unmask_Interrupt or General_Reset (all declared
in package IT_Driver), you can specify:
check unsafe_paired_calls (IT_Driver.Mask_Interrupt,
IT_Driver.Unmask_Interrupt);
check unsafe_paired_calls (IT_Driver.Mask_Interrupt,
IT_Driver.General_Reset);
Normally, the legality of a rule is checked when the rules file is parsed, and execution does not start if there is any error. However, the legality of the provided type can be checked only during the analysis. If the type is incorrect for some reason, a proper error message is issued and execution stops immediately.
Due to a weakness of the ASIS standard, dispatching calls are not considered. Especially, this means that the <Lock type> cannot be class-wide. Such calls are detected by the rule “uncheckable”. See Uncheckable.
Due to a size limitation of internal data structures, this rule can be specified at most 32 times.
This rule controls unchecked conversions between types which are not statically known to have identical sizes.
<control_kind> unsafe_unchecked_conversion
This rule controls instances of Unchecked_Conversion between
types where the following conditions are not met:
Moreover, a special message is given if any of the types is a class-wide type (certainly a very questionable construct!).
Ex:
check unsafe_unchecked_conversion
There are cases where a size clause is given for a type, but AdaControl is unable to evaluate it. This happens especially if the size clause refers to a size attribute of a predefined type, like:
for T'Size use Integer'size;
This can lead to false positives (i.e. detection of instantiations of
Unchecked_Conversion that are actually OK. Such cases are
detected by the rule “uncheckable”. See Uncheckable.
This rule controls how certain entitities (variables, constants, types, procedures, functions, exceptions, tasks, protected objects, and generics) are used.
<control_kind> usage
(variable|object {,[not] <location> | read | written | initialized});
<control_kind> usage
(constant {,[not] <location> | read});
<control_kind> usage
(type {,[not] <location> | used});
<control_kind> usage
(procedure {,[not] <location> | called});
<control_kind> usage
(function {,[not] <location> | called});
<control_kind> usage
(exception {,[not] <location> | raised | handled});
<control_kind> usage
(task {,[not] <location> | called | aborted});
<control_kind> usage
(protected {,[not] <location> | called});
<control_kind> usage
(generic {,[not] <location> | instantiated});
<control_kind> usage
(all {,[not] <location>});
<location> ::= from_visible | from_private | from_spec
The first parameter defines the class of entities to be controlled. “object” stands for both “constant” and “variable”, “type” stands for both types and subtypes, and “all” stands for all classes.
If only one parameter is given, usage of all entities belonging to the indicated class are reported . Otherwise, other parameter(s) are keyword that restrict the kind of usage being controlled.
“[not] from_visible”, “[not] from_private”, and “[not] from_spec” restrict entities being checked to those that appear (or not) in (generic) package specifications, in the visible part, in the private part, or in any part, respectively. Other keywords carry their obvious meaning, and are allowed only where appropriate. The rule will output the information only for objects that match all the conditions given. A combination of parameters can be given only once for each of “check”, “search”, and “count”.
The report includes the kind of unit that declares the entity (normal unit, instantiation, or generic unit), the part where it is declared (visible or private) if it is declared in a (generic) package, and whether the entity is known to be initialized, read, written, raised, handled, called, or aborted, depending on the entity's class. Some combinations give an extra useful message (for example, a variable which is initialized and read but not written will produce a “could be declared constant” message).
Variables of an access type and variables of an array type whose
components are of an access type (or arrays of an access type, etc.)
are always considered initialized, since they are initialized to
null by the compiler.
Variables that cannot be assigned to (i.e. variables of an array type with some null dimension, or variables of a discrete type whose range includes no values) are specially recognized as “pseudo-constants”: there is no message that they are not written to (since it is not possible), but there is an indication that they are pseudo-constants.
The subrules “procedure” and “function” check only regular subprograms, not protected ones. On the other hand, the subrule “protected” controls all calls to any protected subprogram or entry.
Exceptions raised by calling Raise_Exception and tasks aborted by
calling Abort_Task are properly recognized as exceptions being
raised and tasks being aborted, respectively.
In the case of entities declared in generic packages, the rule will report on usage of the entities for each instantiation, as well as on global usage for the generic itself. Usage for an instantiation will include usage in the generic itself (i.e. if the generic writes to a variable, the variable will be marked as “written” for each instantiation). Usage for the generic itself is the union of all usages in all instantiations (i.e., if a variable from any instantiation is written to, the variable from the generic will be marked as written). Therefore, if the rule reports that a variable in a generic package can be declared constant, it means that no instance of this variable from any instantiation is being written to. But bear in mind that this can be trusted only if all units from the program are analyzed. See limitation.
