I develop on an x86_64 machine with 32 GB of RAM. When I try to run this test (in my case, through |cd $OBJDIR; make check|) it completely DoS's my machine. It runs for at least 30 seconds and renders my machine completely unusable for a minute or so thereafter. (I haven't let it run long enough to finish.) If I read this test correctly, it attempts to build a string of 2^80 characters. I don't know what the behavior is on machines with "normal" amounts of memory (fast OOM death?), but it's totally unreasonable on my machine. In theory it could allocate >32GB of memory before being OOM killed, but I haven't measured. I don't know what's supposed to be tested here, but if it's OOM this is a *very* bad test. The more RAM your machine has, the longer the test runs and the worse you suffer afterwards. If this indeed tests OOM handling, can we please make a real test that uses rlimit on machines where that's available, and only runs on platforms with rlimit?

bug 513348 "fixes" this, the test completes much faster

chris, do you agree? (In reply to comment #1) > bug 513348 "fixes" this, the test completes much faster

Seemed to be OK last time I ran the trace-tests (before they were jit-tests).