Some further searching suggests that Java triggers and catches SIGSEGVs
as part of normal operation, and hence is expected to not work under gdb
without "handle SIGSEGV nostop pass". With this, both 6.0.1 and 6.0.2
don't crash in gdb, i.e. the crash in gdb probably isn't this bug.
This suggests that the bug could be a subtle kind of baseline violation:
some assumption that fails on Valgrind's and qemu's emulated CPUs and
very old real CPUs. (As noted above, 6.0.2 does crash in Valgrind.)
It also still could be memory corruption that happens not to trigger on
recent real CPUs, or that is in a code path only used on older CPUs.
I haven't investigated how Ubuntu's 6.0.2 packaging differs from the
failed attempt at 6.0.2 reported above, or whether it fixes any of our
other bugs. However, as the freeze rules do not allow it by default, I
suggest not uploading it to unstable without asking release team *first*.