Note that usage of entities whose declaration is not processed (like,
typically, elements declared in standard packages like
Ada.Text_IO), is not reported. For the same reason, it is not
possible to control usage of predefined operators (since they have no
declaration).
Ex:
-- No variable in package spec; check usage otherwise
Package_Variable: check usage (variable, from_spec);
Constantable : search usage (variable, not from_spec, read,
initialized, not written);
Uninitialized : check usage (variable, not from_spec, read,
not initialized, not written);
Removable : search usage (object, not from_spec, not read);
-- Check exceptions that are never raised
-- generics that are never instantiated
-- and protected objects that are never called
check usage (exception, not raised);
check usage (generic, not instantiated);
check usage (protected, not called);
-- Find how many tasks are declared, and report those
-- that may be aborted
count usage (task);
check usage (task, aborted);
Constants that are never used, exceptions that are never raised or handled, tasks that are never called, etc. are suspicious. Moreover, some useful compiler warnings (like those about variables that should be declared constants) are not output for variables declared in library packages, and even in some other contexts (at least with GNAT). This rule can check these kind of things, project wide.
Some of these checks make sense only for entities declared in package specifications; for example, variables are often discouraged in package specifications, or need at least some extra control. That's why it can be useful to restrict some checks to package specifications.
Note that an unspecified parameter in a rule stands for two rules (positive and negative form of the missing parameter). I.e.:
search usage (variable, from_spec, read, written);
is the same as:
search usage (variable, from_spec, read, written, initialized);
search usage (variable, from_spec, read, written, not initialized);
Therefore, the following example will complain on the second line that the rule has already been given for this combination of parameters:
search usage (variable, from_spec, read, written);
search usage (variable, from_spec, read, written, not initialized);
Note that the notion of constants for this rule includes named numbers.
The report of this rule is output at the end of the run, and is meaningful only for the units that have been processed; i.e., if it reports “variable not read”, it should be understood as “not read by the units given”. In order to have meaningful results, it is therefore advisable to use this rule on the complete closure of the program.
An exception can be raised by passing its 'Identity to a
procedure that will in turn call Raise_Exception (and similarly
for Abort_Task). These cases are not statically determinable,
and therefore not recognized by AdaControl. However, these cases can
be identified by searching the use of the 'Identity attribute
with the following rule:
check entity (all 'Identity);
If an object is the prefix of a 'Access,
'Unchecked_Access, or 'Address attribute, it can be used
through the access (or address) value in ways that are not statically
analyzable. The same happens if objects are targets of dynamic
renamings. Such cases are detected by the rule
“uncheckable”. See Uncheckable.
Due to a weakness of the ASIS standard, usages of variables used as parameters to dispatching calls are ignored. This limitation will be removed as soon as we find a way to work around this problem, but the issue is quite difficult!
This rule controls usage of use clauses.
<control_kind> use_clauses
[([<subrule>,] <package name> {, <package name>})];
<subrule> ::= local | global
The rule controls every useclause, except those that name one of the mentioned packages. It is therefore possible to allow use clauses just for certain packages. If the keyword “global” is given, only use clauses that appear in context clauses (i.e. together with the with clauses) are controlled; if the keyword “local” is given, only use clauses that appear as declarations are controlled. If no “local” or “global” is given, all use clauses are controlled.
This rule can be given at most once for each of check, search and count. This way, it is possible to have a level considered a warning (search), and one considered an error (check).
Ex:
-- Global use clauses are disallowed, local ones only for IO:
check use_clauses (global);
check use_clauses (local, Ada.Text_IO, Ada.Wide_Text_IO);
This rule controls with clauses that should be removed or moved to a better place.
<control_kind> with_clauses [(<subrule> [, <subrule>])];
<subrule> ::= multiple_names | reduceable | inherited
The parameters are subrule keywords that determine which kind of control is performed:
multiple_names controls any with clause that
mentions more than one unit name.
reduceable reports:
Text_IO and Ada.Text_IO). Note that giving a
with clause in a unit, and repeating it in a child unit (or
subunit) is not considered redundant.
inherited controls child units and subunits that reference a
unit which is not directly withed, i.e. when withed only from a parent
(or enclosing) unit. Although Ada rules imply that a with
clause carries on to child units and subunits, it can be considered
better practice to ensure that every compilation unit withes directly
the units it needs.
