When using libjpeg-turbo and enabling IMAGE_DECODER_DOWN_SAMPLING, JPEG decoding failed because of no support for JCS_EXT_RGBA, JCS_EXT_BGRA. Set RGBA values when color space is JCS_EXT_RGBA or JCS_EXT_BGRA.
Created attachment 167962 [details] Patch
Attachment 167962 [details] did not pass style-queue: Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/WebCore/ChangeLog', u'Source/WebCor..." exit_code: 1 Source/WebCore/ChangeLog:1: ChangeLog entry has no bug number [changelog/bugnumber] [5] Total errors found: 1 in 2 files If any of these errors are false positives, please file a bug against check-webkit-style.
Created attachment 167963 [details] Patch
This looks fine, but I'd like Noel to do an unofficial review before r+'ing it.
Looks sane. Sadly no tests exist for the IMAGE_DECODER_DOWN_SAMPLING enable.
Comment on attachment 167963 [details] Patch That's unfortunate, but I assume having this work is better than not. r=me
Comment on attachment 167963 [details] Patch Clearing flags on attachment: 167963 Committed r131075: <http://trac.webkit.org/changeset/131075>
All reviewed patches have been landed. Closing bug.
Someone asked me a question about this. @jaehun - should the #if test be as follows? #if ENABLE(IMAGE_DECODER_DOWN_SAMPLING) && defined(TURBO_JPEG_RGB_SWIZZLE) Alternatively, we could remove all this and ensure that ports that use IMAGE_DECODER_DOWN_SAMPLING can use libjoeg-turbo correctly. TURBO_JPEG_RGB_SWIZZLE should not be taken to mean libjpeg-turbo and I think we should fix that. Thoughts?
(In reply to comment #9) > Someone asked me a question about this. @jaehun - should the #if test be as follows? > > #if ENABLE(IMAGE_DECODER_DOWN_SAMPLING) && defined(TURBO_JPEG_RGB_SWIZZLE) > I considered the same #if statement at first. I have no objection to change #if statement. > Alternatively, we could remove all this and ensure that ports that use IMAGE_DECODER_DOWN_SAMPLING can use libjoeg-turbo correctly. TURBO_JPEG_RGB_SWIZZLE should not be taken to mean libjpeg-turbo and I think we should fix that. Thoughts? I perfectly agree with you.
Just noticed this fix. Nice job. For those images are not scaled when DOWNSAMPLE is enabled, it seems that we can still use the swizzling path?
(In reply to comment #11) > Just noticed this fix. Nice job. For those images are not scaled when DOWNSAMPLE is enabled, it seems that we can still use the swizzling path? Bug 103749 created for this, and patch is uploaded
I have reviewed that patch, looks good. Second thing to do is to handle the scaled case better per comment 10.