From: Junio C Hamano <gitster@pobox.com>
Date: Tue, 11 Jun 2024 15:17:08 GMT
Subject: Re: [PATCH 0/4] varargs functions with __attributes__(())
Message-ID: <xmqqcyony0ej.fsf@gitster.g>
In-Reply-To: <20240611081735.GI3248245@coredump.intra.peff.net>

Jeff King <peff@peff.net> writes:

> On Sat, Jun 08, 2024 at 11:37:43AM -0700, Junio C Hamano wrote:
>
>> There are several varargs functions that take either NULL-terminated
>> list of parameters, or printf-like format followed by list of
>> parameters, that are not declared as such with __attributes__(()).
>> 
>> Adding such a missing attribute to trace2_region_enter_printf()
>> revealed that an existing call to it was trying to format a value of
>> type size_t using "%d", which is not such an excellent idea.  Other
>> functions that were lacking attributes fortunately did not have any
>> broken existing callers.
>
> Great, I am happy to see these. I assume you found them all by grepping?
>
> I wonder if there is a way to convince the compiler (or coccinelle) to
> complain about any varargs function that does not have one of our
> usual annotations. It's possible to have other conventions (e.g., an
> "int" up front specifying the number of entries) but in practice I doubt
> we would ever use one.
>
> Still, I suspect the answer is probably "no", there is not an easy way
> to do it.

I just went from grepping for fixed "...)" in *.[ch] files.  I do
admit I wished some form of automation, but didn't come up with one.

I was happy that the exercise found a real bug ;-)  I started it
only as a clean-up topic.