Each of the keywords can be given at most once. If no keyword is
given, both reduceable and inherited are assumed.
Ex:
check with_clauses (multiple_names, reduceable);
search with_clauses (inherited);
A with clause can safely be removed if it is unused, and no child unit (or subunit) reports that the unit is inherited.
In most projects, there are programming rules that define the way a program should be written. AdaControl performs controls, i.e. it finds occurrences of certain kinds of constructs. In this chapter, we give examples of commonly found programming rules, and how the corresponding controls can be written.
The file gnatcheck.aru in directory rules gives the
AdaControl equivalents of rules checked by Gnatcheck. This version of
AdaControl covers all Gnatcheck rules. For rules where Gnatcheck
requires a parameter, the AdaControl rule is given for the default
value, or with an example value. Small differences in semantics are
indicated by a comment that starts with "Difference:".
This file is not intended to be used directly, but as an example on how to convert Gnatcheck rules into AdaControl rules. Note that in many cases, AdaControl is much more general than Gnatcheck. The file follows as strictly as possible the rules as defined by Gnatcheck, but if you are migrating from Gnatcheck to AdaControl, you may want to use the more powerful forms provided by AdaControl.
The rules directory provides also rules files that can be
sourced to enforce some commonly encountered general rules.
Identifiers from Standard shall not be redefined
Use file no_standard_entity.aru.
Identifiers from System shall not be redefined
Use file no_system_entity.aru.
Predefined IO packages shall not be used
Use File no_io.aru.
Standard package XXX shall not be used
File no_standard_unit.aru controls usage of all standard
packages. Comment out those that you do want to allow.
Obsolescent features shall not be used
Use file no_obsolescent_features.aru. Not all obsolescent features are
controlled, but most of them (those that are most worth checking) are.
Gnat specific attributes shall not be used
Use file no_gnat_attribute.aru
Features from annex X shall not be used
Use file no_annex_X.aru.
The Ravenscar profile shall be enforced
Use file ravenscar.aru.
Note that not all of the restrictions of the Ravenscar profile are currently controlled, but many are, and we expect later releases of AdaControl to increase the number of controlled features. In some cases (like “Detect_Blocking”), AdaControl does a better job than the profile, since it can detect statically situations that the profile only requires to be detected at run-time. The rule file is also slightly more restrictive than the profile; for example, the restriction “no_task_allocation” only disallows task allocators, while this rule file controls the declaration of access types on tasks.
NASA coding guidelines shall be enforced
Use file nasa.aru. This file is an example of how to convert guidelines (available from http://fsw.gsfc.nasa.gov/gds/code_standards_ada.pdf) into an AdaControl rules file.
Ada 83 unit names shall not be used (i.e. use
Ada.Text_IO, not Text_IO)
Use file no_83_unit_name.aru.
New reserved words of Ada 2005 shall not be used
Use file reserved_2005.aru.
Below are examples of rules that can be directly checked by AdaControl.
Goto statement shall not be used
check statements (goto);
Short circuit forms should be preferred over corresponding logical operators
Use_Short_Circuit: search expressions (and, or);
Aggregates should be used for full assignments to structured variables, unless it is a record with a single component
check multiple_assignments (groupable, given 2, ratio 100);
All loops that contain exit statements must be named, and the name must be given in the exit statement
check statements (unnamed_loop_exited);
check statements (unnamed_exit);
All type names must start with “T_”
check naming_convention (type, "^T_");
All program units must repeat their name after the “end”
check style (no_closing_name);
Pragma Suppress is not allowed
check pragmas (suppress);
Ada tasking must not be used
check declarations (task);
“=” and “/=” shall not be used between real types
check expressions (real_equality);
All tasks must provide an exception handler that calls “Failure” in the case of an unhandled exception
check exception_propagation (task);
check silent_exceptions (failure);
Unchecked_Conversion shall not be used
check entities (ada.unchecked_conversion);
No global variable shall be declared in the visible part of a package specification
check usage (variable, from_spec);
Predefined numeric types of the language shall not be used
check entities (standard.Integer,
standard.short_integer,
standard.long_integer,
standard.Float,
standard.short_float,
standard.long_float);
Access to subprograms shall not be used
check declarations (access_to_sp);
Abort statements shall not be used
check statements (abort);
There shall be only one instantiation of Ada.Numerics.Generic_Elementary_Functions for each floating point type
-- Put a --##RULE LINE OFF GEF
-- for the one which is allowed
GEF: check Instantiations (Ada.Numerics.Generic_Elementary_Functions);
A local item shall not hide an outer one with the same name
check Local_Hiding;
There shall be no IOs in exception handlers
check entity_inside_exception (ada.Text_IO.put, ada.Text_IO.put_line,
ada.Text_IO.get, ada.Text_IO.get_line);
Note that this checks for all overloaded procedures, but only those dealing with characters and strings (those defined directly within Ada.Text_IO). If the names “get” and “put” are not used for anything else than IOs, a more general form can be given as:
check entity_inside_exception (all get, all put,
all get_line, all put_line);
This will check that no entity with the corresponding names appear in exception handlers.
