88209 - Xlib-toolkit Mozilla crashes in ZDnet Web benchmark when running on a remote X11 Xserver...

Status: RESOLVED FIXED

(Core :: XUL, defect)

Description

[Roland Mainz]

An attempt to run the ZDnet browser benchmark
(http://i-bench.zdnet.com/ibench/testlist/home_js.php#pOverallScore) using a
2001-06-27-08-trunk Xlib-toolkit Mozilla crashes in various places. It would be
nice that the Zilla get's fixed to pass this test cleanly...

Comment 1

[Roland Mainz]

Example crash:
-- snip --
WEBSHELL+ = 6
Disabling Quirk StyleSheet
Enabling Quirk StyleSheet
###!!! ASSERTION: You can't draw an image with a 0 width or height!: 'aSWidth >
0 && aDWidth > 0 && aSHeight > 0 && aDHeight > 0', file
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/gfx/src/xlib/nsImageXlib.cpp,
line 440
###!!! Break: at file
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/gfx/src/xlib/nsImageXlib.cpp,
line 440
Document http://i-bench.zdnet.com/ibench/xml/xmlcss_perf_done.html loaded
successfully
Disabling Quirk StyleSheet
/ibench/xml/xmlcss_perf_done.html
/iview/ba010_zdnet_ros468x60iframemain/direct/01?click=http://ads3.zdnet.com/c/g=r290&c=a63862&camp=c21191&idx=2001.06.28.10.00.23/
WEBSHELL- = 5

Program received signal SIGSEGV, Segmentation fault.
0xfdcb2388 in nsWindow::UpdateIdle (data=0x0)
    at
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/widget/src/xlib/nsWindow.cpp:142
142             window->Update();
(gdb) where
#0  0xfdcb2388 in nsWindow::UpdateIdle (data=0x0)
    at
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/widget/src/xlib/nsWindow.cpp:142
#1  0xfdca3b20 in nsAppShell::Run (this=0x174790)
    at
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/widget/src/xlib/nsAppShell.cpp:490
#2  0xfe1bbb70 in nsAppShellService::Run (this=0x151ad8)
    at
../../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/xpfe/appshell/src/nsAppShellService.cpp:418
#3  0x1d720 in main1 (argc=2, argv=0xffbef57c, nativeApp=0x0)
    at
../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/xpfe/bootstrap/nsAppRunner.cpp:1200
#4  0x1e600 in main (argc=2, argv=0xffbef57c)
    at
../../../../../../../home/mozilla/src/2001-06-27-08-trunk/mozilla/xpfe/bootstrap/nsAppRunner.cpp:1504
-- snip --

Comment 2

[Roland Mainz]

Still crashes with 2001-06-29-08-trunk build. Running it locally seems to work -
it only crashes in the remote X11 case... ;-(

Comment 3

[Roland Mainz]

Some test results running Mozilla5 via remote X11:

GTK+ build (2001-06-29-08-trunk build with Sun Workshop 6 Update 2 EA2):
   HTML Load Speed
   All iterations                  2157.72
   First iteration (downloaded)     312
   Subsequent iteration (cached)    263.67

Environment:
host running Mozilla5: Ultra5/333MHz/2MB2ndLcache/384MB
client running Xserver: SPARCstation 5/70MHZ/128MB, Xsun 8bit pseudocolor visual
both machines are connected via 10baseT+1HUB (mainly to compansate the 8bit
visual...). 

ToDO: It would be better to add more latency to the network to simulate a longer
distance between server&client...

Comment 4

[Roland Mainz]

I even see the crash when I use DISPLAY=hostname:0.0 instead of DISPLAY=:0.0 ...
;-(

Comment 5

[Roland Mainz]

Attachment: Patch for 2001-06-29-08-trunk

Comment 6

[Roland Mainz]

Patch seems to work...

Benchmark results:

GTK+ build:
   HTML Load Speed    [Updated Test]                   
   All iterations    1759.96              
   First iteration (downloaded)    238.60              
   Subsequent iteration (cached)    217.34              

Xlib build:
   HTML Load Speed    [Updated Test]                   
   All iterations    1498.33              
   First iteration (downloaded)    201.50              
   Subsequent iteration (cached)    185.26  

(3 cycles... all more or less the same result)
Looks like GTK+-tookit Mozilla5 is far slower than Xlib-toolkit Mozilla5... :-)

----

Reassining bug to myself, requesting r= for that patch...

Comment 7

[Stuart Parmenter]

I'd love to see local performance tests.

Comment 8

[Roland Mainz]

pavlov wrote:
> I'd love to see local performance tests.

Xlib-toolkit does not have support for MIT-SHM yet.
Comparing 'GTK+-toolkit with MIT-SHM-"accerlation"' vs. 'plain Xlib-toolkit'
wouldn't be fair, right ?

Comment 9

[tor]

r=tor

Comment 10

[Tomi Leppikangas]

Patch looks good, copy from gtk. Maybe change "tmp_list" to "item" 

r=[e-mail]

Comment 11

[tor]

[s]r=tor

Comment 12

[Roland Mainz]

CC:'ing [e-mail] for checkin, please...

Comment 13

[Mike Kaply [:mkaply]]

Fix checked in

Comment 14

[John Morrison]

May God have mercy on us all. The 212 bug spam-o-rama is Now!

Comment 15

[jag (Peter Annema)]

*** Bug 79937 has been marked as a duplicate of this bug. ***