Re: Segfault when running fibc benchmark in current trunk

[Juhani Rantanen]

Hi,

I have a bad case of flu right now, sorry for the delay in answering.

>> I found a problem that seems to be related to continuations, although
>> guile crashes in gc.
>
> I think I fixed this problem in master. Can you try again?

Works:
guile -s fibc_bug.scm

Works:
guile-tools compile fibc_bug.scm
guile
> (load-compiled "fibc_bug.go")

Segfaults:
guile-tools compile fibc_bug.scm
guile -e "(load-compiled \"fibc_bug.go\")"

Last case is how the "bench" shell script used to run the benchmarks first..

>> I have attached a minimal and self-contained
>> testcase here.
>
> Is this your code? I'd like to get it into Guile's test cases, but I
> think we'd need copyright assignment or a disclaimer.

These benchmarks are from Gambit Scheme distribution package, located in
misc/bench.tgz . I ported these benchmarks to run in guile but other than
that, they are from several sources (Gabriel, K&R and miscellaneous). It
seems to be useful to have them around however..

>> I have also found problems with other tests from the same benchmark
>
> I look forward to the next bug report :)

These benchmarks still expose several different bugs in current mainline
(although many were fixed by your changes!) so I put my version of the
benchmarks to

http://aj-group.net/~misty/bench_misty_hack.tar.bz2

You can run all the tests (-c false leaves temporary files in sys/guile):

./bench -c false guile all

or a single test:

./bench -c false guile ctak

Ctak allocates 1.6GB of memory while running, I don't know whether it is
normal or not.

Check functions guile_comp and guile_exec in the bench script to see how
it compiles and runs the benchmarks or run it with no parameters to find
out more about them.

Best Regards,
Juhani