"Worse is Better" gives an entertaining view on this fact. pepe On 24/08/06, skaller wrote:

On Thu, 2006-08-24 at 08:56 +0100, Simon Marlow wrote:

...

Hi Folks,

Roman Leshchinskiy and I looked into the 6.4.3 crashes on Sparc/Solaris yesterday. I think we may have found the problem, and it seems likely that this is the same problem affecting 6.4.3 on MacOS X. The threaded RTS is assuming in a couple of places that pthread_cond_wait() doesn't spuriously wake up, which is apparently the case on Linux but not true in general.

I don't believe it is the case on Linux either. I had the same problem: on Linux it appears to work .. but Posix allows spurious wakeups, our OSX/Solaris/Windows code failed.

This isn't just for cond_wait: most system calls can return spuriously if an OS signal goes off. Somewhere I read a good article explaining why this is necessary for good performance. A signal to a condition variable should be regarded as a hint to recheck the condition.

-- John Skaller <skaller at users dot sf dot net> Felix, successor to C++: http://felix.sf.net

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

Gregory Wright wrote:

I built HEAD and ran ghc-regress last week. It looks like the compiler crashes are almost all gone. There are still 132 unexpected failures, but only one compiler crash.

Are these crashes all for the threaded2 way? We'll probably need to disable SMP on PPC for the 6.6 release, because I'm fairly sure it doesn't work (the workarounds we need for the weaker memory ordering on PPC aren't implemented). Ian is planning to sort this out before the release. Cheers, Simon

Gregory Wright wrote:

Here's what happened (this is after I applied David Kirkman's SMP.h patch to my tree). When I said 132 failures I was speaking from memory. The actual number was 136, as noted in the log:

OVERALL SUMMARY for test run started at Mon Aug 14 11:20:03 EDT 2006 1412 total tests, which gave rise to 4711 test cases, of which 0 caused framework failures 658 were skipped

3882 expected passes 32 expected failures 3 unexpected passes 136 unexpected failures

Unexpected passes: barton-mangler-bug(optasm) cholewo-eval(optasm) ds052(normal)

Unexpected failures: TH_spliceE5_prof(normal) arr004(normal,opt,optasm) arr007(opt,optasm) barton-mangler-bug(opt,threaded2) cabal01(normal) cabal02(normal) cg016(normal,opt,optasm) cg051(opt,optasm) conc012(normal) conc014(opt,optasm) conc021(opt,optasm) etc.

Interesting - a lot of failures from tests that ran (rather than just compiled), and mostly just when compiled optimised. Thankfully this doesn't appear to be threading-related, as these tests are all running with the single-threaded RTS. Cheers, Simon

I'll run the latest on my G4 tonight and we can see. Best Wishes, Greg On Aug 28, 2006, at 12:05 AM, David Kirkman wrote:

I ran the testsuite against the current HEAD (as of saturday afternoon) and had far fewer failures. Is the last week's work supposed to have fixed so many test cases? Or are different machines giving different results? The tests that resulted in unexpected failures for greg were all run on my machine, but many of them worked on my setup.

-david k.

------ Results for a lightly loaded dual proc G5 running OS 10.4.7.

OVERALL SUMMARY for test run started at Sat Aug 26 19:00:07 PDT 2006 1434 total tests, which gave rise to 6334 test cases, of which 0 caused framework failures 1107 were skipped

5138 expected passes 47 expected failures 3 unexpected passes 39 unexpected failures

Unexpected passes: barton-mangler-bug(optasm) cholewo-eval(optasm,profasm)

Unexpected failures: TH_repPatSig(normal) barton-mangler-bug(opt) conc019(opt) conc024(normal) conc039(threaded1) conc056(threaded1) ffi009(ghci) ffi016(normal,prof) gadt7(normal) galois_raytrace(opt,prof) maessen_hashtab(normal,opt,optasm,prof,profasm,ghci,threaded1) rnfail026(normal) signals002(ghci) tcfail068(normal) tcfail071(normal) tcfail090(normal) tcfail099(normal) tcfail103(normal) tcfail115(normal) tcfail140(normal) tcrun021(normal,opt,optasm,prof,profasm,ghci,threaded1) utf8_002(normal) utf8_003(normal) utf8_004(normal) utf8_005(normal)

On 8/24/06, Gregory Wright wrote:

...