Exceptions shall not be used
No_Exception: check declarations (exception, handlers);
No_Exception: check statements (raise);
No_Exception: check entities (Ada.Exceptions);
This will check that no exception is declared, no exception handler is
provided, and no exception is raised, not even through the services of
the package Ada.Exceptions.
No procedure exported to C shall propagate exceptions
check exception_propagation (interface, C);
There shall be no Unchecked_Conversion to or from Address
check instantiations (ada.unchecked_conversion, system.address);
check instantiations (ada.unchecked_conversion, <>, system.address);
There shall be no use clause except for Text_IO
check use_clauses(ada.text_IO);
Use explicit list of values in case statements rather than “when others”if the “when others” would cover less than 10 values
check Case_Statement(min_others_span, 10);
If a block is more than 20 lines long, it must be named
check Max_Size(unnamed_block, 20);
Exceptions shall not be handled except by main program
check declaration (handlers)
This check will be disabled for the exception handler of the main program.
Each unit has a header starting with a fixed format, and must contain at least 10 lines of comments
check header_comments (model, "header.txt");
check header_comments (minimum, 10);
The file header.txt contains the required header (as regexps), like:
^--*{50}$
^-- This is a header$
Below are examples of rules that require manual inspection, but where AdaControl can be used to identify suspicious areas.
All usages of the 'ADDRESS attribute shall be justified and documented
search entities (all 'address);
Specifying an address for a variable shall be restricted to hardware interfacing
search representation_clauses(address);
There shall be no memory leakage
search Allocators;
This rule identifies all allocations, and thus can be used to check that all allocated elements are properly deallocated.
Many rules can take Ada entities as parameters. Each time a rule uses the category <entity>, it refers to an Ada entity that can be specified with the following syntax:
<entity> ::= <full_name> | "all" <simple_name> | "all" <attribute>
<full_name> is the full name of the Ada entity, using normal
Ada dot notation (with some extensions, see below). Full name means
that you give the full expanded name, starting from a compilation
unit. This name must be the actual full name, i.e. it must not include
any renaming (otherwise the name will not be recognized). For example,
the usual Put_Line must be given as
Ada.Text_IO.Put_Line, not as
Text_IO.Put_Line. Predefined elements (Integer,
Constraint_Error) must be given in the form
Standard.Integer or Standard.Constraint_Error, since
they are logically declared in the package Standard.
<simple_name> is a single identifier, possibly followed by
overloading information. No qualification is allowed.
<Attribute> is an attribute name, including the quote. No
overloading information is allowed.
<full_name> designates a single entity or several overloaded
entities declared in the same place (as identified by the prefix),
while all <simple_name> designates all identifiers with the
given name in the program, irrespectively of where they
appear. all <Attribute> designates all occurrences of the given
attribute, irrespectively of what the attribute applies to.
A utility is provided with AdaControl to help you find the full name of an entity. See pfni. If you are using GPS with AdaControl plug-ins, it can be accessed directly from the contextual menu. See Contextual menu.
In Ada, names can be overloaded. This means that you can have several
procedures P in package Pack, if they differ by the
types of the parameters. If you just give the name Pack.P as
the <entity>, the corresponding rule will be applied to all
elements named P from package Pack. If you want to
distinguish between overloaded names, you can specify a profile after
the element's name. A profile has the syntax:
"{" [ ["access"] <type-name>
{ ";" ["access"] <type-name> } ]
["return" <type-name>] "}"
You must specify the type name, even if the <entity>
declaration uses a subtype of the type; this is because Ada uses types
for overloading resolution, not subtypes. Anonymous access parameters
are specified by putting access in front of the type name. An
overloaded name for a procedure without parameters uses just a pair of
empty brackets. If the subprogram is a function, you must provide the
return <type-name> part for the return type of the
function. The types must also be given as a unique name,
i.e. including the full path: if the type is T declared in
package Pack, you must specify it as Pack.T. As a
convenience, the Standard. is optional for predefined types, so
you can write Standard.Integer as Integer. There is no
ambiguity, since a type is always declared within some construct. Note
that omitting Standard works only for types that are part
of the profile used to distinguish between overloaded Ada entities but
that the Ada entity name must always contain Standard if it is a
predefined element.
