We should consider having the test programs like random_abort and timestamp_abort that have a parent kill a child process use SIGQUIT or SIGABRT instead of SIGKILL so that the child process generates a core file. That way if the parent verification phase detects a problem we can see exactly what the child threads were doing at the time.

Some considerations:

• Where do the core files end up? If it is in the top level test/csuite directory then a make check will cause many of them to be generated in that directory. If it is in the database home directory, then only the most recent run will be there and there will be no confusion.
• What are the core files named? Again if they end up in the database home, it is clear what core file goes with what program. If they're in the test/csuite directory and named core.<pid> only, then it may be difficult to know which core file goes with a failure.