Here's what happened (this is after I applied David Kirkman's SMP.h patch to my tree). When I said 132 failures I was speaking from memory. The actual number was 136, as noted in the log:

OVERALL SUMMARY for test run started at Mon Aug 14 11:20:03 EDT 2006 1412 total tests, which gave rise to 4711 test cases, of which 0 caused framework failures 658 were skipped

3882 expected passes 32 expected failures 3 unexpected passes 136 unexpected failures

Unexpected passes: barton-mangler-bug(optasm) cholewo-eval(optasm) ds052(normal)

Unexpected failures: TH_spliceE5_prof(normal) arr004(normal,opt,optasm) arr007(opt,optasm) barton-mangler-bug(opt,threaded2) cabal01(normal) cabal02(normal) cg016(normal,opt,optasm) cg051(opt,optasm) conc012(normal) conc014(opt,optasm) conc021(opt,optasm) conc024(normal,opt,optasm) conc034(normal,opt,optasm) conc037(threaded2) conc039(threaded1) conc052(opt) conc056(threaded1,threaded2) concprog001(normal,opt,optasm) concprog002(threaded2) dsrun005(opt,optasm) dsrun007(opt,optasm) dsrun008(opt,optasm) enum01(normal,opt,optasm) enum02(normal,opt,optasm) enum03(normal,opt,optasm) exceptions001(normal,opt,optasm) exceptions002(opt,optasm) forkprocess01(ghci) galois_raytrace(opt,threaded2) ghci013(ghci) ghcpkg04(normal) joao-circular(optasm) list001(normal,opt,optasm) maessen_hashtab(normal,opt,optasm,ghci,threaded1,threaded2) mod1(normal) mod150(normal) mod151(normal) mod152(normal) mod153(normal) mod2(normal) net001(normal,opt,optasm,ghci,threaded1,threaded2) net002(normal,opt,optasm,ghci,threaded1,threaded2) newtype(opt) pkg02_b(normal,opt,optasm) read003(normal) rw(normal) seward-space-leak(ghci) signals002(ghci) simpl011(normal,opt,optasm) strun002(opt) tc102(normal,opt,optasm) tc134(normal,opt,optasm) tc136(normal,opt,optasm) tc141(normal,opt,optasm) tcfail068(normal) tcfail071(normal) tcfail082(normal) tcfail090(normal) tcfail099(normal) tcfail103(normal) tcfail115(normal) tcfail140(normal) tcrun021(normal,opt,optasm,ghci,threaded1,threaded2) uri001(normal,opt,optasm,ghci,threaded1,threaded2) utf8_002(normal) utf8_003(normal) utf8_004(normal) utf8_005(normal)

...

Cheers, Simon

/Greg _______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

Hi Simon, On Aug 24, 2006, at 7:34 AM, Simon Marlow wrote:

Gregory Wright wrote:

...

I built HEAD and ran ghc-regress last week. It looks like the compiler crashes are almost all gone. There are still 132 unexpected failures, but only one compiler crash.

Are these crashes all for the threaded2 way? We'll probably need to disable SMP on PPC for the 6.6 release, because I'm fairly sure it doesn't work (the workarounds we need for the weaker memory ordering on PPC aren't implemented). Ian is planning to sort this out before the release.

