"Premul" alpha problem? ( gcc4.0.1 vs gcc4.2.1 )

[edit: building blender with gcc4.2.1 and…]

i started running into some strange errors:
http://lh5.google.com/treatkor/R_PZ_50i6fI/AAAAAAAABAw/L00WUFS2yDA/s288/smc_lightbulb12_14307.png

i’m trying to figure how to fix these alpha “holes” in my renders. they seem to be happening at the edges of render parts (xparts, yparts) when i have the “Premul” option clicked in the Scene Buttons > Render tab.

so i did a test using the same blend file in several different builds of blender.
and got these results: (click image to see larger and corresponding build info)
http://treatkor.googlepages.com/testresults.png

i just opened up this blend file : smc_lightbulb12.blend in different builds and pressed F12.

the blend file was created in a recent SVN build, so maybe that is the reason for the differences between some builds. (but the differences isn’t really my issue; my issue is how can i use “Premul” and not get these holes)

i discovered that when i render with “Sky” instead of “Premul” the problem goes away. (i only checked this in rev 14307)

btw, i’m using a macbook 2Ghz Core2Duo with 2G ram

:

ok, so what I know is that 1. each new blender version adds new stuff to the render options and pipeline, so different versions MAY render the same scene differently based on the option set in effect and the effect that the default settings have.
2. graphicall SVN builds are EXPERIMENTAL. You might be able to drive yourself nuts trying to figure out why an experiment fails. For sanity, you should stick tot he SVN build if you want to investigate an error or discrepancy or test the upcoming release for regression.

  1. I’m not seeing any bugs in the 2.45. In the 2.46RC1, with Sky, the bulb is dark because you have translucency turned off. I would think that the inside backside of the3 glass of the bulb would reflect some light that is coming in from the frontside. Setting Tralu to .4 looks about “real” to me, assuming I was looking at a smoky bulb.

Attachments


PapaSmurf, thanks for the reply. i should have mentioned that these are my own builds. and more recently i’d been using gcc42 as compiler, so i did a compile of most recent svn revision with gcc4.0.1 and the problem went away. so i’m almost certain it is the optimization flags i’m using. i’m going to change (remove) the optimization flags i’ve been using with gcc42 and see what the result is.
also thanks for the material settings tips!

edit: ok, so i compilied latest svn with gcc42 and NO optimization flags, and i still get this problem.