Re: [AD] Fixing a pair of trivial warnings in Allegro 4.4

[SiegeLord <siegelordex@xxxxxxxxxx>]

In that case, I guess the best solution would be to set that flag, but 
for only that file. Something along the lines of

'set_source_files_properties(src/unix/uossmidi.c PROPERTIES 
COMPILE_FLAGS -fno-strict-aliasing)'

That'll have to be conditioned on the compiler being GCC, naturally.

-SL

On 09/01/2015 11:20 PM, Boon947 wrote:
> The aliasing bug is upstream, in linux/soundcard.h
>
> src/unix/uossmidi.c: In function 'oss_midi_key_on':
> src/unix/uossmidi.c:529:7: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
>         SEQ_SET_PATCH(seq_device, hwvoice, inst);
>         ^
> src/unix/uossmidi.c:533:4: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
>      SEQ_CONTROL(seq_device, hwvoice, CTL_PAN, pan);
>      ^
> src/unix/uossmidi.c:534:4: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
>      SEQ_BENDER(seq_device, hwvoice, 8192 + bend);
>      ^
> src/unix/uossmidi.c: In function 'oss_midi_set_volume':
> src/unix/uossmidi.c:562:4: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
>      SEQ_CONTROL(seq_device, voice, CTL_MAIN_VOLUME, vol);
>      ^
> src/unix/uossmidi.c: In function 'oss_midi_set_pitch':
> src/unix/uossmidi.c:572:4: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
>      SEQ_CONTROL(seq_device, voice, CTRL_PITCH_BENDER, 8192 + bend);
>      ^
>
>
> #define SEQ_CONTROL(dev, chn, controller, value) \
> 		_CHN_COMMON(dev, MIDI_CTL_CHANGE, chn, controller, 0, value)
>
> #define SEQ_BENDER(dev, chn, value) \
> 		_CHN_COMMON(dev, MIDI_PITCH_BEND, chn, 0, 0, value)
>
> #define _CHN_COMMON(dev, event, chn, p1, p2, w14) \
> 					{_SEQ_NEEDBUF(8);\
> 					_seqbuf[_seqbufptr] = EV_CHN_COMMON;\
> 					_seqbuf[_seqbufptr+1] = (dev);\
> 					_seqbuf[_seqbufptr+2] = (event);\
> 					_seqbuf[_seqbufptr+3] = (chn);\
> 					_seqbuf[_seqbufptr+4] = (p1);\
> 					_seqbuf[_seqbufptr+5] = (p2);\
> 					*(short *)&_seqbuf[_seqbufptr+6] = (w14);\
> 					_SEQ_ADVBUF(8);}
>
> *(short *)&_seqbuf[_seqbufptr+6] is an aliasing rule violation, AFAIU.
>
> What course of action do you suggest?
>
> Regards.
>
>
> On 02/09/2015 03:15, SiegeLord wrote:
>
> > Thanks, applied. The best way to send patches these days is through github.
> >
> > I'd prefer to fix the aliasing bugs, rather than disabling the aliasing
> > optimizations, but if that's not possible, it'd be ok by me to add that
> > flag.
> >
> > -SL
> >
> > On 09/01/2015 01:46 PM, Boon947 wrote:
> > > Hello,
> > >
> > > I've posted this patch to the forums, but the discussion seems
> > > to have died down. Is the mailing list a better place to discuss
> > > patches?
> > >
> > > The attached patch fixes a pair of trivial warnings in Allegro 4.4
> > >
> > > I also noticed a warning about an aliasing problem in the OSS code;
> > > with compilers becoming more aggressive at high optimization levels,
> > > it might be prudent to add -fno-strict-aliasing to the CFLAGS, as
> > > Linux has been doing for ages.
> > >
> > > Regards.