130272 - segfault/critical errors while starting 0.9.9

Status: VERIFIED INVALID

(SeaMonkey :: General, defect)

Description

[Wouter Van Hemel]

0.9.9 compiled from SRPM crashes on startup. Compilation itself was ok (I have
been compiling them since 0.9.1 or so).

Errors shown:
[cocaine:/opt/gnome/lib/mozilla] run-mozilla.sh 
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/libgfx_gtk.so) Load FAILED
with error: /opt/gnome/lib/mozilla/components/libgfx_gtk.so: undefined symbol:
GetGlobalPrintSettings__14nsPrintOptionsPP16nsIPrintSettings
**************************************************
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/libgfxps.so) Load FAILED with
error: /opt/gnome/lib/mozilla/components/libgfxps.so: undefined symbol:
AbortDocument__17DeviceContextImpl
**************************************************
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/liblocalmail.so) Load FAILED
with error: /opt/gnome/lib/mozilla/components/liblocalmail.so: undefined symbol:
OpenNetworkSocketWithInfo__13nsMsgProtocolPCciT1P12nsIProxyInfoP21nsIInterfaceRequestor
**************************************************
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/libmsgcompose.so) Load FAILED
with error: /opt/gnome/lib/mozilla/components/libmsgcompose.so: undefined
symbol: GetSpec__16nsMsgMailNewsUrlR10nsACString
**************************************************
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/libmsgimap.so) Load FAILED
with error: /opt/gnome/lib/mozilla/components/libmsgimap.so: undefined symbol:
GetPerformingBiff__19nsMsgIncomingServerPi
**************************************************
**************************************************
nsNativeComponentLoader:
SelfRegisterDll(/opt/gnome/lib/mozilla/components/libmsgnews.so) Load FAILED
with error: /opt/gnome/lib/mozilla/components/libmsgnews.so: undefined symbol:
OpenNetworkSocketWithInfo__13nsMsgProtocolPCciT1P12nsIProxyInfoP21nsIInterfaceRequestor
**************************************************
[cocaine:/opt/gnome/lib/mozilla] _

Are these SRPMS (ftp, releases/0.9.9/SRPMS/) working? This is the first time i
got an error.

Comment 1

[Wouter Van Hemel]

regxpcom crashes, during installation of rpms

/opt/gnome/lib/mozilla/rebuild-databases.sh: line 17: 29701 Segmentation fault
    MOZILLA_FIVE_HOME=/opt/gnome/lib/mozilla /opt/gnome/lib/mozilla/regxpcom
>/dev/null 2>/dev/null

--
(gdb) r
Starting program: /opt/gnome/lib/mozilla/regxpcom 
(no debugging symbols found)...(no debugging symbols found)...
(no debugging symbols found)...[New Thread 1024 (LWP 29717)]

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 1024 (LWP 29717)]
0x8048e44 in nsCOMPtr_base::~nsCOMPtr_base () at eval.c:88
88      eval.c: No such file or directory.
(gdb) bt
#0  0x8048e44 in nsCOMPtr_base::~nsCOMPtr_base () at eval.c:88
#1  0x8048fc0 in _IO_stdin_used () at eval.c:88
#2  0x401802eb in __libc_start_main (
    main=0x8048d88 <nsCOMPtr_base::~nsCOMPtr_base(void)+1364>, argc=1, 
    ubp_av=0xbffff954, init=0x804871c <_init>, fini=0x8048f14 <_fini>, 
    rtld_fini=0x4000c130 <_dl_fini>, stack_end=0xbffff94c)
    at ../sysdeps/generic/libc-start.c:129
--

I suspect all the other things are just consequential.

(I'm not sure where this belongs, it looks pretty bad, so I leave it up to
somebody who knows better where to assign this bug)

Comment 2

[Boris Zbarsky [:bzbarsky]]

To blizzard... Looks like the printsettings and IDN landings are not correctly
merged into the RPMs.

Comment 3

[Christopher Blizzard (:blizzard)]

Works fine for me here.  Where are you getting those rpms, anyway? /opt/gnome is
not where I install things.

Comment 4

[Wouter Van Hemel]

The srpms are the ones on ftp.mozilla.org. I just patched the spec-file to use
/opt/gnome instead of /usr - that's not the problem, I've been doing that since
0.9.1 or something.

It's cleaner to install like that on my slackware system, which has mozilla in
/opt/gnome.

Whatever the problem is, I don't think regxpcom is supposed to segfault anyway,
is it? ...

If regxpcom would register the components properly, maybe mozilla itself
wouldn't show the errors it does now.

I've checked all build-files, and recompiled a couple of times, but in vain -
the compilation itself works out fine (just like all previous versions I did
from srpm), but 0.9.9 won't run.

If there's some test you want done, tell me. I'm a bit clueless here.

