Review: texture race condition fix

Fix a subtle race condition in the texture code: in the bilinear and bicubic sampling code, the case where all samples lie on the same tile wasn't stashing the reference-counted tile pointer in a safe place for the duration of time that it would be potentially reading texel values out of the tile.

It was not symptomatic, and I think it would take a rare circumstance to fail -- basically, the whole tile cache would need to be cycled through between the time it left the block with the ref-counted tile and the time it finished using the pointers to the texels within the tile, thus freeing the tile and invalidating the memory.

Just something I noticed hunting a different bug and thought I should plug the hole.

[ckulla, 2010-12-02 17:02:46]

Are you sure this is needed? In the "one tile" case - won't thread_info->tile
keep pointing to the tile until the same thread does another lookup?

[lg_larrygritz.com, 2010-12-02 17:35:00]

Yes!

Thanks, I didn't think of that.  I won't make this change, then, but I will add
a comment to clarify it for future observers.


On Dec 2, 2010, at 9:02 AM, ckulla@gmail.com wrote:

> Are you sure this is needed? In the "one tile" case - won't
> thread_info->tile keep pointing to the tile until the same thread does
> another lookup?
> 
> http://codereview.appspot.com/3405041/
> _______________________________________________
> Oiio-dev mailing list
> Oiio-dev@lists.openimageio.org
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org

--
Larry Gritz
lg@larrygritz.com