Remove not needed __builtin_expect due to malloc predictor.

[official-gcc.git] / gcc / ada / gnat-style.texi

\input texinfo   @c -*-texinfo-*-
@c %**start of header

@c oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo
@c                                                                            o
@c                           GNAT DOCUMENTATION                               o
@c                                                                            o
@c                     G N A T   C O D I N G   S T Y L E                      o
@c                                                                            o
@c                     Copyright (C) 1992-2012, AdaCore                       o
@c                                                                            o
@c oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo

@setfilename gnat-style.info

@copying
Copyright @copyright{} 1992-2012, AdaCore

Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with no
Invariant Sections, with no Front-Cover Texts and with no Back-Cover
Texts.  A copy of the license is included in the section entitled
``GNU Free Documentation License''.
@end copying

@settitle GNAT Coding Style
@setchapternewpage odd

@include gcc-common.texi

@dircategory Software development
@direntry
* gnat-style: (gnat-style).      GNAT Coding Style
@end direntry

@macro syntax{element}
@t{\element\}
@end macro
@c %**end of header

@titlepage
@titlefont{GNAT Coding Style:}
@sp 1
@title A Guide for GNAT Developers
@subtitle GNAT, The GNU Ada Compiler
@versionsubtitle
@author Ada Core Technologies, Inc.
@page
@vskip 0pt plus 1filll

@insertcopying
@end titlepage

@raisesections

@node    Top,       General,         , (dir)
@comment node-name, next,    previous, up

@ifnottex
@noindent
GNAT Coding Style@*
A Guide for GNAT Developers
@sp 2
@noindent
GNAT, The GNU Ada Compiler@*

@noindent
@insertcopying
@end ifnottex


@menu
* General::
* Lexical Elements::
* Declarations and Types::
* Expressions and Names::
* Statements::
* Subprograms::
* Packages::
* Program Structure::
* GNU Free Documentation License::
* Index::
@end menu

@c  -------------------------------------------------------------------------
@node    General, Lexical Elements, Top, Top
@section General
@c  -------------------------------------------------------------------------

@noindent
Most of GNAT is written in Ada using a consistent style to ensure
readability of the code.  This document has been written to help
maintain this consistent style, while having a large group of developers
work on the compiler.

For the coding style in the C parts of the compiler and run time,
see the GNU Coding Guidelines.

This document is structured after the @cite{Ada Reference Manual}.
Those familiar with that document should be able to quickly
lookup style rules for particular constructs.


@c  -------------------------------------------------------------------------
@node    Lexical Elements, Declarations and Types, General, Top
@section Lexical Elements
@c  -------------------------------------------------------------------------
@cindex Lexical elements

@subsection Character Set and Separators
@c  -------------------------------------------------------------------------
@cindex Character set
@cindex ASCII
@cindex Separators
@cindex End-of-line
@cindex Line length
@cindex Indentation

@itemize @bullet
@item
The character set used should be plain 7-bit ASCII@.
The only separators allowed are space and the end-of-line sequence.
No other control character or format effector (such as @code{HT},
@code{VT}, @code{FF})
should be used.
The normal end-of-line sequence is used, which may be
@code{LF}, @code{CR/LF} or @code{CR},
depending on the host system.  An optional @code{SUB}
(@code{16#1A#}) may be present as the
last character in the file on hosts using that character as file terminator.

@item
Files that are checked in or distributed should be in host format.

@item
A line should never be longer than 79 characters, not counting the line
separator.

@item
Lines must not have trailing blanks.

@item
Indentation is 3 characters per level for @code{if} statements, loops, and
@code{case} statements.
For exact information on required spacing between lexical
elements, see file @file{style.adb}.
@cindex @file{style.adb} file
@end itemize


@subsection Identifiers
@c  -------------------------------------------------------------------------
@itemize @bullet
@cindex Identifiers

@item
Identifiers will start with an upper case letter, and each letter following
an underscore will be upper case.
@cindex Casing (for identifiers)
Short acronyms may be all upper case.
All other letters are lower case.
An exception is for identifiers matching a foreign language.  In particular,
we use all lower case where appropriate for C@.

@item
Use underscores to separate words in an identifier.
@cindex Underscores

@item Try to limit your use of abbreviations in identifiers.
It is ok to make a few abbreviations, explain what they mean, and then
use them frequently, but don't use lots of obscure abbreviations.  An
example is the @code{ALI} word which stands for Ada Library
Information and is by convention always written in upper-case when
used in entity names.

