Re: [PATCH 06/11] config: use git_config_string() in easy cases

From: Jeff King <peff@peff.net>
Date: Apr 7, 2024, 02:47 UTC
Message-ID: <20240407024733.GB1585186@coredump.intra.peff.net>
In-Reply-To: <CAPig+cST40pM5PdGA==o8aGFwv8kStzzC8=28NULUtV8OFJmLg@mail.gmail.com>

On Sat, Apr 06, 2024 at 10:05:00PM -0400, Eric Sunshine wrote:

Show 6 quoted lines

> On Sat, Apr 6, 2024 at 9:02 PM Jeff King <peff@peff.net> wrote:
> > config: use git_config_string() in easy cases
> 
> Did you instead mean?
> 
>     config: use git_config_string_dup() in easy cases

Oops, yes.

Show 9 quoted lines

> > (like the one above), we can easily convert them to use a variant of the
> > helper that frees the existing value before writing over it. That works
> > because free(NULL) is a noop. What we can't change, though, is cases
> > where the pointer is initialized to a string literal, like:
> 
> Fun historical (historic?) fact: On NeXTSTEP operating system,
> free(NULL) and realloc(NULL,n) would crash the program, so getting
> non-native (open-source) code to run correctly often involved more
> than a little effort..

I'm happy that's no longer something we need to worry about. ;) IIRC we had some discussion long ago about whether free(NULL) needed to be protected, and came to the conclusion that it's OK everywhere.

(There's a discussion in a nearby thread about a possible git_free() wrapper. The point there is errno handling, but of course it could also cover this case for us if we found a modern platform where it was a problem).

-Peff