Overloaded names can be also be used with the all <simple_name>
form of the <entity>. In this case, the rule will be applied to
all names that are subprograms with the given identifier and matching
the given profile, irrespectively of where they appear.
Note that if you use an overloaded name, all overloadable names that are part of the <entity>, including those of the profile, must use the overloaded syntax. For example, given the following program
procedure P is
procedure Q (I : Integer) is
...
end Q;
procedure Q (F : Float) is
...
end Q;
begin
...
end P;
If you want to distinguish between the two procedures Q, you
must specify them as P{}.Q{Integer} and
P{}.Q{Float} (note the P{} which specifies an
overloaded name for a procedure P without parameters).
The names of entities which can not be overloaded (like package,
exception, ...) must not be suffixed by braces
(e.g. Ada.Text_IO.Put_Line{Standard.String}).
Following normal Ada rules, an enumeration literal is considered a parameterless function. If you want to distinguish between overloaded enumeration literals, you can use overloaded names for them. For example, given:
package Pack is
type T1 is (A, B);
type T2 is (B, C);
end Pack;
Ada entities names are:
Pack.B{return Pack.T1}
Pack.B{return Pack.T2}
AdaControl handles operators (i.e. functions like "+")
correctly. Of course, you must specify such operations using normal
Ada syntax: if you define the integer type T in package
Pack, an overloaded name for the addition would be
Pack."+"{Pack.T; Pack.T return Pack.T}.
It is also possible to designate attributes, using the normal notation
(i.e. Standard.Integer'First). If the name of an attribute which
is a function appears in a name that uses the overloaded syntax, it is
not necessary (and actually not allowed) to provide its profile, since
there is no possible ambiguity in that case. For example, given:
procedure P (I : Integer) is
type T is range 1 .. 10;
begin
...
end P;
You can designate the 'Image attribute for type T as
P{Standard.Integer}.T'Image (the profile of the 'Image
function is not given, as would be necessary for a normal function).
There is a special case for elements that are defined (directly or
indirectly) within unnamed loops or block statements. Everything
happens as if the unnamed construct was named _anonymous_. So
if you have the following program:
procedure P is
begin
for I in 1..10 loop
declare
J : Integer;
begin
...
end;
end loop;
end P;
You can refer to I as P._anonymous_.I, and to J
as P._anonymous_._anonymous_.J.
You can designate the name of a record or protected type component (a “field” name), but to identify it uniquely, you must precede its name by the name of the type. This is a small extension to Ada syntax, but it is the simplest and most natural way to deal with this case. For example, given:
procedure P is
type T is
record
Name : Integer;
end record;
...
The Ada entity name is P.T.Name.
Similarly, you can designate the formal of an access to subprogram type by prefixing it by the access type. For example, given:
procedure P is
type T is access procedure (X : Integer);
...
The Ada entity name of the formal is P.T.X.
Due to a limitation of ASIS for Gnat, it is not possible to specify a profile with predefined operators; predefined operators without a profile work normally.
-- This will not recognize "<" on Standard.Integer:
check entities (Standard."<"{Standard.Integer,
Standard.Integer
return Standard.Boolean});
-- This will correctly recognize all predefined "<":
check entities (Standard."<");
The following syntax gives the complete definition of regular
expressions, as used by several rules. It is taken from the
specification of the package gnat.regpat, where additional
information is available.
regexp ::= expr
::= ^ expr -- anchor at the beginning of string
::= expr $ -- anchor at the end of string
expr ::= term
::= term | term -- alternation (term or term ...)
