That's great Anton!
Laurence, please try a build with this patch in it, hopefully you will have more luck:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20080623/064290.html
-Chris
Dear Chris,
Thanks for your help, and good luck debugging!
Ok, looks like I've nailed down the tablegen bug and even reproduced the
assertion on linux.Steps to reproduce: just apply the attached patch. This makes iteration
over PatterFragments map little bit slower, but stable. This explains, why
assertions is seen on OpenBSD only half times.I don't have idea, what's happening there, but I guess, that some pattern
inlining changes behaviour of InferAllTypes().That's great Anton!
Laurence, please try a build with this patch in it, hopefully you will have
more luck:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20080623/064290
.html
As Anton says, this appears to fix the problem on my boxes - I've tried
compiling the same bit over and over and no problems have arisen yet. Thanks
Anton and Chris for your speedy response (I notice it's already in svn!).
Yours,
Laurie
Laurence Tratt wrote:
As Anton says, this appears to fix the problem on my boxes - I've tried
compiling the same bit over and over and no problems have arisen yet. Thanks
Anton and Chris for your speedy response (I notice it's already in svn!).
I have patched llvm-2.3 and i have had 5 consequtive successful builds with gcc 3.3.5! Rejoice!
Just tidying the port ready for port@openbsd.org list now.