835904 - test_prompt.html fails on mozilla-central

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[(no longer active)]

We saw this over in bug 806737 for the first time.  Unfortunately, this seems to be a real regression in the HTTP implementation.  If you run toolkit/components/passwordmgr/test/test_prompt.html, you'll get the following js error:

 1:08.18 JavaScript error: file:///Users/ehsanakhgari/moz/pb/obj-ff-dbg/dist/NightlyDebug.app/Contents/MacOS/components/nsLoginManagerPrompter.js, line 577: NS_ERROR_XPC_JS_THREW_STRING: 'proxy auth needs nsIProxyInfo' when calling method: [nsIPromptService2::promptAuth]
 1:08.18 JavaScript error: file:///Users/ehsanakhgari/moz/pb/obj-ff-dbg/dist/NightlyDebug.app/Contents/MacOS/components/nsLoginManagerPrompter.js, line 577: NS_ERROR_XPC_JS_THREW_STRING: 'proxy auth needs nsIProxyInfo' when calling method: [nsIPromptService2::promptAuth]

Patrick, can you please look into this?  The first step would be to bisect this, I suppose (but that's a bit harder than it should be since test_prompt.html has to be manually enabled for each bisection step.)

Comment 1

[Patrick McManus [:mcmanus]]

just a comment to say this is on my radar but I'm going into a vacation period for the better part of a week..

Comment 2

[(no longer active)]

Thanks, Patrick!  Assigning it to you then.  I'll try to find some time today to help with the bisection...

Comment 3

[(no longer active)]

OK, I tried bisecting this, and in every step of the bisection, the error was reproducible.  Then I looked more closely, and suddenly it occured to me what's going on with the test.  This test ends up calling SimpleTest.finish() too soon.  You can observe it by running the test stand-alone, and seeing the green table pop up before the JS error occurs.  This causes the test failure to be masked away when we run all of the tests in the directory (like we do on tinderbox machines) since finish() will be called and we will move on to the next test before we hit the failing code.  In other words, a large portion of this test has been getting skipped for quite some time, and it's not easy to say when this exactly broke.  :(

Comment 4

[Andres Hernandez [:andreshm]]

That's what I was trying to explain on bug 806737 comment 5 :)

Comment 5

[(no longer active)]

(In reply to comment #4)
> That's what I was trying to explain on bug 806737 comment 5 :)

Right, yeah I missed that.  Sorry!

Comment 6

[Patrick McManus [:mcmanus]]

The issue is what I poorly described over in:

https://bugzilla.mozilla.org/show_bug.cgi?id=806737#c10

proxy info is determined asynchronously now (before it was causing jank when done synchronously during channel creation). test_prompt.html simply creates the channel and then queries the PI.. this is going to fail.

The solution is just to wait for something to happen on the channel (observe on-modify-request for example) to signal that the PI is ready.

Comment 7

[Patrick McManus [:mcmanus]]

Attachment: patch 0

Comment 8

[Patrick McManus [:mcmanus]]

Ehsan, Andres -

my patch should fix up the proxy auth cases (508, 509, 510).

The test still fails for me near the end in test 1003. But that has something to do with popupNotifications which I really know nothing about - can you find a new bug and owner for that? I'm quite sure its unrelated.

Comment 9

[Patrick McManus [:mcmanus]]

  https://hg.mozilla.org/integration/mozilla-inbound/rev/d0c699e76236

Comment 10

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/d0c699e76236

Comment 11

[(no longer active)]

Thanks Patrick!