[Gc] Re: gcc 3.4 gc-test program boehm-gc/tests/test.c fails on linux 2.6.7

John Lumby johnlumby at hotmail.com
Tue Aug 17 08:27:10 PDT 2004


I'm sorry, I do not follow gcc development, therefore I don't know whether 
the tree I downloaded is after "Bryce finished the 6.3 merge".    How would 
I tell?    I don't even understand that version number since the versions 
I'm familiar with are like 3.4.1 etc.

I did this on Sunday 15th:
   CVS_RSH=ssh CVSROOT=:ext:anoncvs at savannah.gnu.org:/cvsroot/gcc cvs -z9 
checkout  -P gcc
which has const char version_string[] = "3.5.0 20040815 (experimental)";
then configure and make and then make -k check which failed with :
make[3]: Entering directory 
`/home/gcc_from_cvs/040815/build_dir/i686-pc-linux-gnu/boehm-gc'
Switched to incremental mode
Emulating dirty bits with mprotect/signals
Segfault at 0x1040ef64
Unexpected bus error or segmentation fault
FAIL: gctest

(and just to clarify one thing, I did not alter any source in this tree at 
all).

This failure looks slightly different from the one wih the 3.4.1 (more 
messages about the error, whereas on 3.4.1 all I see is "Killed").

My environment is i686 pentium III with glibc 2.3.2 and kernel 2.6.7      
(same failure on another machine which is same except pentium II).

Yes, I was not intending to publish my workaround fix of disabling 
incremental although I'm happy to do so if need be.     A real fix would be 
better.     However I can confirm that I've gone ahead and installed my 
3.4.1 with that change and oher compiles since then appear to be fine

John
----Original Message Follows----
From: Hans Boehm <Hans.Boehm at hp.com>
To: John Lumby <johnlumby at hotmail.com>
CC: gc at napali.hpl.hp.com, Hans.Boehm at hp.com
Subject: RE: [Gc] Re: gcc 3.4 gc-test program boehm-gc/tests/test.c fails 
onlinux 2.6.7
Date: Mon, 16 Aug 2004 21:38:27 -0700 (PDT)

I had missed the other bug report.

Did you try after Bryce finished the 6.3 merge?  If that still fails, I'd
like to track down the problem.  I did some testing on a 2.6 Itanium
machine, but my X86 testing probably used older kernels.

If you were planning on having this patch checked into the 3.4.2 tree
we should post it on java-patches.  It may be better to backport a
real patch, once we're sure we have the correct one.  I think either one
is fine for 99.5% of applications, but there may be someone out there
trying to use the incremental collector by turning it on in native code.

Hans

On Mon, 16 Aug 2004, John Lumby wrote:

 > Thank-you Hans.    I made that change and also made the following two
 > changes to the compiler itself to be consistent:
 > . in gcc-3.4.1/boehm-gc/misc.c
 >        ifdef-out entire function GC_enable_incremental
 > . in gcc-3.4.1/boehm-gc/pcr_interface.c
 >       change call to GC_enable_incremental() into an ABORT
 > This is on the assumption that the gcc never uses this function as you
 > indicated.
 >
 > With those changes, build ok and make check ok.
 >
 > Perhaps you could kindly confirm that these changes are ok for gcc and I 
can
 > then install it.
 >
 > By the way, just for interest I tried building and make check'ing the gcc
 > tree from cvs and it also fails in the boehm-gc test, so the changes you
 > mentioned that you recently made
 > (pthread_stop_world.c)  are apparently not enough.  I also see 
gcc-Bugzilla
 > Bug 15812 open.  (I personally am not intending to instal gcc 3.5 until 
it
 > becomes stable release)
 >
 > John
 >
 > ----Original Message Follows----
 > From: Hans Boehm <Hans.Boehm at hp.com>
 > To: John Lumby <johnlumby at hotmail.com>
 > CC: gc at napali.hpl.hp.com, Hans.Boehm at hp.com
 > Su  bject: RE: [Gc] Re: The gcc 3.4 gc-test program boehm-gc/tests/test.c
 > failson linux 2.
 > Date: Sun, 15 Aug 2004 21:54:37 -0700 (PDT)
 >
 > The easiest patch may be to remove the GC_enable_incremental call around
 > line 1809 in tests/test.c, or change the surrounding #if to a #if 0.
 >
 > (I'm not sure about the line number in your version, but it's the last
 > such call in the file.)
 >
 > The other workaround is probably to build with --enable-parallel-mark,
 > which is probably a good idea if you are targetting mostly
 > multiprocessors.
 >
 > Hans

_________________________________________________________________
Designer Mail isn't just fun to send, it's fun to receive. Use special 
stationery, fonts and colors. 
http://join.msn.com/?pgmarket=en-ca&page=byoa/prem&xAPID=1994&DI=1034&SU=http://hotmail.com/enca&HL=Market_MSNIS_Taglines 
  Start enjoying all the benefits of MSN® Premium right now and get the 
first two months FREE*.



More information about the Gc mailing list