I've noticed that a page http://complex.pl fails to open inside Safari 3.0.3 for Windows and the latest Webkit r25282. Steps to reproduce: 1. Go to a page: http://complex.pl 2. Wait a couple of seconds. 3. An error message appears: "Safari can’t open the page. Too many redirects occurred trying to open “http://complex.pl/index.php?sid=1bead20a9cdb4133caf2e175fb0cb49e&clear=1&go=17&bMenu2=1&menu2=231&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498?sid=1bead20a9cdb4133caf2e175fb0cb49e&go=17&id_k=498”. This might occur if you open a page that is redirected to open another page which then is redirected to open the original page." Expected result: Safari should open the site. Current result: Safari fails to open the site. Other tested browsers: Opera 9, Firefox 2 and Internet Explorer 7 have no problem with the site.
Confirmed on Mac OS X; not a regression. The server sends a Location response header that only has a query part and expects it to replace the existing query, but WebKit appends it to URL instead of replacing. Location: ?sid=ad062e33b1831d1cd31c62520b524b9&go=17&id_k=498
This looks like an underlying closed source framework bug to me: -[WebCoreResourceHandleAsDelegate connection:willSendRequest:redirectResponse:] already receives an incorrect new request. DOM manipulations such as window.location="?1" appear to work correctly. I haven't investigated what happens in the Windows port, but I suppose it's similar.
<rdar://problem/5451301>
Web developers of complex.pl fixed the problem on their side. So the described issue is no longer able to reproduce on reported site, therefore I remove the URL.
FWIW, I get the same problem on http://www.sjscycles.co.uk with WebKitGTK+ 1.2.6, but it's working fine with svn trunk@77060 268f45cc-cd09-0410-ab3c-d52691b4dbfc so I guess it's fixed.
The Gtk issue is definitely different, as the original bug was not even in open source WebKit. Marking resolved now, as this has been fixed server-side on http://complex.pl.