Cannot reproduce on r48999 Linux x64 SVN

Can you provide a .blend and maybe two output files that show the flicker?

My bad Gottfried. Should have done a little due diligence.

Summary:

It appears the background of the stamp (and likely the foreground?) is not being correctly linearized / de-linearized in the VSE. This can be illustrated by either A) selecting a color for the stamp in the VSE and looking at the result, or B) using alpha overs to illustrate the error.

The attached sample uses a simple background and a simple foreground with alpha. The alpha over image is set to "Convert Float", which, while not linearizing the values, is somehow routing the correct way through the rendering pipeline. The background, which is not converted to float (and is in the uchar format), is rendering the stamp colors incorrectly.

I suspect it has to do with the UChar / Float paths in the Stamp feature?

Looked at the code, and the buf_rectfill_area does a blind fill of a given color (Grr on the blind hard coded color assumption!) so it doesn't pertain to this issue.

From what I can guess it is probably the "Convert Float" feature of the VSE. Guessing, but my hunch is that the rect_float assumption as linear space that the VSE violates (it is actually merely a value/255 sRGB conversion) is somehow bogging up the rendering of the stamp, which is likely expecting correctly linearized color values.

Only the combination of "premultiply" and "convert float" produces the issue here. Doen't happen if only one of the options is set.

Btw: When rendering the file here, no flicker occurs. Just a change in colors for the first "stamp-bug.png"-strip.

Bug does not occur when you turn of color management btw...

"Btw: When rendering the file here, no flicker occurs. Just a change in colors for the first "stamp-bug.png"-strip."

That's the flicker. I originally had somewhat misdiagnosed it so my rapid series of strips caused something akin to a flicker.

I believe Sergey is working on gutting the rendering backend on the VSE to fit OCIO under the hood, which should fix this.

"Bug does not occur when you turn of color management btw..."

This would be because the linearization code paths are skipped. Soon this toggle should be removed altogether anyways.

fixed r49860, font color had the same issue too which is now fixed.