[Gc] Re: GC_MALLOC() Failure
yazicivo at ttmail.com
Tue May 11 13:01:14 PDT 2010
Thanks for your prompt answers. (I needed to wait for running tests to
end on the related machine, and just now I had chance to test your
suggestions.) Findings are as follows:
- I couldn't reproduce the same problem with CVS tip.
- I learnt that strace bundled with the Debian GNU/Linux (lenny) doesn't
support mmap as a system call. OTOH, somehow, "strace -f -etrace=mmap"
and "strace -f 2>&1 | grep ^mmap" gives nearly identical results.
On Mon, 10 May 2010, Andrew Haley <aph at redhat.com> writes:
> strace -f -etrace=mmap
> and then look for which mmap() call fails.
On Mon, 10 May 2010, Ivan Maidanski <ivmai at mail.ru> writes:
> I'd recommend you:
> 0. use the latest BoehmGC from CVS (or, at least, gc-7.2alpha4.tar.gz);
> 1. first compile libgc without -DUSE_MUNMAP -DTHREAD_LOCAL_ALLOC -DPARALLEL_MARK -DNO_DEBUGGING but with -DGC_ASSERTIONS;
> 2. identify the reason GC_allochblk() returns 0 (every GC_malloc for a large object calls GC_allochblk).
More information about the Gc