term ::= item
::= item item ... -- concatenation (item then item)
item ::= elmt -- match elmt
::= elmt * -- zero or more elmt's
::= elmt + -- one or more elmt's
::= elmt ? -- matches elmt or nothing
::= elmt *? -- zero or more times, minimum number
::= elmt +? -- one or more times, minimum number
::= elmt ?? -- zero or one time, minimum number
::= elmt { num } -- matches elmt exactly num times
::= elmt { num , } -- matches elmt at least num times
::= elmt { num , num2 } -- matches between num and num2 times
::= elmt { num }? -- matches elmt exactly num times
::= elmt { num , }? -- matches elmt at least num times
non-greedy version
::= elmt { num , num2 }? -- matches between num and num2 times
non-greedy version
elmt ::= nchr -- matches given character
::= [range range ...] -- matches any character listed
::= [^ range range ...] -- matches any character not listed
::= . -- matches any single character
-- except newlines
::= ( expr ) -- parens used for grouping
::= \ num -- reference to num-th parenthesis
range ::= char - char -- matches chars in given range
::= nchr
::= [: posix :] -- any character in the POSIX range
::= [:^ posix :] -- not in the POSIX range
posix ::= alnum -- alphanumeric characters
::= alpha -- alphabetic characters
::= ascii -- ascii characters (0 .. 127)
::= cntrl -- control chars (0..31, 127..159)
::= digit -- digits ('0' .. '9')
::= graph -- graphic chars (32..126, 160..255)
::= lower -- lower case characters
::= print -- printable characters (32..127)
::= punct -- printable, except alphanumeric
::= space -- space characters
::= upper -- upper case characters
::= word -- alphanumeric characters
::= xdigit -- hexadecimal chars (0..9, a..f)
char ::= any character, including special characters
ASCII.NUL is not supported.
nchr ::= any character except \()[].*+?^ or \char to match char
\n means a newline (ASCII.LF)
\t means a tab (ASCII.HT)
\r means a return (ASCII.CR)
\b matches the empty string at the beginning or end of a
word. A word is defined as a set of alphanumerical
characters (see \w below).
\B matches the empty string only when *not* at the
beginning or end of a word.
\d matches any digit character ([0-9])
\D matches any non digit character ([^0-9])
\s matches any white space character. This is equivalent
to [ \t\n\r\f\v] (tab, form-feed, vertical-tab,...
\S matches any non-white space character.
\w matches any alphanumeric character or underscore.
This include accented letters, as defined in the
package Ada.Characters.Handling.
\W matches any non-alphanumeric character.
\A match the empty string only at the beginning of the
string, whatever flags are used for Compile (the
behavior of ^ can change, see Regexp_Flags below).
\G match the empty string only at the end of the
string, whatever flags are used for Compile (the
behavior of $ can change, see Regexp_Flags below).
... ::= is used to indication repetition (one or more terms)
Embedded newlines are not matched by the ^ operator. It is possible to retrieve the substring matched a parenthesis expression. Although the depth of parenthesis is not limited in the regexp, only the first 9 substrings can be retrieved.
The operators '*', '+', '?' and '{}' always match the longest possible substring. They all have a non-greedy version (with an extra ? after the operator), which matches the shortest possible substring.
For instance:
regexp="<.*>" string="<h1>title</h1>" matches="<h1>title</h1>"
regexp="<.*?>" string="<h1>title</h1>" matches="<h1>"
'{' and '}' are only considered as special characters if they appear in a substring that looks exactly like '{n}', '{n,m}' or '{n,}', where n and m are digits. No space is allowed. In other contexts, the curly braces will simply be treated as normal characters.
Note that if you compiled AdaControl with the
String_Matching_Portable package, only basic wildcards are
available, i.e. only “*” and “?” are supported, where “*”
matches any string of character and “?” matches a single character.
This chapter is intended to users of a previous version of AdaControl, who want to migrate rule files to the latest version. Although we understand the burden of non upward-compatible changes, we consider that making AdaControl more powerful and easier to use is sometimes more important than strict compatibility. Moreover, in most cases the changes are very straightforward and can be done easily by hand, or with scripts if many files are involved.
The subrule Real_Equality does not control user-defined
equality operators any more. This is intended to be more of an
improvement than an incompatibily.
Since the number of subrules is growing, and do not only address `special” comments, this rule has been renamed to “comments”.
Due to a bug/feature of the GPS interface, if a units file was specified, it did not reappear later in the corresponding box of the Switch/AdaControl dialog. This has been fixed, but you must reenter the units file name in the dialog.
