We have been seeing the JavaTest harness crash java spuriously.
Not a Java backtrace, but a full and immediate core dump from "java".

I successfully fired up dbx on the core file, and got the following
at the head of the trace:

=>[1] kill(0x0, 0x6, 0x0, 0x0, 0xffffffff, 0x58038), at 0xef673ecc
  [2] abort(0x6, 0xef3fec98, 0xef3fead8, 0xf026b494, 0x0, 0x1), at 0xef6395f4
  ---- called from signal handler with signal 6 (SIGABRT) ------
  [3] kill(), at 0xef673ecc
  [4] abort(0x29, 0x29, 0xb41f0, 0x2, 0xef3fedf4, 0xef673f0c), at 0xef6395a8
  [5] java_lang_UNIXProcess_exec(0xb7e00, 0xb7e04, 0xee30a9c8, 0x4, 0x1, 0x1), at 0x3ada0


I think this puts the problem at an explicit call of sysAbort() at the end
of java_lang_UNIXProcess_exec in process_md.c.
(Didn't I just file the err = 1! bug against process_md.c!)

Calling sysAbort because you fail to exec the command the use supplied
is unreasonable. You should raise some exception.