mu/HACKING

122 lines
4.2 KiB
Plaintext

* HACKING
Here are some guidelines for hacking on the 'mu' source code. This is fairly
long list -- this is not meant to discourage anyone from working on mu; I
think most of the rules are common sense anyway (or should be), and some of
the more stylistic-aesthetic rules are clearly visible in current source code,
so as long as any new code 'fits in', it should go a long way in satisfying
them.
I should add some notes for the Lisp/Scheme code as well...
** Coding style
For consistency and, even more important, to keep things understandable, mu
attempts to follow the following rules:
1. basic code layout is like in the Linux kernel coding style, with the '{'
on the same line as the statement, except for functions. Tabs/spaces
have width 8.
2. lines must not exceed 80 characters (C) or 100 characters (C++)
3. functions must not exceed 35 lines. You can easily check if any functions
violate this rule with 'make line35', which lists all functions with more
than 35 non-comment lines.
4. source files should not exceed 1000 lines
5. a function's cyclomatic complexity should not exceed 10 (there could be
rare exceptions, currently there are none in mu). You can test the
cyclomatic complexity with the pmccabe tool; if you installed that, you
can use 'make cc10' to list all functions that violate this rule; there
should be none.
6. filenames have their components separated with dashes (e.g, 'mu-log.h'),
and start with 'mu-' where appropriate.
7. global functions have the prefix based on their module, e.g., mu-foo.h
declares a function of 'mu_foo_bar (int a);', mu-foo.c implements this.
8. non-global functions *don't* have the module prefix, and are declared
static.
9. functions have their return type on a separate line before the function
name, so:
int
foo (const char* bar)
{
....
}
There is no non-aesthetic reason for this.
10. in C code, variable-declarations are at the beginning of a block; in
principle, C++ follows that same guideline, unless for heavy yet
uncertain initializations following RAII. In C code, the declaration does
*not* initialize the variable. This will give the compiler a chance to
warn us if the variable is not initialized in a certain code path.
11. returned strings of type char* must be freed by the caller; if they are
not to be freed, 'const char*' should be used instead
12. functions calls have a space between function name and arguments, unless
there are none, so:
foo (12, 3);
and
bar();
after a comma, a space should follow.
13. functions that do not take arguments are explicitly declared as f(void)
and not f(). Reason: f() means that the arguments are /unspecified/ (in
C)
** Logging
For logging, mu uses the GLib logging functions/macros as listed below,
except when logging may not have been initialized.
The logging system redirects most logging to the log file (typically,
~/.mu/mu.log). g_warning, g_message and g_critical are shown to the user,
except when running with --quiet, in which case g_message is *not* shown.
- g_message is for non-error messages the user will see (unless running
with --quiet)
- g_warning is for problems the user may be able to do something about (and
they are written on stderr)
- g_critical is for mu bugs, serious, internal problems (g_return_if_fail and
friends use this). (and they are written on stderr)
- don't use g_error
if you just want to log something in the log file without writing to screen,
use MU_LOG_WRITE, as defined in mu-util.h
** Compiling from git
For hacking, you're strongly advised to use the latest git
version. Compilation from git should be straightforward, if you have the
right tools (automake, autoconf and friends) installed -- but if you do
development, you'd probably have those.
Anyhow, to compile straight from git:
$ git clone https://github.com/djcb/mu
$ cd mu
$ autoreconf -i
$ ./configure
$ make
# Local Variables:
# mode: org; org-startup-folded: nofold
# End: