2011-01-21
Should be:
diff -u
unifed context format preferred.Do not make large commits that only change code style unless you have
previously had this agreed or it is in code under major refactoring
where a diff is not a large concern. There is always diff -b
when
large code style (whitespace) changes are made.
2 spaces. No tabs.
All code must be wrapped to 80 chars as far as is possible. Function
definitions or calls should indent the parameters to the left (
.
Redland libraries use very long function names following the naming convention which can make linebreaking very hard. In this case, indent function parameters on new lines 4 spaces after the function name like this:
var = function_name_with_very_long_name_that_is_hard_to_wrap_args(
argument1_with_very_long_name_or_expression,
argument2,
..)
Use no space between a keyword followed by braces argument.
For example, use if(cond)
rather than if (cond)
(ditto for
while, do etc.) and
functionname(...)
rather than functionname (...)
in
both definition and calls of functions.
Put spaces around operators in expressions, assignments, tests, conditions
GOOD:
a += 2 * x
if(a < 2)
BAD:
a+=2*x
if(a<2)
When comparing to 0
or a NULL
pointer, use the idiomatic form
that has no comparison.
GOOD:
if(!ptr)
if(!index)
BAD:
if(ptr == NULL)
if(index == 0)
if(0 == index)
When comparing a variable to a constant, the code has currently used
if(var == constant)
rather than the slightly safer, and easier to
compile check, if(constant == var)
.
In general add {}s around blocks in if else chains when one of the blocks has more than 1 line of code. Try not to mix, but the final case if it is one line, can be braceless.
if(var == 1) {
... multiple lines of code ...
} else {
... multiple lines of code ...
}
or
if(var == 1)
... one line of code
else
... one line of code
or
if(var == 1) {
... multiple lines of code ...
} else if(var == 2) {
... multiple lines of code and / or more if conditions ...
} else
... one line of code ...
If using if else chains on an enumeration, don't do that, use a
switch()
which GCC can use to find missing cases when they get added.
switch(enum_var) {
case ENUM_1:
... code ...
break;
case ENUM_2:
... code ...
break;
case ENUM_DONT_CARE:
default:
... code ...
break;
}
There should ALWAYS be a default:
case.
Declare functions in this format:
returntype
functionname(type1 param1, type2 param2, ...)
{
type3 var1;
type4 var2;
... first line of code ...
tidy:
... cleanup code...
return value;
}
Notes:
{}
blocks. The
form if(1) { ... var decls ...; more code }
may be used but
a code rewrite is preferable.return
are allowed but for obvious error or result
returns. Do not twist the code to enable a single return.goto
may be used for resource cleanup and result return
where control flow only goes forward.Public functions, types, enumerations and defines must have autodocs -
the structured comment block before the definition. This is read by
gtk-doc(1)
to generate reference API documentation.
Format:
/**
* functionname:
* @param1: Description of first parameter
* @param2: Description of second parameter (or NULL)
* ... more params ...
*
* Short Description
*
* Long Description.
*
* Return value: return value
*/
returntype
functionname(...)
{
... body ...
}
The Short Description have several commmon forms:
The latter is used for autodocs for internal functions either as internal documentation or for APIs that may one day be public.
The (or NULL) phrase is used for pointer parameters that may be omitted. This is usually tested by the function as an assertion. In some functions there are more complex conditions on which optional parameters are allowed, these are described in the Long Description.
The general standard for Redland libraries using GIT is a merge of the GIT standards format and GNU ChangeLog
First line summaries what commit does - this goes into the GIT short log
(function1, function2): what changed
(function3): Added, deprecating function4()
(function4): Deleted, replaced by function3()
struct foo gains field ...
struct bar loses field ...
enum blah gains new value BLAH_2 which ...
Use name()
in the description for references to functions.
Make sure to do (function1, function2) NOT (function1,function2) as
it makes things easier to format later.
Sometimes it's short enough (good) that it all can be done in the first line, pretty much only if it's a small change to a single function.
If the change is trivial / typo and IMPORTANT NOT a commit to code, then the commit can start with '#'. This may get filtered out of commit log message notifications and ChangeLog.
e.g. #spelling
or #ws
the latter is whitespace changes for some reason
The changes will semi-automatically be added to the ChangeLog files following the GNU style, indented and word wrapped, and adding the list of files at the start. So the commit message above ends up looking something like:
2010-08-23 User Name <user@example.org>
* dir/file1.c, dir2/file2.c: First line summaries what commit
does - this goes into the GIT short log
(function1, function2): what changed
(function3): Added, deprecating function4()
(function4): Deleted, replaced by function3()
struct foo gains field ...
struct bar loses field ...
enum blah gains new value BLAH_2 which ...
Copyright (C) 2010 Dave Beckett