@smallexample @c adanocomment
       procedure Find_ALI_Files;
@end smallexample

@item
Don't use the variable name @code{I}, use @code{J} instead; @code{I} is too
easily confused with @code{1} in some fonts.  Similarly don't use the
variable @code{O}, which is too easily mistaken for the number @code{0}.
@end itemize

@subsection Numeric Literals
@c  -------------------------------------------------------------------------
@cindex Numeric literals

@itemize @bullet
@item
Numeric literals should include underscores where helpful for
readability.
@cindex Underscores

@smallexample
      1_000_000
      16#8000_0000#
      3.14159_26535_89793_23846
@end smallexample
@end itemize

@subsection Reserved Words
@c  -------------------------------------------------------------------------
@cindex Reserved words

@itemize @bullet
@item
Reserved words use all lower case.
@cindex Casing (for reserved words)

@smallexample @c adanocomment
       return else
@end smallexample

@item
The words @code{Access}, @code{Delta} and @code{Digits} are
capitalized when used as @syntax{attribute_designator}.
@end itemize

@subsection Comments
@c  -------------------------------------------------------------------------
@cindex Comments

@itemize @bullet
@item
A comment starts with @code{--} followed by two spaces.
The only exception to this rule (i.e.@: one space is tolerated) is when the
comment ends with a single space followed by @code{--}.
It is also acceptable to have only one space between @code{--} and the start
of the comment when the comment is at the end of a line,
after some Ada code.

@item
Every sentence in a comment should start with an upper-case letter (including
the first letter of the comment).
@cindex Casing (in comments)

@item
When declarations are commented with ``hanging'' comments, i.e.@:
comments after the declaration, there is no blank line before the
comment, and if it is absolutely necessary to have blank lines within
the comments, e.g. to make paragraph separations within a single comment,
these blank lines @emph{do} have a @code{--} (unlike the
normal rule, which is to use entirely blank lines for separating
comment paragraphs).  The comment starts at same level of indentation
as code it is commenting.
@cindex Blank lines (in comments)
@cindex Indentation

@smallexample @c adanocomment
       z : Integer;
       --  Integer value for storing value of z
       --
       --  The previous line was a blank line.
@end smallexample

@item
Comments that are dubious or incomplete, or that comment on possibly
wrong or incomplete code, should be preceded or followed by @code{???}@.

@item
Comments in a subprogram body must generally be surrounded by blank lines.
An exception is a comment that follows a line containing a single keyword
(@code{begin}, @code{else}, @code{loop}):

@smallexample @c adanocomment
@group
       begin
          --  Comment for the next statement

          A := 5;

          --  Comment for the B statement

          B := 6;
       end;
@end group
@end smallexample

@item
In sequences of statements, comments at the end of the lines should be
aligned.
@cindex Alignment (in comments)

@smallexample @c adanocomment
        My_Identifier := 5;      --  First comment
        Other_Id := 6;           --  Second comment
@end smallexample

@item
Short comments that fit on a single line are @emph{not} ended with a
period.  Comments taking more than a line are punctuated in the normal
manner.

@item
Comments should focus on @emph{why} instead of @emph{what}.
Descriptions of what subprograms do go with the specification.

@item
Comments describing a subprogram spec should specifically mention the
formal argument names.  General rule: write a comment that does not
depend on the names of things.  The names are supplementary, not
sufficient, as comments.

@item
@emph{Do not} put two spaces after periods in comments.
@end itemize

@c  -------------------------------------------------------------------------
@node    Declarations and Types, Expressions and Names, Lexical Elements,Top
@section Declarations and Types
@c  -------------------------------------------------------------------------
@cindex Declarations and Types

@itemize @bullet
@item
In entity declarations, colons must be surrounded by spaces.  Colons
should be aligned.
@cindex Alignment (in declarations)

@smallexample @c adanocomment
        Entity1   : Integer;
        My_Entity : Integer;
@end smallexample

@item
Declarations should be grouped in a logical order.
Related groups of declarations may be preceded by a header comment.

@item
All local subprograms in a subprogram or package body should be declared
before the first local subprogram body.

@item
Do not declare local entities that hide global entities.
@cindex Hiding of outer entities

@item
Do not declare multiple variables in one declaration that spans lines.
Start a new declaration on each line, instead.

@item
The @syntax{defining_identifier}s of global declarations serve as
comments of a sort.  So don't choose terse names, but look for names
that give useful information instead.

@item
Local names can be shorter, because they are used only within
one context, where comments explain their purpose.

@item
When starting an initialization or default expression on the line that follows
the declaration line, use 2 characters for indentation.

