Remove URL decoding in srcset handling
Created attachment 212068 [details] Patch
As pointed out to me by Blink's Christian Biesinger, the URL decoding in the srcset algorithm can break encoded URLs (both in src and in srcset) Since the new parser is in place (which splits the attribute on white spaces rather than commas), data URIs are handled properly, and decoding URLs doesn't really offer any benefits. Therefore, I removed decoding to avoid breaking these URLs.
Thanks for the follow up.
Comment on attachment 212068 [details] Patch Clearing flags on attachment: 212068 Committed r156140: <http://trac.webkit.org/changeset/156140>
All reviewed patches have been landed. Closing bug.
(In reply to comment #4) > (From update of attachment 212068 [details]) > Clearing flags on attachment: 212068 > > Committed r156140: <http://trac.webkit.org/changeset/156140> This change broke development on native windows systems. '?' is not a valid character in filenames on windows. Can you fix the filename (by changing the test to a sever side script which checks for the question mark or any other special character in the URL) or roll out the patch?
(In reply to comment #6) > (In reply to comment #4) > > (From update of attachment 212068 [details] [details]) > > Clearing flags on attachment: 212068 > > > > Committed r156140: <http://trac.webkit.org/changeset/156140> > > This change broke development on native windows systems. '?' is not a valid character in filenames on windows. > Can you fix the filename (by changing the test to a sever side script which checks for the question mark or any other special character in the URL) or roll out the patch? I'm extremely sorry :( Will fix the filename ASAP
Looks like Windows was fixed in <http://trac.webkit.org/changeset/156161>.
And then more in <http://trac.webkit.org/r156179>.
It should be possible to test question marks with a cgi script that intercepts all requests within a directory. We do this in LayoutTests/http/tests/uri/intercept tests.