[Gc] Suggestion for Windows version.
Hans.Boehm at hp.com
Fri Feb 29 11:24:10 PST 2008
This looks like a bug to me. The current behavior was certainly
not intended. I'll check in a patch after some minimal testing.
On Fri, 29 Feb 2008, Roger Osborn wrote:
> First, as someone new here, I'd like to thank the people who maintain
> the list and who answer questions.
> We are using Boehm GC in a program that runs right up to the 2Gb user
> memory limit in 32 bit Windows. Because of this we quite often bounce
> off the condition where the VirtualAlloc call in GC_win32_get_mem
> returns zero. This ultimately causes an abort with the "Too many heap
> sections" condition.
> I was wondering whether there was any reason the canonical code could
> not be changed so that it only does the:
> GC_heap_bases[GC_n_heap_bases++] = result;
> if result is non zero? We have changed this is in our code, but at first
> sight it looks like it might be the sort of change that would make
> things better for everyone, in that there is more chance of the out of
> memory callback function being called reliably rather than the program
> ending abruptly.
> Roger Osborn
> Gc mailing list
> Gc at linux.hpl.hp.com
More information about the Gc