Patch: FYI: small test suite change

[Tom Tromey]

>>>>> "Mark" == Mark Wielaard <mark@klomp.org> writes:

>> * libjava.lang/SyncTest.java (run): Cache .class value.

Mark> 1. That .class constructs were not cached at all.

There is a PR for this.

We can't rely on SyncTest to test for this bug anyway.
It might or might not time out.  We should have a separate test case
that tests for exactly what we require.  Perhaps it could work by
running jcf-dump and examining the output?

Mark> 2. That when doing a class lookup (and stack walk) using the
Mark> interpreter was horribly slow (which seem to have been fixed in
Mark> 3.4).

This is an argument against the patch, but...

It would be nice to have a test case that tests for speed regressions
in stack crawling.  I could easily imagine another libffi buglet like
the fixed one slipping in (if not on x86 then some other platform).
I'm not exactly sure how to write one though, especially one that is
future-proof against faster machines, etc.

Tom