@smallexample @c adanocomment
        Entity1 : Integer :=
          Function_Name (Parameters, For_Call);
@end smallexample

@item
If an initialization or default expression needs to be continued on subsequent
lines, the continuations should be indented from the start of the expression.

@smallexample @c adanocomment
        Entity1 : Integer := Long_Function_Name
                               (parameters for call);
@end smallexample

@end itemize


@c  -------------------------------------------------------------------------
@node    Expressions and Names, Statements, Declarations and Types, Top
@section Expressions and Names
@c  -------------------------------------------------------------------------
@cindex Expressions and names

@itemize @bullet

@item
Every operator must be surrounded by spaces. An exception is that
this rule does not apply to the exponentiation operator, for which
there are no specific layout rules. The reason for this exception
is that sometimes it makes clearer reading to leave out the spaces
around exponentiation.
@cindex Operators

@smallexample @c adanocomment
       E := A * B**2 + 3 * (C - D);
@end smallexample

@item
Use parentheses where they clarify the intended association of operands
with operators:
@cindex Parenthesization of expressions
@smallexample @c adanocomment
       (A / B) * C
@end smallexample
@end itemize

@c  -------------------------------------------------------------------------
@node    Statements, Subprograms, Expressions and Names, Top
@section Statements
@c  -------------------------------------------------------------------------
@cindex Statements

@subsection Simple and Compound Statements
@c  -------------------------------------------------------------------------
@cindex Simple and compound statements

@itemize @bullet
@item
Use only one statement or label per line.
@item
A longer @syntax{sequence_of_statements} may be divided in logical
groups or separated from surrounding code using a blank line.
@end itemize

@subsection If Statements
@c  -------------------------------------------------------------------------
@cindex @code{if} statement

@itemize @bullet
@item
When the @code{if}, @code{elsif} or @code{else} keywords fit on the
same line with the condition and the @code{then} keyword, then the
statement is formatted as follows:
@cindex Alignment (in an @code{if} statement)

@smallexample @c adanocomment
@group
        if @var{condition} then
           ...
        elsif @var{condition} then
           ...
        else
           ...
        end if;
@end group
@end smallexample

@noindent
When the above layout is not possible, @code{then} should be aligned
with @code{if}, and conditions should preferably be split before an
@code{and} or @code{or} keyword a follows:

@smallexample @c adanocomment
@group
        if @var{long_condition_that_has_to_be_split}
          and then @var{continued_on_the_next_line}
        then
           ...
        end if;
@end group
@end smallexample

@noindent
The @code{elsif}, @code{else} and @code{end if} always line up with
the @code{if} keyword.  The preferred location for splitting the line
is before @code{and} or @code{or}.  The continuation of a condition is
indented with two spaces or as many as needed to make nesting clear.
As an exception, if conditions are closely related either of the
following is allowed:

@smallexample
@group
     if x = lakdsjfhlkashfdlkflkdsalkhfsalkdhflkjdsahf
          or else
        x = asldkjhalkdsjfhhfd
          or else
        x = asdfadsfadsf
     then
       ...
     end if;
@end group

@group
     if x = lakdsjfhlkashfdlkflkdsalkhfsalkdhflkjdsahf or else
        x = asldkjhalkdsjfhhfd                         or else
        x = asdfadsfadsf
     then
       ...
     end if;
@end group
@end smallexample

@item
Conditions should use short-circuit forms (@code{and then},
@code{or else}), except when the operands are boolean variables
or boolean constants.
@cindex Short-circuit forms

@item
Complex conditions in @code{if} statements are indented two characters:
@cindex Indentation (in @code{if} statements)

@smallexample @c adanocomment
@group
      if @var{this_complex_condition}
        and then @var{that_other_one}
        and then @var{one_last_one}
      then
         ...
      end if;
@end group
@end smallexample

@noindent
There are some cases where complex conditionals can be laid out
in manners that do not follow these rules to preserve better
parallelism between branches, e.g.

@smallexample @c adanocomment
@group
      if xyz.abc (gef) = 'c'
           or else
         xyz.abc (gef) = 'x'
      then
         ...
      end if;
@end group
@end smallexample


@item
Every @code{if} block is preceded and followed by a blank line, except
where it begins or ends a @syntax{sequence_of_statements}.
@cindex Blank lines (in an @code{if} statement)

@smallexample @c adanocomment
@group
        A := 5;

        if A = 5 then
           null;
        end if;

        A := 6;
@end group
@end smallexample
@end itemize

@subsection Case Statements
@cindex @code{case} statements

@itemize @bullet
@item
Layout is as below.  For long @code{case} statements, the extra indentation
can be saved by aligning the @code{when} clauses with the opening @code{case}.

