From: Jonathan Nieder <jrnieder@gmail.com>
Date: Wed, 16 Mar 2011 06:28:48 GMT
Subject: Re: [PATCH/RFC] reflog: silence -O3 -Wuninitialized warning
Message-ID: <20110316062831.GA5252@elie>
In-Reply-To: <7vfwqnabbi.fsf@alter.siamese.dyndns.org>

Junio C Hamano wrote:

> We could obviously do (2) or (3), but the thing is, I don't think we can
> have much confidence on -Wuninitialized warnings from this compiler once
> we go down that route.  Is it _guaranteed_ that the compiler bug _always_
> err on the false-positive side?
>
> IOW, I'd very much prefer (1) for this particular case and if somebody
> really cares (2).

*nod*.  The tricky heuristics here are that (A) a call to the function

 extern void foo(int *var);

is assumed to initialize *var (to support idioms like strbuf_init), while
(B) no assumption is made about the return value of a call to the function

 extern int bar(void);

The assumption (B) is the scourge of -Wuninitialized users.  It is
always going produce a lot of false positives --- the compiler simply
doesn't have enough information to completely analyze the flow through
a function.  (Even if it did have enough information, completely
solving the problem is Turing-complete.)

Problems from (A) are more rare.  -Wuninitialized will produce _some_
false negatives, but problems it misses would typically be due to
failure to check for errors and return early, which is not what
-Wuninitialized is about.

I don't mind the warning, especially since it only appears with -O3.
What makes this interesting to me is that it is a reminder that the
compiler has very little information about the flow of control.  A
simple

 #define error(...) (report_error(__VA_ARGS__), -1)

could open the door to some nice micro-optimizations.