GC missed reference on Windows

GC missed reference on Windows

[Neil Jerram]

I've built a native libguile DLL for Windows (using MSVC, from the 1.6.4 
tarball - I'll release the DLL and build details shortly).  It generally 
works well, but I occasionally get instances of two errors:

- Access violation in a call to SCM_NEWCELL (from various callers), 
trying to write at address 0x00000083.

- Scheme-level error when printing a list that appears to contain a 
freed cell - as shown by "freed cell: GC missed a reference" in the 
backtrace.  The apparently freed cell is the CDDR of a list that is 
bound to a top level definition in a loaded module, so I find it 
difficult to believe that the GC sweep has really missed it.

If anyone has seen anything like these problems before, or has any tips 
on how to debug them, I'd appreciate your input.

Thanks,
      Neil



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Rob Browning]

Neil Jerram <neil@ossau.uklinux.net> writes:

> If anyone has seen anything like these problems before, or has any
> tips on how to debug them, I'd appreciate your input.

Can you easily try 1.6 CVS?  Some of the fixes since 1.6.4 may be
relevant.

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Neil Jerram]

Rob Browning wrote:

> Neil Jerram <neil@ossau.uklinux.net> writes:
> 
> 
>> If anyone has seen anything like these problems before, or has any
>> tips on how to debug them, I'd appreciate your input.
> 
> 
> Can you easily try 1.6 CVS?  Some of the fixes since 1.6.4 may be
> relevant.

Yes, thanks - good idea.  Is there a nightly snapshot of 1.6 anywhere? 
(I haven't got CVS up and running right now, although it wouldn't be a 
big deal to reinstate it.)

     Neil



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Rob Browning]

Neil Jerram <neil@ossau.uklinux.net> writes:

> Yes, thanks - good idea.  Is there a nightly snapshot of 1.6 anywhere?
> (I haven't got CVS up and running right now, although it wouldn't be a
> big deal to reinstate it.)

I'm not sure, although occasional snapshots would be a good idea.
Alternately, if you're not in a big hurry, I'm hoping to produce a
1.6.5 release candidate moderately soon.

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Ludovic Courtès]

Hi,

One day, 2 hours, 15 minutes, 33 seconds ago, 
Neil Jerram wrote:
> If anyone has seen anything like these problems before, or has any tips 
> on how to debug them, I'd appreciate your input.

Running Guile 1.6.4 within Valgrind [1] actually shows quite a lot of
suspicious things, most notably use of uninitialized values or
conditional jumps that depend on uninitialized values in the GC code
(eg. in `scm_gc_mark ()' IIRC).  I haven't tried to dig into this so far
due to lack of time and I don't have Valgrind right here so I can't give
you more details.  Did anyone try that before?

Thanks,
Ludovic.

[1] http://valgrind.kde.org/
    Note that Valgrind is only available for GNU/Linux on x86 machines.


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Rob Browning]

Ludovic Courtès <ludo@chbouib.org> writes:

> Running Guile 1.6.4 within Valgrind [1] actually shows quite a lot
> of suspicious things, most notably use of uninitialized values or
> conditional jumps that depend on uninitialized values in the GC code
> (eg. in `scm_gc_mark ()' IIRC).  I haven't tried to dig into this so
> far due to lack of time and I don't have Valgrind right here so I
> can't give you more details.  Did anyone try that before?

I think so.  Guile does some unusual things with memory (as compared
to other programs), so valgrind may or may not be reporting real bugs.
Here's an earlier thread on the same topic:

  http://lists.gnu.org/archive/html/guile-devel/2003-01/msg00005.html

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Neil Jerram]

Rob Browning wrote:

> Neil Jerram <neil@ossau.uklinux.net> writes:
> 
> 
>> Yes, thanks - good idea.  Is there a nightly snapshot of 1.6 anywhere?
>> (I haven't got CVS up and running right now, although it wouldn't be a
>> big deal to reinstate it.)
> 
> 
> I'm not sure, although occasional snapshots would be a good idea.
> Alternately, if you're not in a big hurry, I'm hoping to produce a
> 1.6.5 release candidate moderately soon.

I took a look at the diffs between 1.6.4 and 1.6 CVS today.  The ones 
that looked relevant to me were (i) the reordering of check_mtrigger and 
realloc in gc.c and (ii) the addition of a missing va_end in list.c. 
Are those what you had in mind?  (I'm afraid I didn't get as far as 
trying out the changes yet, not least because I haven't got a reliable 
repro scenario yet.)

Regards,
      Neil



_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel

Re: GC missed reference on Windows

[Marius Vollmer]

Neil Jerram <neil@ossau.uklinux.net> writes:

> Is there a nightly snapshot of 1.6 anywhere?

Yes, here

   http://www.gnu.org/software/guile/ftp-snapshots.html

-- 
GPG: D5D4E405 - 2F9B BCCC 8527 692A 04E3  331E FAF8 226A D5D4 E405


_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-user

Re: GC missed reference on Windows

[Rob Browning]

Neil Jerram <neil@ossau.uklinux.net> writes:

> I took a look at the diffs between 1.6.4 and 1.6 CVS today.  The
> ones that looked relevant to me were (i) the reordering of
> check_mtrigger and realloc in gc.c and (ii) the addition of a
> missing va_end in list.c. Are those what you had in mind?

I think the first one may have been the one I was thinking of.

-- 
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4


_______________________________________________
Guile-devel mailing list
Guile-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/guile-devel