libjava testsuite failures (regressions 3.2 -> 3.3)

[Andrew Haley]

Jeff Sturm writes:
 > On 18 Jan 2003, Mark Wielaard wrote:
 > > We could do the same kind of thing, but this does not seem to be the
 > > root cause of this problem. The SyncTest as compiled by gcj -C seems to
 > > have a real deadlock some of the times (this does never happen with gij
 > > 3.2 it seems, which is also much faster on this test then gij 3.3).

Good catch!

 > I see... Class.forName() now calls _Unwind_FindEnclosingFunction which
 > needs dl_iterate_phdr.  (That also explains why 3.3 is much slower on this
 > test.)  Thanks for the stack dump.  I hadn't gotten a deadlock yet, but I
 > may have just been lucky.
 > 
 > I guess the simplest workaround is to #undef HAVE_DL_ITERATE_PHDR?

I think so.  This has been a latent bug for a long time -- other
places in libgcc also call dl_iterate_phdr().

Andrew.