Re: [Chromium-users] Compressed textures

[Martins D I. <mi...@cc...>]

>>>>>>>> Hi,
>>>>>>>>      I get a segfault when viewing an OpenSceneGraph model with
>>>>>>>> compressed textures through tilesort.  It works fine with 
>>>>>>>> uncompressed
>>>>>>>> textures.  Stack trace attached.
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> I don't see the attachment.
>>>>>>> 
>>>>>>> I just tried a Mesa texture compression test and it seemed to work,
>>>>>>> with a little hack.  I had to disable the check for the extension
>>>>>>> GL_EXT_texture_compression_s3tc since it's not fully supported in
>>>>>>> Chromium.  It works to the extent that you can call
>>>>>>> glCompressedTexImage2DARB() with a pre-compressed image.  But Chromium
>>>>>>> doesn't have the S3TC compression/decompression code needed for some
>>>>>>> of other paths.
>>>>>>> 
>>>>>>> -Brian
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Sorry, here's the trace.  The textures are precompressed so looks like 
>>>>>> it
>>>>>> should work based on what you say above.
>>>>>> 
>>>>>> Doesn't seem like it failing on the extension check, as its actually
>>>>>> making the glCompressedTexImage2D call.
>>>>>> 
>>>>>> I don't fully understand the code path, but seems like server_index is 
>>>>>> not
>>>>>> being set.
>>>>> 
>>>>> 
>>>>> 
>>>>> Which version of Chromium are you using?  The line numbers in the
>>>>> stack trace don't match current CVS.  You might try the 1.9 release
>>>>> candidate at http://chromium.sourceforge.net/beta/
>>>> 
>>>> 
>>>> 
>>>> I was using 1.9-rc1.  Tried cvs from today with the same result.  Trace 
>>>> attached.
>>> 
>>> 
>>> OK, I reproduced the problem here.  There's two things to try:
>>> 
>>> 1. Disable the assertion.  I think it may be bogus.
>>> 2. Increase your MTU in your .conf file so it can hold any compressed tex 
>>> image.
>> 
>> 
>> Increasing the MTU (1024*1300) made it not crash, but that caused there to 
>> be untextured polygons, as well as textures applied to the incorrect 
>> polygons.
>> 
>> OR
>> 
>> Disabling the assertion caused problems further along. See attached.
>
> Hmmm.
>
>
>> Doesn't "server_index=-1" cause "buffer" to be invalid at line 80 of 
>> tilesortspu_flush.c ?
>
> server_index is always >= 0 for me.  I checked in some new assertions to be 
> sure.  I guess we're hitting slightly different problems.

OK. Yeah, server_index is -1 for me.

>
> What's involved in running your OpenSceneGraph program?  Could I get a copy?
>

Just running the model through osgviewer shows the problem for me.  I'll 
send you a copy of the model off-list.  I'm running OSG 1.0 on x86_64 if 
that makes a difference.

Martins