Re: [PATCH] builtin/init-db: preemptively clear repo_fmt to avoid leaks

[Andrzej Hunt <andrzej@ahunt.org>]

Hi Johannes,

On 05/05/2021 21:28, Johannes Schindelin wrote:
> Hi Andrzej,
> 
> On Wed, 5 May 2021, Andrzej Hunt via GitGitGadget wrote:
> 
>> diff --git a/builtin/init-db.c b/builtin/init-db.c
>> index 31b718259a64..b25147ebaf59 100644
>> --- a/builtin/init-db.c
>> +++ b/builtin/init-db.c
>> @@ -512,6 +512,7 @@ int init_db(const char *git_dir, const char *real_git_dir,
>>   			       git_dir, len && git_dir[len-1] != '/' ? "/" : "");
>>   	}
>>
>> +	clear_repository_format(&repo_fmt);
> 
> I am afraid that this might not be correct, as t0410.27 experienced a
> segmentation fault (see
> https://github.com/git/git/pull/1018/checks?check_run_id=2511749719#step:5:2845
> for the full details):


Thanks for spotting that. On further investigation this looks like a 
preexisting issue on seen (my github PR was based on seen - in hindsight 
that was probably not a good idea) - here's a CI run from seen this 
morning exhibiting the same failure:
   https://github.com/git/git/runs/2515095446?check_suite_focus=true

To gain more confidence, I've rebased my patch onto next, and I no 
longer hit any CI failures:
   https://github.com/ahunt/git/runs/2515309312?check_suite_focus=true
(I ran this on my fork because changing the base branch appears to have 
confused the PR's CI configuration.)


Nevertheless, I was still curious about what might be causing the 
failure in the first place: it appears to only happen in CI with gcc on 
Linux, and I was not able to repro on my machine using (tested with 
gcc-7.5 and gcc-10, with seen from this morning):
   make CC=gcc-10 T=t0410-partial-clone.sh test

I'm guessing that understanding this failure might require the help of 
someone with an Ubuntu install to debug (unless there's some easy way of 
using Github's Actions to run a bisect)?

-- snip --

ATB,

   Andrzej