[vlc-devel] [vlc-commits] src/missing: fix a bunch of clang compilation warnings and update copyright header

[Rémi Denis-Courmont]

Le lundi 3 mars 2014, 18:50:52 Felix Paul Kühne a écrit :
> Hey Rémi,
> 
> On 03.03.2014, at 18:42, Rémi Denis-Courmont <remi at remlab.net> wrote:
> > Le lundi 3 mars 2014, 18:37:21 Felix Paul Kühne a écrit :
> >> vlc | branch: master | Felix Paul Kühne <fkuehne at videolan.org> | Mon Mar 
> >> 3
> >> 18:28:42 2014 +0100| [374bd117c4390bcfbc1b56f40b1183560f256b0c] |
> >> committer: Felix Paul Kühne
> >> 
> >> src/missing: fix a bunch of clang compilation warnings and update
> >> copyright
> >> header
> > 
> > There should be no warning on debug builds, since assert(0) is effectively
> > a synonym for abort().
> 
> Well, this is true indeed. However, clang 5 seems to believe differently.

clang is right for the release case, but it is totally silly for the debug 
case. Indeed, GCC does not warn about this here. There might be a "bug" in 
Apple libc, failing to flag the assert failure function as non-returning.

> Here are 2 logs with the code state prior to my patch:
> with --disable-debug: https://pastee.org/y79r5 (70 warnings)
> with --enable-debug: https://pastee.org/jye6f (55 warnings)
> 
> > And now, there are a lot of unreachable code warnings…
> 
> Meh. I'll happily revert part of the patch (the return values) since the
> remainder of unused variables already saves 55 warnings for me.

There are two problems here. The first problem is that there is a legitimate 
warning in release (NDEBUG) build. To fix it, we would need a lot of ifdefs. I 
am not sure this is worth the effort, as developers should use debug builds.

The second problem is that unreachable code and missing return value warnings 
are fundamentally contradictory. We can absolutely not fix both, so for debug 
builds, we have to choose whether we optimize for stupid or smart SDK's.

-- 
Rémi Denis-Courmont
http://www.remlab.net/