Bug 33242 – Does not send an Accept header for subresources, causing some sites to break

• Status: RESOLVED
• Resolution: FIXED
• Component: Page Loading

[Gustavo Noronha (kov)]

Initially reported in Debian (see http://bgus.debian.org/563736). WebKitGTK+ is unable to display the PNGs used in this site: http://www.inkscapegallery.net/quickstart. I tracked this to be working when the request is done using HTTP/1.0, and when it is done with HTTP/1.1 and with an Accept header. That is why it works if you copy the image address and open it - WebCore sends a default Accept header for main resources.

Safari seems to send an Accept header; Chromium sends Accept: */*, and works. Firefox sends a more specific header saying it prefers image/png, but accepts any image/* type. I guess it decided that because the resource is being loaded from an <img> tag.

[Gustavo Noronha (kov)]

Created attachment 48812 [details]
Send Accept: image/* when obtaining image resources

[Brady Eidson]

Comment on attachment 48812 [details]
Send Accept: image/* when obtaining image resources

r- 

Mucking with Accept headers is a notoriously dangerous thing to do.  

Safari and Chrome send out "valid" accept headers here that they've found to work for them.  WebKitGtk+ alone has the problem, so the fix needs to be targeted there.

[Gustavo Noronha (kov)]

Created attachment 48827 [details]
Make soup backend always send a "valid" Accept header, when one is not explicitly set

Given beidson's comments, I made the change local to our network backend. I tried putting the test in platform/gtk, but it seems like the special-casing of the http tests makes this difficult. If it turns out we need more exceptions like this, I'll take a look at fixing the special casing.

[Xan Lopez]

Comment on attachment 48827 [details]
Make soup backend always send a "valid" Accept header, when one is not explicitly set

If Chromium is doing the same thing then I guess it's a reasonable thing to do for compatibility.

[Gustavo Noronha (kov)]

Thanks! Landed as r54892.

[Tony Chang]

Committed r54941: <http://trac.webkit.org/changeset/54941>

[Tony Chang]

(In reply to comment #6)
> Committed r54941: <http://trac.webkit.org/changeset/54941>

Err, ignore this.  I think webkit-patch got confused.