RE: [Gc-announce] Gc6.2alpha1

["Boehm, Hans" <hans_boehm at hp dot com>]

It depends on the 3.3 timing, which unfortunately I've lost track of.  I was just told that 6.2alpha1 is completely broken on FreeBSD (which I can't test, but will try to fix immediately).  I would expect at least one or two more problems along these lines.  I did test the dl_iterate_phdr related changes.  But they're fairly subtle, and I don't yet have a great deal of confidence in them either.

I would be hesitant to put 6.2alpha1 in a release for which we didn't have at least 2 or 3 weeks of testing time left.

(The 6.2alpha1 release message forgot to explicitly mention a win32-specific change.  The collector now tries much harder to find address regions owned by the system memory allocator, so that it can avoid tracing those.  This makes things much more consistent with other platforms.  In some cases it's likely to be a significant performance improvement.  Bit it also involves some risk.)

Hans

> -----Original Message-----
> From: Andrew Haley [mailto:aph@redhat.com]
> Sent: Friday, January 24, 2003 11:16 AM
> To: Boehm, Hans
> Cc: java@gcc.gnu.org
> Subject: [Gc-announce] Gc6.2alpha1
> 
> 
> Boehm, Hans writes:
> 
>  > I just released 6.2alpha1.  This is labeled as an alpha release in
>  > spite of the fact that it mostly involves fixes for mostly minor
>  > bugs.  A couple of those bug fixes are fairly risky, and have
>  > received minimal testing.  In particular, this attempts to fix a
>  > problem with occasional deadlocks involving the glibc
>  > dl_iterate_phdr() call, which required some code restructuring.
> 
> Hans,
> 
> I wonder if we need to use this in gcc Release 3.3, soon to go gold.
> We have seen deadlocks in Java code.  What do you think?
> 
> Andrew.
>