Comment 5

[timeless]

i'm not sure the stack trace makes sense, but assuming it does, please see 
www.mozilla.org/unix/debugging-faq.html

Comment 6

[Wouter Van Hemel]

I will... but regxpcom crashes during installing of the (bin) rpms build from
the srpm. I think there's no point in debugging mozilla if the components
already can't register?...

What's the exact function of that regxpcom? I guess it's not supposed to
segfault whatever it's arguments are. This is weird, I will do a memory- and
hdcheck and re-download the srpm (or does somebody have a md5-sum --
88a192953f33350a624bdaece074f48e  /usr/src/rpm/SRPMS/mozilla-0.9.9-0.src.rpm).

I've built and rebuilt it about 5 times, with the same result: regxpcom
segfaults in rebuild-databases.sh when installing the rpm on a _clean_ system;
mozilla refuses to start with the errors mentioned above.

Are the mozilla errors a possible cause of regxpcom segfaulting?

Comment 7

[timeless]

sure there is, you can debug regxpcom...

Comment 8

[irb]

For what it's worth, I get this same crash in regxpcom on both glibc2.1.3 and
glibc2.2.5, both compiled with gcc-2.95.4, but built from mozilla.org tarball,
not srpm. CFLAGS/CXXFLAGS were set to:

-O6 -ffast-math -fexpensive-optimizations -march=i686 -mcpu=i686
-malign-functions=2 -malign-jumps=2 -malign-loops=2 -fomit-frame-pointer -pipe

.mozconfig:

~# egrep -v ^# ~/.mozconfig

ac_add_options --disable-tests
ac_add_options --disable-debug
ac_add_options --enable-strip
ac_add_options --enable-strip-libs
ac_add_options --enable-xft
ac_add_options --disable-mailnews
ac_add_options --enable-optimize=-O6
ac_add_options --with-pthreads
ac_add_options --with-system-jpeg
ac_add_options --with-system-png
ac_add_options --with-system-mng
ac_add_options --with-system-zlib
ac_add_options --enable-crypto
ac_add_options --enable-default-mozilla-five-home=/usr/local/lib/mozilla
ac_add_options --prefix=/usr/local/lib/mozilla

I'll try a less aggressively optimized build and see if that works.

/i.

Comment 9

[Christopher Blizzard (:blizzard)]

ac_add_options --enable-xft

That flat out won't work.  I know that.

Comment 10

[irb]

Nope, rebuilding without --enable-xft didn't work either. Same environments
(glibc 2.1 and 2.2), still crashes in regxpcom. Will try the less optimized
build tonight, if possible.

/i.

Comment 11

[Wouter Van Hemel]

I didn't have '--enable-xft' - I used the stock .spec file from the srpm (with a
slight path change).

I also rebuilt the thing with debug enabled, and without optimalisations -
regxpcom crashes. I updated my gdb to the latest version, since 5.0 didn't come
up with anything usable.

###!!! ASSERTION: Null nsIComponentRegistrar: 'registrar', file regxpcom.cpp,
line 194
###!!! Break: at file regxpcom.cpp, line 194
###!!! ASSERTION: You can't dereference a NULL nsCOMPtr with operator->().:
'mRawPtr != 0', file ../../../dist/include/xpcom/nsCOMPtr.h, line 650
###!!! Break: at file ../../../dist/include/xpcom/nsCOMPtr.h, line 650

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 1024 (LWP 516)]
0x0804a210 in __builtin_delete ()

This is the message I get from regxpcom now. I'm not sure if the segfault comes
from the assertion-errors - something else you want me to try at this point?

I never had a problem before with regxpcom (or the whole mozilla-code), so I
think it must be a recent change in this code...

I'm using glibc-2.2.3 and gcc 2.95.3 (unpatched release, AFAIK - the one with
slackware 8.0).

Please tell me if you want me to try something else.

Comment 12

[timeless]

reporter,
1. please try deleting component.reg
2. please read the debugging faq and get a stack trace for the null pointer 
deref (setenv XPCOM_DEBUG_BREAK abort || setenv XPCOM_DEBUG_BREAK suspend)

Comment 13

[Wouter Van Hemel]

I traced regxpcom like you said, and traced it back to a library problem. I
investigated my lib-dirs (like /usr/lib) which shouldn't contain mozilla libs,
and I found a couple of old versions there (dated from 11.feb.2002) - not of all
libs, just a few (regxpcom and nss/nspr stuff). Never having installed any of
those there, I strongly suspect one of the programs that use mozilla in one way
or another has copied/installed there something that it shouldn't have touched.

Since /usr/lib got checked before the other dirs, some invalid libs were used -
after deletion everything worked fine (Is there no version check of some kind?...).

Closing this 'bug' - sorry about your time.

(It's worth remembering though I think, this might happen to more people. As of
now, I don't know which program it could have been - I compile new programs
daily, and I can't remember what I did yesterday, left alone a month ago. ;) )

Comment 14

[timeless]

doh. thanks for following up.  i need to remember to suggest path issues first. 
I'm glad you found the problem.

vrfy.