Bug 82484 – [GTK] 3 fast/ layout tests failing after upgrading libsoup to 2.37.92

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Sergio Villar Senin]

These are the tests that timeout.

fast/encoding/percent-escaping.html
fast/loader/file-URL-with-port-number.html
fast/loader/location-port.html

I'm skipping them for the moment as it seems it's a bug in libsoup. The culprit seems to be this commit 38aab0e2b.

[Dan Winship]

fast/loader/file-URL-with-port-number.html passes for me

[Dan Winship]

Created attachment 134692 [details]
libsoup patch to fix fast/encoding/percent-escaping

Can you try this patch against libsoup? It definitely fixes the percent-escaping test, but I seem to get lots of other differences in the test output (mostly in -stderr.txt files) as well, but I can't be sure if that's because of an actual behavior change, or just random trying-to-run-tests-on-a-non-blessed-machine flakiness.

[Sergio Villar Senin]

(In reply to comment #2)
> Created an attachment (id=134692) [details]
> libsoup patch to fix fast/encoding/percent-escaping
> 
> Can you try this patch against libsoup? It definitely fixes the percent-escaping test, but I seem to get lots of other differences in the test output (mostly in -stderr.txt files) as well, but I can't be sure if that's because of an actual behavior change, or just random trying-to-run-tests-on-a-non-blessed-machine flakiness.

It fixes that one indeed but the other two are still failing.

Wouldn't it be simpler just to back to soup_uri_decode()/g_file_new_for_path() and put the windows code inside some guards? Yeah I know it isn't very beautiful and that libsoup code does not have things like that, but if the glib implementations behave in a different way ... I mention that because I asked the author of the commit and he kind of remembers some issues with slashes when using g_file_new_for_path(), that's why he switched to g_file_new_for_uri().

[Dan Winship]

(In reply to comment #3)
> It fixes that one indeed but the other two are still failing.

Ah, I thought fast/loader/file-URL-with-port-number.html was passing, because it shows "SUCCESS" if you run it in GtkLauncher, but that's just because the test isn't distinguishing the requested resource from an error page...

However, fast/loader/location-port.html is completely unrelated; it fails with any version of libsoup. Perhaps it's because of something that changed in ResourceHandleSoup.cpp rather than something that changed in libsoup?

[Dan Winship]

fast/encoding/percent-escaping.html and fast/loader/file-URL-with-port-number.html are now fixed with the latest libsoup, and the fix will also be in 2.38.1. I don't know if it's worth bumping the requirements over. (If we want to un-skip the tests we could just update the packages on the bots without changing the required minimum version.)

[Martin Robinson]

(In reply to comment #5)
> fast/encoding/percent-escaping.html and fast/loader/file-URL-with-port-number.html are now fixed with the latest libsoup, and the fix will also be in 2.38.1. I don't know if it's worth bumping the requirements over. (If we want to un-skip the tests we could just update the packages on the bots without changing the required minimum version.)

We can just update the version in the WebKit jhbuild and unskip.

[Dominik Röttsches (drott)]

Same failures in EFL of course, I'll take a look.

[Dominik Röttsches (drott)]

Created attachment 137890 [details]
Bumping libsoup to 2.38.1

I tested the previously failing tests that were moved to the skip list as well as http/*

On my machine, for GTK I got a couple of unexpected passes afterwards. Maybe Philippe can unskip them in a separate commit if that's verified on the buildbot? I figure these are related to bug 52798.

Test / Expected Failure
http/tests/loading/preload-append-scan.php	fail
http/tests/xmlhttprequest/redirect-cross-origin-sync.html	timeout
http/tests/xmlhttprequest/re-login-async.html	fail
http/tests/xmlhttprequest/redirect-cross-origin-post-sync.html	timeout
http/tests/xmlhttprequest/redirect-cross-origin-sync-double.html	timeout
http/tests/xmlhttprequest/basic-auth-default.html	fail
http/tests/xmlhttprequest/cross-origin-authorization.html	fail
http/tests/xmlhttprequest/xmlhttprequest-contenttype-empty.html	fail
http/tests/xmlhttprequest/simple-cross-origin-progress-events.html	timeout
http/tests/media/video-cross-site.html	fail
http/tests/workers/text-encoding.html	fail
http/tests/cache/subresource-fragment-identifier.html	fail
http/tests/misc/dns-prefetch-control.html	fail
http/tests/misc/iframe-reparenting-id-collision.html	fail
http/tests/misc/favicon-loads-with-images-disabled.html	fail
http/tests/appcache/crash-when-navigating-away-then-back.html	fail
http/tests/appcache/auth.html	fail
http/tests/appcache/interrupted-update.html	fail
http/tests/incremental/split-hex-entities.pl	timeout
http/tests/security/cross-frame-access-object-getPrototypeOf.html	fail
http/tests/security/local-CSS-from-remote.html	fail
http/tests/security/local-iFrame-from-remote.html	fail
http/tests/security/local-user-CSS-from-remote.html	fail
http/tests/security/isolatedWorld/context-destroy.html	fail
http/tests/security/xssAuditor/xss-filter-bypass-sjis.html	fail
http/tests/security/xssAuditor/xss-filter-bypass-big5.html	fail
http/tests/security/local-image-from-remote.html	fail
http/tests/security/local-JavaScript-from-remote.html	fail
http/tests/security/drag-drop-same-unique-origin.html	fail

[WebKit Review Bot]

Comment on attachment 137890 [details]
Bumping libsoup to 2.38.1

Clearing flags on attachment: 137890

Committed r114653: <http://trac.webkit.org/changeset/114653>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Raphael Kubo da Costa (:rakuco)]

Please remember to raise the libsoup versions in configure.ac and Source/cmake/OptionsEfl.cmake.

[Martin Robinson]

(In reply to comment #11)
> Please remember to raise the libsoup versions in configure.ac and Source/cmake/OptionsEfl.cmake.

This isn't actually a requirement in this case. The libsoup version in configure.ac is the necessary API level, whereas the version in jhbuild is what's necessary to pass the tests.