@smallexample @c adanocomment
@group
       case @var{expression} is
          when @var{condition} =>
             ...
          when @var{condition} =>
             ...
       end case;
@end group
@end smallexample
@end itemize

@subsection Loop Statements
@cindex Loop statements

@itemize @bullet
@item
When possible, have @code{for} or @code{while} on one line with the
condition and the @code{loop} keyword.

@smallexample @c adanocomment
@group
       for J in S'Range loop
          ...
       end loop;
@end group
@end smallexample

@noindent
If the condition is too long, split the condition (see ``If
statements'' above) and align @code{loop} with the @code{for} or
@code{while} keyword.
@cindex Alignment (in a loop statement)

@smallexample @c adanocomment
@group
      while @var{long_condition_that_has_to_be_split}
        and then @var{continued_on_the_next_line}
      loop
         ...
      end loop;
@end group
@end smallexample

@noindent
If the @syntax{loop_statement} has an identifier, it is laid out as follows:

@smallexample @c adanocomment
@group
      Outer : while not @var{condition} loop
         ...
      end Outer;
@end group
@end smallexample
@end itemize

@subsection Block Statements
@cindex Block statement

@itemize @bullet
@item
The @code{declare} (optional), @code{begin} and @code{end} words
are aligned, except when the @syntax{block_statement} is named.  There
is a blank line before the @code{begin} keyword:
@cindex Alignment (in a block statement)

@smallexample @c adanocomment
@group
      Some_Block : declare
         ...

      begin
         ...
      end Some_Block;
@end group
@end smallexample

@end itemize

@c  -------------------------------------------------------------------------
@node    Subprograms, Packages, Statements, Top
@section Subprograms
@c  -------------------------------------------------------------------------
@cindex Subprograms

@subsection Subprogram Declarations
@c  -------------------------------------------------------------------------
@itemize @bullet

@item
Do not write the @code{in} for parameters.

@smallexample @c adanocomment
      function Length (S : String) return Integer;
@end smallexample

@item
When the declaration line for a procedure or a function is too long to fit
the entire declaration (including the keyword procedure or function) on a
single line, then fold it, putting a single parameter on a line, aligning
the colons, as in:

@smallexample @c adanocomment
@group
     procedure Set_Heading
       (Source : String;
        Count  : Natural;
        Pad    : Character := Space;
        Fill   : Boolean   := True);
@end group
@end smallexample

@noindent
In the case of a function, if the entire spec does not fit on one line, then
the return may appear after the last parameter, as in:

@smallexample @c adanocomment
@group
      function Head
        (Source : String;
         Count  : Natural;
         Pad    : Character := Space) return String;
@end group
@end smallexample

@noindent
Or it may appear on its own as a separate line. This form is preferred when
putting the return on the same line as the last parameter would result in
an overlong line. The return type may optionally be aligned with the types
of the parameters (usually we do this aligning if it results only in a small
number of extra spaces, and otherwise we don't attempt to align). So two
alternative forms for the above spec are:

@smallexample @c adanocomment
@group
      function Head
        (Source : String;
         Count  : Natural;
         Pad    : Character := Space)
         return   String;

      function Head
        (Source : String;
         Count  : Natural;
         Pad    : Character := Space)
         return String;
@end group
@end smallexample

@end itemize

@subsection Subprogram Bodies
@c  -------------------------------------------------------------------------
@cindex Subprogram bodies

@itemize @bullet
@item
Function and procedure bodies should usually be sorted alphabetically. Do
not attempt to sort them in some logical order by functionality. For a
sequence of subprogram specs, a general alphabetical sorting is also
usually appropriate, but occasionally it makes sense to group by major
function, with appropriate headers.

@item
All subprograms have a header giving the function name, with the following
format:

@smallexample @c adanocomment
@group
      -----------------
      -- My_Function --
      -----------------

      procedure My_Function is
      begin
        ...
      end My_Function;
@end group
@end smallexample

@noindent
Note that the name in the header is preceded by a single space,
not two spaces as for other comments. These headers are used on
nested subprograms as well as outer level subprograms. They may
also be used as headers for sections of comments, or collections
of declarations that are related.

@item
Every subprogram body must have a preceding @syntax{subprogram_declaration},
which includes proper client documentation so that you do not need to
read the subprogram body in order to understand what the subprogram does and
how to call it. All subprograms should be documented, without exceptions.