The introduction of categories made some subrules syntactically ambiguous or redundant. In consequence, the subrules “derived_record”, “extension_record”, and “tagged_record” have been removed, and the subrules “record”, “incomplete_record”, and “non_contiguous_record” have been renamed as “layout”, “incomplete_layout”, and “non_contiguous_layout” respectively. Change:
check representation_clause (derived_record);
check representation_clause (extension_record);
check representation_clause (tagged_record);
check representation_clause (record);
check representation_clause (incomplete_record);
check representation_clause (non_contiguous_record);
to:
check representation_clause (new layout);
check representation_clause (extension layout);
check representation_clause (tagged layout);
check representation_clause (layout);
check representation_clause (incomplete_layout);
check representation_clause (non_contiguous_layout);
The subrule “Max_Length” has been changed to “Length”, with the possibility to specify both min and max values. Change:
check array_declarations (max_length, 100);
to:
check array_declarations (length, max 100);
The subrule names “initialized_record_field”, “uninitialized_record_field”, “initialized_protected_field”, and “uninitialized_protected_field” have been changed to “initialized_record_component”, “uninitialized_record_component”, “initialized_protected_component”, and “uninitialized_protected_component”, respectively, to be more consistent with official Ada terminology. Change:
check declarations (initialized_record_field,
uninitialized_record_field,
initialized_protected_field,
uninitialized_protected_field);
to:
check declarations (initialized_record_component,
uninitialized_record_component,
initialized_protected_component,
uninitialized_protected_component);
The subrule “aliased” has been split into “aliased_constant” and “aliased_variable”. The old rule controlled both at the same time, but did not control aliased components (there are now other subrules to that effect). Change:
check declarations (aliased);
to:
check declarations (aliased_constant, aliased_variable);
The <place> is no more allowed to be “all”, because it was ambiguous with the “all <name>” syntax of <entity>. If you used “all”, duplicate the control with “calls” and “instantiations”. Change:
My_label : check default_parameter (all, ...);
to:
My_label : check default_parameter (calls, ...),
check default_parameter (instantiations, ...);
By default, variables declared directly within (generic) package specifications and bodies are no more checked. To get the previous behaviour, add the “package” modifier. Change:
check improper_initialization (variable);
to:
check improper_initialization (package variable);
If the output format is CSV or CSVX, the file name, line number and column number are generated as three different spreadsheet columns, instead of forming a single message. This makes it easier to use a spreadsheet program for per-file statistics.
Due to the introduction of the “positional” keyword, “not used” is now spelled “not_used”. Change:
check default_parameter (proc, param, not used);
to:
check default_parameter (proc, param, not_used);
This rule has been changed into a subrule of the (new) rule “Dependencies”. Change:
check Other_Dependencies (pack1, pack2);
to:
check Dependencies (others, pack1, pack2);
Due to the introduction of another subrule, add “pattern” as the first parameter to the rule. Change:
check Special_Comments ("TBSL");
to:
check Special_Comments (pattern, "TBSL");
The “raise” subrule now reports all occurrences of the raise statement, even if another control is applicable to the same statement.
The “reraise” subrule now reports calls to
Ada.Exceptions.Reraise_Occurrence.
The “raise_standard” subrule now reports exceptions raised by calls to
Ada.Exceptions.Raise_Exception.
This rule now allows the specification of both min and max values for each subrule. Subrule names have been changed accordingly. Change:
check Case_Statement (max_range_span, 5);
check Case_Statement (max_values, 10);
check Case_Statement (min_others_span, 4);
check Case_Statement (min_paths, 6);
to:
check Case_Statement (range_span, max 5);
check Case_Statement (values, max 10);
check Case_Statement (others_span, min 4);
check Case_Statement (paths, min 6);
This rule has been changed into a subrule of the (new) rule “Parameter_Declarations”. Change:
check Max_Parameters (10);
to:
check Parameter_Declarations (Max_Parameters, 10);
The message is now syntactically a string, and must always be enclosed in double quotes (quotes were optional in previous versions).
If a “source” command is given in a rules file, and the sourced file is given with a relative path, it is interpreted relatively to the sourcing file (it was interpreted relatively to the current directory previously). This should make “chained” sourcing easier, since the interpretation does not depend on where the sourcing file is being called from.
This rule is now called “Characters” and can process other kinds of characters in addition to control characters. Control characters correspond to the “control” parameter of the rule. Change:
check control_characters;
to:
check characters (control);
This rule has been changed into a subrule of the (new) rule “simplifiable_statements”. Change:
check if_for_case;
to:
check simplifiable_statements (if_for_case);
The rule does not print the number of instantiations any more, since the same effect can be achieved with the “count” control kind.
This rule has been removed, since its effect can now be achieved with other rules: the rule “declarations” to check for local instantiations of any generic, and the rule “instantiations” to check for local instantiations of specified generics. Change:
R1: check Local_Instantiation;
R2: search Local_Instantiation (Ada.Unchecked_Conversion);
to:
R1: check declarations (local instantiation);
R2: search Instantiations (local Ada.Unchecked_Conversion);
Quotes are no more optional around patterns.
