Re: [AD] Bug in unstable developers release, sound.c

[Eric Botcazou <ebotcazou@xxxxxxxxxx>]

> I have come across a bug in the unstable developers release stream (at
> least 4.1.4-4.1.9) in the file src/sound.c.

Note that Chris Jones changed the type of 's' from 'signed short' to 'int' in 
the 4.1.5 release.

> In the function 'load_wav' (line 951 of the 4.1.9 source), a signed short
> is compared against EOF, and will prematurely stop if it encounters valid
> data of -1. This only occurs on 16 bit data.

This bug did exist in the 4.1.4 release but was fixed in the 4.1.5 by Chris.

> Also, in line 944, the result of a 'pack_fread' is compared to the wrong
> variable, though it would only matter on a truncated stereo 8 bit wav file.

Confirmed, my fault. Thanks for spotting it!

> The stable releases are unaffected

Yes, thanks for checking.

> --- 940,956 ----
>   	 spl = create_sample(bits, ((channels == 2) ? TRUE : FALSE), freq,
> len);
>
>   	 if (spl) {
> ! 	    if (pack_fread(spl->data, length, f) < length) {
>   		  destroy_sample(spl);
>   		  spl = NULL;
> ! 		  goto getout;
>   	    }
> ! 	    if(bits == 16) {
> ! 		p = (unsigned short *)spl->data;
> ! 		i = length / 2;
> ! 		while(i--)
> ! 			*p++ ^= 0x8000;
>   	    }
>   	    length = 0;
>   	 }
>         }

The 16-bit part doesn't work for big-endian CPUs, which we try to support as 
much as possible. 16-bit WAV files are stored in little-endian format (Intel 
format, hence the use of pack_igetw) while Allegro's internal buffers are 
supposed to be stored in native format.

-- 
Eric Botcazou