Hi,

Recently I got some error reports from testing a neko app. The error message
was 'Fatal error in gc / Bad stack base'. There were some similar topics on
the neko/haxe lists, with more or less resolution:
http://haxe.1354130.n2.nabble.com/Bad-stack-base-in-GC-register-my-thread-inner-td5623216.html
http://www.mail-archive.com/[email protected]/msg01934.html
http://www.0x61.com/forum/programming-haxe-f233/neko-fatal-error-in-gc-t1285880.html

After some googling, I came over the following recent posts:
http://zachsaw.blogspot.com/2010/11/wow64-bug-getthreadcontext-may-return.html
http://comments.gmane.org/gmane.comp.programming.garbage-collection.boehmgc/4188

The posts may be related to the appearing issues (if we disregard the other
possible option of antivirus software messing around), since they inform
about a (allegedly rare) bug in the 32-bit emulation layer of 64-bit
windows, which makes it possible for a thread to read an invalid ESP value
(well, it is complicated and I don't say I understand it 100%, but the posts
describe it well enough), which statedly can have bad effect on GCs,
including boehm gc used by neko.

Now I built a boehm gc with debug infos, and also a neko without Use_DllMain
as suggested, and waiting for feedback.
The other alternative seems to be building neko for win32-x64, which does
not suffer from the issue (already did some progress with this).

BR,
Robin
-- 
Neko : One VM to run them all
(http://nekovm.org)

Reply via email to