This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: 3.3-branch QA assessment


Hi,
On Sat, Jan 25, 2003 at 09:23:56PM -0700, Tom Tromey wrote:
> I haven't investigated SyncTest yet :-(

An analysis can be found at:
http://gcc.gnu.org/ml/java/2003-01/msg00297.html

There are a couple of issues here:
- This test calls the ClassName.class construct alot in a loop.
  We don't optimise this by doing the initialization once but we
  call Class.forName() each time. This was always the case but...
- Class.forName() has become a bit slower in 3.3 since it now calls
  calls _Unwind_FindEnclosingFunction. Which wouldn't be so bad except
  for the fact that it needs dl_iterate_phdr
- And the garbage collecter also calls dl_iterate_phdr from a different
  thread. Which means that we sometimes get a deadlock situation.
  (The same deadlocks we were seeing with Eclipse btw.)

Andrew Haley proposed to import the new Gc6.2alpha which has a workaround
for this issue. But Hans Boehm isn't sure that it is ready yet to be
imported. See: http://gcc.gnu.org/ml/java/2003-01/msg00368.html

Which is all unfortunate since it is a nasty regression.

Cheers,

Mark


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]