[Webkit-unassigned] [Bug 42273] Convolution computation causes bad alpha channel values

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=42273





--- Comment #15 from Alex Nicolaou <anicolao at chromium.org>  2010-07-15 00:22:59 PST ---
OK I think this last patch satisfies all comments so far on the bug. Please take another look.

@Dirk - no, the particular complaint I have isn't yet raised against the specification. When I become 100% sure about it I'll try to figure out how to ask someone to add it as an issue. Put as simply as I can phrase it, premultiplied alpha is an optimization, and whether we store the bytes of an image premultiplied or not should not affect the output of a filter like a convolution. However, as the specification currently stands, the same image premultiplied will produce a different output than the non-premultipled version after applying certain convolution filters...it seems to me that if I apply the same filter, and if the image looked the same to begin with, the output should be the same, whether the user agent works with premultiplied image data or not. I am, separately, working on an example to demonstrate the issue.

@Zoltan - yes, I had the same idea as I tried to clean up the code per Eric's request. Have a look at the latest patch.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.