The <location> modifier is now before the <filter_kind> (it was before the pattern previously). This may require splitting the rule in two in some cases. For example, change:
check naming_convention (object, local "^L_", global "^G_");
to:
check naming_convention (local object, "^L_");
check naming_convention (global object, "^G_");
The name of this rule has been changed to “improper_initialization”, since it now controls other cases of improper initialization.
Quotes are no more optional around patterns.
Two subrules of this rule have migrated to the new rule “simplifiable_statements” (with slightly different names). Change:
check statements (unnecessary_null);
check statements (while_true);
to:
check simplifiable_statements (null);
check simplifiable_statements (loop);
The subrule “Formal_In_Out” has been renamed as “In_Out_Generic_Parameter”, for consistency with the new “In_Out_Parameter” subrule.
The subrules “renames” and “not_operator_renames” have been renamed to “renaming” and “not_operator_renaming”.
As a consequence of being able to specify the location of any construct, the subrules “nested_function_instantiation”, “nested_generic_function”, “nested_generic_package”, “nested_generic_procedure”, “nested_package”, “nested_package_instantiation”, and “nested_procedure_instantiation” have been removed and replaced with the corresponding general construct (without “nested_”). You can have the same effect by specifying the “nested” modifier in front of them. I.e., change:
check declarations (nested_generic_function);
to:
check declarations (nested generic_function);
The <location> keyword is placed before the <Filter_Kind> keyword instead of before the <Pattern>, which looks more natural. The “Any” keyword has been removed, since omitting the <location> keyword has the same effect. Change:
check naming_convention (variable, global "^G_");
check naming_convention (package, any "^Pack_");
to:
check naming_convention (global variable, "^G_");
check naming_convention (package, "^Pack_");
This rule is now called Non_Static, since it is no more restricted to constraints. The parameters “index” and “discriminant” have been changed to “index_constraint” and “discriminant_constraint”, respectively. Change:
check non_static_constraint (index, discriminant);
to:
check non_static (index_constraint, discriminant_constraint);
This rule has been renamed to Insufficient_Parameters, since it does no more
handle the “maximum” subrule. Controlling positional parameters according to their number
is now done by the rule style (positional_association). Change:
check positional_parameters (maximum, 3);
check positional_parameters (insufficient, 2, Boolean);
to:
check style (positional_association, call, 3);
check insufficient_parameters (2, Boolean);
This rule is no more a rule of its own, it is a subrule of the (new) rule Expressions, whose name is Real_Equality. Change:
check Real_Operators;
to:
check expressions (Real_Equality);
The name of the subrule “casing” has been changed to “casing_identifier” since the casing of attributes and pragmas can now also be checked. The casing style is no more optional.
The name of the subrule “literal” has been changed to “numeric_literal” (since characters and strings are also literals, but are not handled by this subrule).
The subrule “exposed_literal” now requires an extra parameter to tell whether it applies to integer literals, real literals, character literals or string literals. Allowed values are provided after this parameter, and must of course be of the appropriate type. In short, if you had:
check style (exposed_literal, 0, 1, 0.0, 1.0);
you must change it to:
check style (exposed_literal, integer, 0, 1)
check style (exposed_literal, real, 0.0, 1.0);
The “aggregate” parameter of the subrule “positional_association” has been split into “array_aggregate” and “record_aggregate”. For example, change:
check style (positional_association, aggregate);
into:
check style (positional_association, record_aggregate, array_aggregate);
The XML file used to describe AdaControl features to GPS used to be
called adactl.xml. It is now called zadactl.xml, since
GPS processes its initialization files in alphabetical order. This
avoids shuffling the menus when AdaControl support is activated.
Make sure to remove the old adactl.xml file from the GPS
plug-ins directory before installing the new version.
The parameters “access” and “access_subprogram” have been changed to “access_type” and “access_subprogram_type”, for consistency with the new parameters.
A keyword has been added to specify the required number of comment lines. Change:
check Header_Comments (10);
to:
check Header_Comments (minimum, 10);
This rule is now part of the “style” rule. Change:
check|search|count No_Closing_Name;
to:
check|search|count Style (No_Closing_Name);
This rule is now part of the “usage” rule. Change:
check|search|count Specification_Objects (<parameters>);
to:
check|search|count Usage (Object, From_Spec, <parameters>);
Name changed from “statement” to “statements” (added an 's'), to be consistent with other rules.
This rule is now part of the “statements” rule. Change:
check|search|count When_Others_Null (case);
check|search|count When_Others_Null (exception);
to:
check|search|count Statements (case_others_null);
check|search|count Statements (exception_others_null);