76604 - mozilla + gcc3 + -O2 == crash on startup

Status: VERIFIED WORKSFORME

(SeaMonkey :: Build Config, defect, P1)

Description

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

Something changed in the gcc3 branch between 20010309 (I think) and the 20010326
snapshot that causes mozilla to crash on startup when compiled with -O2.  I've
seen the problem in the weekly snapshots dated 20010326, 20010402, 20010409, and
20010416.  It works fine when compiled without optimization, and I haven't
tested -O.

When I compile optimized non-debug, the crash is in the CheckForBrowser function
in layout/html/document/src/nsFrameFrame.cpp (and was there across a number of
different gcc versions and mozilla tree pulls).  When I compile optimized debug,
the crash is here (notice something is messed up with line numbers):

#4  <signal handler called>
#5  0xfffdd78f in ?? ()
#6  0x40df369c in imgRequest::OnDataAvailable(imgIRequest*, nsISupports*,
gfxIImageFrame*, nsRect const*) (this=0x8404170, request=0x0, cx=0x0, 
    frame=0x848d298, rect=0xbfffea98)
    at ../../../dist/include/xpcom/nsISupportsUtils.h:1241
#7  0x40e2344a in HaveDecodedRow(void*, unsigned char*, unsigned char*, int,
int, int, int, unsigned char, int) (aClientData=0x844f970, 
    aRowBufPtr=0x8474790 "", aRGBrowBufPtr=0x8252e80 "", aXOffset=0, 
    aLength=11, aRowNumber=0, aDuplicateCount=1, aDrawMode=4, aInterlacePass=1)
    at ../../../../dist/include/xpcom/nsCOMPtr.h:647

I really don't know if this is a Mozilla bug or a gcc bug.  It could be that the
gcc folks turned on some optimization that breaks some incorrect code in Mozilla,
or it could be a gcc bug...

Somebody else please take this bug. :-)

Comment 1

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

The patch I will attach moves the crash to some other function.  printfs in
other parts of CheckForBrowser don't have the same effect.  It prints "(nil)"
and "(nil)" for the 2 values.

Comment 2

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

Attachment: patch that moves crash elsewhere

Comment 3

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

Problem does not occur with 3-12 snapshot.

Comment 4

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

I just filed a bug report in GNATS, and I'm currently building the May 21
snapshot.  This should have been my top priority a while back... and I'll try to
make it that when I have some more time next week.  P1/0.9.2

Comment 5

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

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=2917

Comment 6

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

Attachment: build errors with May 21 snapshot

Comment 7

[Dan Mosedale (:dmosedale, :dmose)]

It might be interesting to try compiling -O2 -ggdb, and see what sort of
backtrace information you get from that.

Comment 8

[cls]

With a gcc 3.0 cvs pull from last night, gcc3 -O2 builds seem to be working now.

Comment 9

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

Yeah, I discovered the same thing yesterday, but forgot to comment on the bug.

Comment 10

[Jon Granrose]

verified.