| Summary: | [Windows 7 Release Tests] http/tests/xmlhttprequest/basic-auth-nouser.html and basic-auth-nopassword.html failing since introduction in r74609 | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Jessie Berlin <jberlin> |
| Component: | New Bugs | Assignee: | Brent Fulgham <bfulgham> |
| Status: | NEW --- | ||
| Severity: | Normal | CC: | abarth, ap, aroben, bfulgham, eric, jberlin, jhanssen, jhoneycutt, webkit.review.bot |
| Priority: | P2 | Keywords: | InRadar, LayoutTestFailure, PlatformOnly |
| Version: | 528+ (Nightly build) | ||
| Hardware: | All | ||
| OS: | Windows 7 | ||
|
Description
Jessie Berlin
2010-12-24 11:16:56 PST
(In reply to comment #0) > I am not entirely sure why it is not finding those php files, but I am not at all familiar with how we are serving those files. Any ideas? > > I will commit these failing results soon so that the bot is green. The URL is wrong at least, the initial "/xmlhttprequest/" part is what makes this fail most likely, though I'm not sure where that comes from. Perhaps some peculiarity with how DRT's queueLoad() works on Windows? Failing results landed in r74645: http://trac.webkit.org/changeset/74645 There is a FIXME: We should do real relative URL resolution here in LayoutTestControllerWin's queueLoad. Maybe that is the culprit? (In reply to comment #2) > Failing results landed in r74645: > http://trac.webkit.org/changeset/74645 > > There is a FIXME: We should do real relative URL resolution here in LayoutTestControllerWin's queueLoad. Maybe that is the culprit? Looking at the code it certainly looks that way, it doesn't seem to support absolute URL input at all. The thing that gets passed to the WorkQueue to load from LayoutTestControllerWin's queueLoad is "file:///C:/cygwin/home/jessieberlin/WebKit/OpenSource/LayoutTests/http/tests/xmlhttprequest/http://user:@localhost:8000/xmlhttprequest/resources/basic-auth-nouserpass/basic-auth-nouserpass.php" so the issue is probably somewhere in how WorkQueue loads that (because the output makes it appear that it knows enough to strip out the first part but is failing to strip out the /xmlhttprequest/ part of the first part). Right now the only connection to a Windows machine I have is a remote connection, so it will take me a bit to do any further debugging (so painfully slow ...) (In reply to comment #4) > Right now the only connection to a Windows machine I have is a remote connection, so it will take me a bit to do any further debugging (so painfully slow ...) The problem appears to be in LayoutTestController::queueLoad() itself: // FIXME: We should do real relative URL resolution here. int lastSlash = responseURL.rfind('/'); if (lastSlash != -1) responseURL = responseURL.substr(0, lastSlash); wstring wURL = jsStringRefToWString(url); wstring wAbsoluteURL = responseURL + TEXT("/") + wURL; I haven't actually executed this code (No easy access to a Windows box here either at the moment), but since wURL at that point is the URL input to queueLoad(), responseURL will be "/xmlhttprequest/". The code appears to attempt to generate an absolute URL out of a relative one. (In reply to comment #5) > I haven't actually executed this code (No easy access to a Windows box here either at the moment), but since wURL at that point is the URL input to queueLoad(), responseURL will be "/xmlhttprequest/". The code appears to attempt to generate an absolute URL out of a relative one. A possible solution could be to check if the URL is an absolute one (i.e. starts with http://) and then skip this part if that is the case. (In reply to comment #5) > (In reply to comment #4) > > Right now the only connection to a Windows machine I have is a remote connection, so it will take me a bit to do any further debugging (so painfully slow ...) > > The problem appears to be in LayoutTestController::queueLoad() itself: > > // FIXME: We should do real relative URL resolution here. > int lastSlash = responseURL.rfind('/'); > if (lastSlash != -1) > responseURL = responseURL.substr(0, lastSlash); > > wstring wURL = jsStringRefToWString(url); > wstring wAbsoluteURL = responseURL + TEXT("/") + wURL; > > I haven't actually executed this code (No easy access to a Windows box here either at the moment), but since wURL at that point is the URL input to queueLoad(), responseURL will be "/xmlhttprequest/". The code appears to attempt to generate an absolute URL out of a relative one. responseURL is actually "ile:///C:/cygwin/home/jessieberlin/WebKit/OpenSource/LayoutTests/http/tests/xmlhttprequest/" (since it is rfind and this is actually what I saw when stepping through the function). I don't know why it later is truncated to /xmlhttprequest/. I would be fine with changing it to check if the URL is absolute. I will try out that solution when I get the windows machine to respond and see if it breaks anything else. http/tests/xmlhttprequest/basic-auth-nopassword.html has been timing out on the Windows XP Debug test bots for longer than we have records of the layout test results: http://build.webkit.org/old-results/Windows%20XP%20Debug%20(Tests)/r76975%20(24675)/results.html http://build.webkit.org/results/Windows%20XP%20Debug%20(Tests)/r79718%20(25674)/results.html I am going to add it to the Windows XP skipped list soon. (In reply to comment #9) > http/tests/xmlhttprequest/basic-auth-nopassword.html has been timing out on the Windows XP Debug test bots for longer than we have records of the layout test results: > > http://build.webkit.org/old-results/Windows%20XP%20Debug%20(Tests)/r76975%20(24675)/results.html > http://build.webkit.org/results/Windows%20XP%20Debug%20(Tests)/r79718%20(25674)/results.html > > I am going to add it to the Windows XP skipped list soon. I wish we had a debug skipped list. Presumably the timeout is due to Debug builds being slow, not due to something about Windows XP. Did we ever fix the URL-resolution issue? (In reply to comment #11) > Did we ever fix the URL-resolution issue? No, I didn't get around to it yet. Filed a bug about having Debug vs. Release skipped lists: https://bugs.webkit.org/show_bug.cgi?id=55254 Added http/tests/xmlhttprequest/basic-auth-nopassword.html to the win-xp skipped list in http://trac.webkit.org/changeset/79723 http://trac.webkit.org/changeset/79723 might have broken GTK Linux 32-bit Debug and Qt Linux Release The following tests are not passing: fast/table/fixed-with-auto-with-colspan-vertical.html http/tests/xmlhttprequest/basic-auth-nouser.html has been timing out on Windows XP, too: http://build.webkit.org/results/Windows%20XP%20Debug%20(Tests)/r79988%20(25825)/results.html Added http/tests/xmlhttprequest/basic-auth-nouser.html to the Skipped file Committed r79994: <http://trac.webkit.org/changeset/79994> http://trac.webkit.org/changeset/79994 might have broken GTK Linux 32-bit Debug Moved the skipped entries from win-xp to win. ...in r85329. See also bug 52978. This whole directory is flaky on Mac. We have some sort of marauding http-test failure which we don't understand. I've done some investigation in: https://bugs.webkit.org/show_bug.cgi?id=51613 We've also had to skip tests in that directory before: https://bugs.webkit.org/show_bug.cgi?id=32961 I wonder if CFNetwork may have some http-auth oddities. It seems (at present) to be a security settings issue: --- /cygdrive/c/Projects/WebKit/OpenSource/WebKitBuild/Release/bin32/layout-test-results/http/tests/xmlhttprequest/basic-auth-nouser-expected.txt +++ /cygdrive/c/Projects/WebKit/OpenSource/WebKitBuild/Release/bin32/layout-test-results/http/tests/xmlhttprequest/basic-auth-nouser-actual.txt @@ -1,4 +1,4 @@ -Not Found +Forbidden -The requested URL /xmlhttprequest/http://:password@localhost:8000/xmlhttprequest/resources/basic-auth-nouserpass/basic-auth-nouserpass.php was not found on this server. -Apache/1.3.33 Server at 127.0.0.1 Port 8000 +You don't have permission to access /xmlhttprequest/http://:password@localhost:8000/xmlhttprequest/resources/basic-auth-nouserpass/basic-auth-nouserpass.php on this server. +Apache/2.4.10 (Win32) OpenSSL/1.0.1h PHP/5.4.0 Server at 127.0.0.1 Port 80 |