Re: Bugreport from mozilla bug database

At 11:21 AM 7/7/2004 -0400, Glenn Randers-Pehrson wrote:
>At 12:06 AM 7/7/2004 +0200, Gerard Juyn wrote:
>>Glenn,
>>
>>There's a "smart" decision in the glue-code to use a different canvas-style (to 
>>reduce memory-overhead) which is based on libmng prediction of an alpha-
>>channel being present or not. These problems sound like there's something 
>>fishy in that area.
>> 
>
>Thanks for the hint.  That was it.  The mistake is in libmng, however;
>it is returning "iAlphadepth==0" instead of "1" for standalone PNG files
>even when there is "cheap" binary transparency conveyed by the tRNS chunk.
>I'll check in the fix today.

The fix is checked in.  Mozilla now produces the correct results for all
of jason's test images under MNG_BUILD_RAW_MNG and MNG_BUILD_FULL_MNG
configurations.  Under the MNG_BUILD_MOZ_MNG configuration, tests
T9, T10, G7, and G8 fail, but this cannot be helped because the
16-bit transparency information is deliberately thrown away.

Glenn


--
Send the message body "help" to [email protected]