Re: [PATCH v2 4/4] ref-filter: use oid_object_info() to get object

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 17, 2018, 22:17 UTC
Message-ID: <xmqqtvoxld68.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <CAL21BmnDqv9yib5Vqh4q7zj1hd_bmhO+xnDVJBi-zM0Cqtsevw@mail.gmail.com>

Оля Тележная  <olyatelezhnaya@gmail.com> writes:

Show 7 quoted lines

>> Hmph, doesn't this belong to the previous step?  In other words,
>> isn't the result of applying 1-3/4 has a bug that can leave eaten
>> uninitialized (and base decision to free(buf) later on it), and
>> isn't this change a fix for it?
>
> Oh. I was thinking that it was new bug created by me. Now I see that
> previously we had the same problem.

The original said something like:

	int eaten;
	void *buf = get_obj(..., &eaten);
	...
	if (!eaten)
		free(buf);

and get_obj() left eaten untouched when it returned NULL. As a random uninitialized cruft in eaten that happened to be "true" would just cause free(NULL) on many archs, there was no practical problem in such a code, but it is undefined behaviour nevertheless.

And the previous step made it a bit more alarming ;-)