232721 - firebird crashes at http://www.ez74.com

Status: VERIFIED DUPLICATE of bug 233399

(Firefox :: General, defect)

Description

[Jungshik Shin]

How to reproduce:

1. go to http://www.ez74.com
2. set the character coding to EUC-KR (in View | Character Coding)
3. firebird crashes

Mozilla doesn't crash at the same site. Both firebird 0.7 and nightly crashes on
Windows. I have yet to try it on Linux. Mozilla on Linux doesn't crash, either.

Comment 1

[Isaac Hwak Han]

Confirming crash with a clean profile on lastest trunk nightly build
(20040130/WinXP). The crash only happens in non-debug build, so I can't upload
strace.

QA to self and minimized testcase coming...

Comment 2

[Isaac Hwak Han]

Attachment: Testcase (Crash!)

The content of testcase is:

<style>-가</style>

Comment 3

[Jungshik Shin]

With your testcase, firebird 0.7 doesn't crash. Instead, it goes into an
infinite loop. Actually, it seems like only one thread goes into the loop while
the UI thread is still responsive. So I can quit it, but according to the task
manager firebird is still running. When I launch it again, firebird wouldn't use
the profile I used for the previous instance saying the profile is in use. 

On Linux, fb 0.7 doesn't crash. Neither does 0.8nightly I downloaded from
ftp.mozilla.org. It was built with 'disable-debug enable-optimize=-Os -g'. 

If a debug build doesn't crash while opt. build does, it's nasty... Can you
build an optimized debug build on Windows and see if you can get stack trace?
 

Comment 4

[Jungshik Shin]

The external stylesheet used at the url has an invalid (html-style comment) with
a Korean string. It's <http://www.ez74.com/nam.css>. With the invalid comment
removed , firebird doesn't creash when accesing a local copy. 

What I can't understand is why firebird crashes while Mozilla doesn't. bz, do
you know of any difference between firebird and mozilla in stylesheet handling? 

Comment 5

[Boris Zbarsky [:bzbarsky]]

Not that I'm aware of, no.  But who knows what all they've changed....

Comment 6

[Isaac Hwak Han]

I tried optimized debug build (--enable-optimize --enable-debug), but FB refuses
to start. It shows several MSCRT Debug Assertion dialogs complaining debug heap
was damaged or not valid. When I ignore the dialogs, FB crashes. (Well, it may
be a different bug or that build option combo may be unsupported.)

I couldn't have any opportunity to load the testcase or URL, so no useful info
here. I agree this bug is getting quite nasty.

Comment 7

[Jungshik Shin]

Sorry I was lazy (I had known how to build an optimized build with debug symbols
on Windows, but I couldn't remember when I wrote the previous comment and  just
assumed that you'd do the 'right thing' :-)).  I should have referred to the
Win32 debugging FAQ:

http://www.mozilla.org/build/win32-debugging-faq.html#opt

You have to set MOZ_PROFILE to 1 before building an opt. build. 

Comment 8

[Jungshik Shin]

It's becoming more interesting. My optimized debug build with the fresh source
gree off the cvs doesn't crash. The latest nightly downloaded from
ftp.mozilla.org still crashes (2004-01-31). I compared about:buildconfig of two
builds. Notable differences between two are:

 1. build target: it is i686-pc-cygwin in mine and it is i586-pc-msvc in the
nightly. However, I guess this doesn't matter.
 2. configure arguments: the nightly has '--disable-profile-sharing
--enable-static --disable-shared' while mine doesn't. 
 3. set of enabled extensions:  inspector is enabled only in the nightly. others
are common in both builds.
 4. mine was build with MOZ_PROFILE set to 1 while the nightly was not (judging
from it doesn't have any debug symbol) 

What's suspicious is 'enable-static disable-shared'. Mozilla isn't build with
either of them. It's not very likely that they are to blame, but I can't find
any other culprit. Anyway, I wonder why they're used for firebird. 

Comment 9

[Boris Zbarsky [:bzbarsky]]

> Anyway, I wonder why they're used for firebird.

Because they make startup a fraction of a percent faster.

Comment 10

[Isaac Hwak Han]

Thanks for your comment. My optimized build with MOZ_PROFILE=1 didn't crash either.

As for static vs. dynamic build issue, I don't think this is real culprit. See
following test (all of these are static build):

030516 (official 0.6)            ok
030612 (aebrahim with -Ox)       ok
030702 (aebrahim with -Ox)       CRASH
031007 (official 0.7)            CRASH
040128 (aebrahim with -Oxs)      CRASH
040130 (official nightly)        CRASH
040130 (aebrahim with -O2)       ok
040131 (aebrahim with -O1 -Oi)   CRASH

Official FTP has no old nightly, so I tested with aebrahim's builds. (from
http://pryan.org/firebird/aebrahim ). I couldn't narrow down the regression
window since there is no binary between 030612 and 030702.

Interestingly, -O2 build doesn't crash. (There are other problems though)

Comment 11

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

*** This bug has been marked as a duplicate of 233399 ***