Re: static variables

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Dec 11, 2013, 01:45 UTC
Message-ID: <20131211014501.GI2311@google.com>
In-Reply-To: <CAHOQ7J-rO-KjHyYk1Gw6Wv+iH_M7DPr76t3G7YN_sUv3YqcJcg@mail.gmail.com>

Stefan Zager wrote:

> This is probably a naive question, but: there are quite a lot of static
> variables in the git code where it's really unnecessary.  Is that just a
> historical artifact, or is there some reason to prefer them?

Sometimes it's for convenience. Other times it's to work around C89's requirement that initializers can't include pointers to automatic variables, so when using parse_options, old commands tend to use statics for the variables initialized by options. (Since then, git has stopped following that so rigidly, which is probably a good thing.)

Worse, some functions have static buffers when they need a large buffer and want to avoid too much allocation churn. As a general rule, historically very little of git's code (mostly pack related) needed to be usable with threads, though of course it would be excellent to fix more code to be thread-safe.

> As an example, here's an excerpt from symlnks.c.  In addition to being
> static, if I'm reading this right, it appears that the 'removal' variable
> is used before it's initialized:

statics are allocated from the .bss section, where they are zeroed automatically.

> static struct removal_def {
>   char path[PATH_MAX];
>   int len;
> } removal;

Plumbing this through the call stack instead of using a static sounds like a good idea. That would mean allocating the removal_def in unpack-trees.c::check_updates, I think (see v1.6.3-rc0~147^2~16, "unlink_entry(): introduce schedule_dir_for_removal()", 2009-02-09 for context). Then the loop could be divided into chunks that each use their own removal_def or something.

Sometimes when git needs parallelism and threads don't work, it uses fork + exec (aka run_command). Making the relevant functionality thread-safe is generally much nicer, though.

Thanks and hope that helps, Jonathan