[Gc] Re: recent commit breaks default config tests

Ivan Maidanski ivmai at mail.ru
Mon Nov 2 00:12:16 PST 2009

Andreas Tobler <andreast-list at fgznet.ch> wrote:
> Hi Ivan,
> this commit breaks the tests on darwin x86:
> https://bdwgc.cvs.sourceforge.net/viewvc/bdwgc/bdwgc/alloc.c?r1=1.52&r2=1.53
> I get a failure here:
> Unexpected heap growth - collector may be broken

Ok. The recent patch just highlights the real problem. See: https://thread.gmane.org/gmane.comp.programming.garbage-collection.boehmgc/3502

> Test failed
> /bin/sh: line 1: 62744 Abort trap              ${dir}$tst
> FAIL: gctest
> 1 of 6 tests failed
> Please report to Hans.Boehm at hp.com
> The config is w/o any arguments:
> ../config.status --recheck
> running CONFIG_SHELL=/bin/sh /bin/sh ../bdwgc/configure
> --build=i686-apple-darwin9 --target=i686-apple-darwin9
> Just a fyi. I'll try to dig in later.

It would be nice.

> The failure does not occur when I
> use --enable-munmap=6 but then I also do not test the:
> #Switched to incremental mode
> #Emulating dirty bits with mprotect/signals

Because OSX isn't so mature as Win32 ;) The incremental mode is working on OSX only with MPROTECT_VDB but:

# ifdef USE_MUNMAP
    /* FIXME: Remove this undef if possible.    */
#   undef MPROTECT_VDB  /* Can't deal with address space holes. */
# endif

I think we need a volunteer here to either workaround this limitation or prove this is impossible.

> Andreas

Thanks for the report.


More information about the Gc mailing list