Bug 66929 – [Qt] REGRESSION(93767): Two tests fail with newer pywebsocket

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Fehér Zsolt]

I am skipping the tests for now.

[Fehér Zsolt]

The tests were skipped in http://trac.webkit.org/changeset/93775.

[Csaba Osztrogonác]

Any idea what happened?

--- /ramdisk/qt-linux-release/build/layout-test-results/http/tests/websocket/tests/hixie76/close-before-open-expected.txt 
+++ /ramdisk/qt-linux-release/build/layout-test-results/http/tests/websocket/tests/hixie76/close-before-open-actual.txt 
@@ -1,4 +1,5 @@
 CONSOLE MESSAGE: line 0: WebSocket is closed before the connection is established.
+CONSOLE MESSAGE: line 0: Unexpected response code: 403
 Test if Web Socket fires close event when WebSocket is opened and closed fore open event is received.
 
 On success, you will see a series of "PASS" messages, followed by "TEST COMPLETE".


--- /ramdisk/qt-linux-release/build/layout-test-results/http/tests/websocket/tests/hixie76/close-unref-websocket-expected.txt 
+++ /ramdisk/qt-linux-release/build/layout-test-results/http/tests/websocket/tests/hixie76/close-unref-websocket-actual.txt 
@@ -1,4 +1,5 @@
 CONSOLE MESSAGE: line 0: WebSocket is closed before the connection is established.
+CONSOLE MESSAGE: line 0: Unexpected response code: 403
 Test if Web Socket is closed while handshaking and unreferenced, it should fire close event at most once.
 
 On success, you will see a series of "PASS" messages, followed by "TEST COMPLETE".

[Yuta Kitamura]

I rolled out r93767, which should fix these failures.

[Yuta Kitamura]

r93767 caused pywebsocket server to generate some suspicious logs on bots including Qt and SnowLeopard, thus it was rolled out. See https://bugs.webkit.org/show_bug.cgi?id=66924#c7 for details.

I think we can unskip these two tests now.

[Takashi Toyoshima]

Created attachment 105301 [details]
Patch

[Takashi Toyoshima]

Previous change works on Chromium and Apple ports.
Qt port seems to have a problem that close before open could not raise close event.
I'll check it later.

[Takashi Toyoshima]

Created attachment 105318 [details]
Patch

[Takashi Toyoshima]

A Qt port issue I commented previously seems to be my misunderstanding.

The second patch work at Chrome, Apple and Qt ports.
I'm happy if anyone review and land it.

Thanks.

[Takashi Toyoshima]

Created attachment 105361 [details]
Patch (update ChangeLog)

[Takashi Toyoshima]

Kent-san,
Could you review this change?

This issue block some changes to support WebSocket::close's code and reason.

[Kent Tamura]

Comment on attachment 105361 [details]
Patch (update ChangeLog)

View in context: https://bugs.webkit.org/attachment.cgi?id=105361&action=review

Need more explanations in ChangeLog.

> LayoutTests/ChangeLog:3
> +        Two tests fail with newer pywebsocket.

Please write what problem is solved, or what we change concretely.

> LayoutTests/ChangeLog:10
> +        close-unref-websocket tests call close() before GC. But, close()
> +        itself queues close event. These close() must not be called to test
> +        GC causes close event.

Why does this cause a problem on Qt + new pywebsocket, and not on other platforms?

> LayoutTests/ChangeLog:13
> +        its implementation is rough and results in unstable behavior.

Could you explain how rough it is, why does it make tests unstable, and why has Qt a problem please?

[Takashi Toyoshima]

Created attachment 105578 [details]
Patch

[Kent Tamura]

Comment on attachment 105578 [details]
Patch

Looks reasonable.

[WebKit Review Bot]

Comment on attachment 105578 [details]
Patch

Clearing flags on attachment: 105578

Committed r94052: <http://trac.webkit.org/changeset/94052>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Csaba Osztrogonác]

Reni, could you unskip the tests if this patch really fixed it?