@item
@cindex Blank lines (in subprogram bodies)
A sequence of declarations may optionally be separated from the following
begin by a blank line.  Just as we optionally allow blank lines in general
between declarations, this blank line should be present only if it improves
readability. Generally we avoid this blank line if the declarative part is
small (one or two lines) and the body has no blank lines, and we include it
if the declarative part is long or if the body has blank lines.

@item
If the declarations in a subprogram contain at least one nested
subprogram body, then just before the @code{begin} of the enclosing
subprogram, there is a comment line and a blank line:

@smallexample @c adanocomment
@group
    --  Start of processing for @var{Enclosing_Subprogram}

    begin
      ...
    end @var{Enclosing_Subprogram};
@end group
@end smallexample

@item
When nested subprograms are present, variables that are referenced by any
nested subprogram should precede the nested subprogram specs. For variables
that are not referenced by nested procedures, the declarations can either also
be before any of the nested subprogram specs (this is the old style, more
generally used). Or then can come just before the begin, with a header. The
following example shows the two possible styles:

@smallexample @c adanocomment
@group
    procedure Style1 is
       Var_Referenced_In_Nested      : Integer;
       Var_Referenced_Only_In_Style1 : Integer;

       proc Nested;
       --  Comments ...


       ------------
       -- Nested --
       ------------

       procedure Nested is
       begin
          ...
       end Nested;

    --  Start of processing for Style1

    begin
       ...
    end Style1;

@end group

@group
    procedure Style2 is
       Var_Referenced_In_Nested : Integer;

       proc Nested;
       --  Comments ...

       ------------
       -- Nested --
       ------------

       procedure Nested is
       begin
          ...
       end Nested;

       --  Local variables

       Var_Referenced_Only_In_Style2 : Integer;

    --  Start of processing for Style2

    begin
       ...
    end Style2;

@end group
@end smallexample

@noindent
For new code, we generally prefer Style2, but we do not insist on
modifying all legacy occurrences of Style1, which is still much
more common in the sources.

@end itemize


@c  -------------------------------------------------------------------------
@node    Packages, Program Structure, Subprograms, Top
@section Packages and Visibility Rules
@c  -------------------------------------------------------------------------
@cindex Packages

@itemize @bullet
@item
All program units and subprograms have their name at the end:

@smallexample @c adanocomment
@group
      package P is
         ...
      end P;
@end group
@end smallexample

@item
We will use the style of @code{use}-ing @code{with}-ed packages, with
the context clauses looking like:
@cindex @code{use} clauses

@smallexample @c adanocomment
@group
      with A; use A;
      with B; use B;
@end group
@end smallexample

@item
Names declared in the visible part of packages should be
unique, to prevent name clashes when the packages are @code{use}d.
@cindex Name clash avoidance

@smallexample @c adanocomment
@group
      package Entity is
         type Entity_Kind is ...;
         ...
      end Entity;
@end group
@end smallexample

@item
After the file header comment, the context clause and unit specification
should be the first thing in a @syntax{program_unit}.

@item
Preelaborate, Pure and Elaborate_Body pragmas should be added right after the
package name, indented an extra level and using the parameterless form:

@smallexample @c adanocomment
@group
      package Preelaborate_Package is
         pragma Preelaborate;
         ...
      end Preelaborate_Package;
@end group
@end smallexample

@end itemize

@c  -------------------------------------------------------------------------
@node    Program Structure, GNU Free Documentation License, Packages, Top
@section Program Structure and Compilation Issues
@c  -------------------------------------------------------------------------
@cindex Program structure

@itemize @bullet
@item
Every GNAT source file must be compiled with the @option{-gnatg}
switch to check the coding style.
(Note that you should look at
@file{style.adb} to see the lexical rules enforced by
@option{-gnatg}).
@cindex @option{-gnatg} option (to gcc)
@cindex @file{style.adb} file

@item
Each source file should contain only one compilation unit.

@item
Filenames should be 8 or fewer characters, followed by the @code{.adb}
extension for a body or @code{.ads} for a spec.
@cindex File name length

@item
Unit names should be distinct when ``krunch''ed to 8 characters
(see @file{krunch.ads}) and the filenames should match the unit name,
except that they are all lower case.
@cindex @file{krunch.ads} file
@end itemize


@c **********************************
@c * GNU Free Documentation License *
@c **********************************
@node GNU Free Documentation License,Index, Program Structure, Top
@unnumberedsec GNU Free Documentation License
@set nodefaultgnufreedocumentationlicensenode
@include fdl.texi
@c GNU Free Documentation License
@cindex GNU Free Documentation License

@node Index,,GNU Free Documentation License, Top
@unnumberedsec Index

@printindex cp

@contents

@bye