Cheers, Simon

For the record, here are the crashes (not test failures) from build of HEAD on OS X ppc. I will sort through them to find out how many distinct failures there seem to be. Fortunately, the all knowing haskell-cafe list has explained my laziness to me and thereby fixed my stack overflow woes. Henceforth I promise to be, if not diligent and hardworking, at least strict :-) And thanks to them I should be able to find a few moments to look at the crash logs. Best, greg gregory-wrights-powerbook-g4-17> cat crashes Aug 14 11:23:42 gregory-wrights-powerbook-g4-17 crashdump[18790]: arr007 crashed Aug 14 11:23:42 gregory-wrights-powerbook-g4-17 crashdump[18790]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ arr007.crash.log Aug 14 11:23:53 gregory-wrights-powerbook-g4-17 crashdump[18806]: arr007 crashed Aug 14 11:23:53 gregory-wrights-powerbook-g4-17 crashdump[18806]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ arr007.crash.log Aug 14 11:38:30 gregory-wrights-powerbook-g4-17 crashdump[22522]: cg016 crashed Aug 14 11:38:30 gregory-wrights-powerbook-g4-17 crashdump[22522]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ cg016.crash.log Aug 14 11:50:37 gregory-wrights-powerbook-g4-17 crashdump[25697]: concprog001 crashed Aug 14 11:50:37 gregory-wrights-powerbook-g4-17 crashdump[25697]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ concprog001.crash.log Aug 14 11:51:34 gregory-wrights-powerbook-g4-17 crashdump[25773]: concprog001 crashed Aug 14 11:51:34 gregory-wrights-powerbook-g4-17 crashdump[25773]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ concprog001.crash.log Aug 14 12:00:32 gregory-wrights-powerbook-g4-17 crashdump[27844]: conc012 crashed Aug 14 12:00:32 gregory-wrights-powerbook-g4-17 crashdump[27844]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc012.crash.log Aug 14 12:03:59 gregory-wrights-powerbook-g4-17 crashdump[29119]: conc024 crashed Aug 14 12:03:59 gregory-wrights-powerbook-g4-17 crashdump[29119]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc024.crash.log Aug 14 12:05:30 gregory-wrights-powerbook-g4-17 crashdump[29822]: conc034 crashed Aug 14 12:05:30 gregory-wrights-powerbook-g4-17 crashdump[29822]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc034.crash.log Aug 14 12:05:34 gregory-wrights-powerbook-g4-17 crashdump[29842]: conc034 crashed Aug 14 12:05:34 gregory-wrights-powerbook-g4-17 crashdump[29842]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc034.crash.log Aug 14 12:05:37 gregory-wrights-powerbook-g4-17 crashdump[29858]: conc034 crashed Aug 14 12:05:37 gregory-wrights-powerbook-g4-17 crashdump[29858]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc034.crash.log Aug 14 12:09:03 gregory-wrights-powerbook-g4-17 crashdump[1346]: conc052 crashed Aug 14 12:09:03 gregory-wrights-powerbook-g4-17 crashdump[1346]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ conc052.crash.log Aug 14 12:16:48 gregory-wrights-powerbook-g4-17 crashdump[4545]: dsrun005 crashed Aug 14 12:16:48 gregory-wrights-powerbook-g4-17 crashdump[4545]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun005.crash.log Aug 14 12:16:50 gregory-wrights-powerbook-g4-17 crashdump[4561]: dsrun005 crashed Aug 14 12:16:50 gregory-wrights-powerbook-g4-17 crashdump[4561]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun005.crash.log Aug 14 12:17:10 gregory-wrights-powerbook-g4-17 crashdump[4723]: dsrun007 crashed Aug 14 12:17:10 gregory-wrights-powerbook-g4-17 crashdump[4723]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun007.crash.log Aug 14 12:17:12 gregory-wrights-powerbook-g4-17 crashdump[4739]: dsrun007 crashed Aug 14 12:17:12 gregory-wrights-powerbook-g4-17 crashdump[4739]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun007.crash.log Aug 14 12:17:22 gregory-wrights-powerbook-g4-17 crashdump[4814]: dsrun008 crashed Aug 14 12:17:22 gregory-wrights-powerbook-g4-17 crashdump[4814]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun008.crash.log Aug 14 12:17:24 gregory-wrights-powerbook-g4-17 crashdump[4830]: dsrun008 crashed Aug 14 12:17:24 gregory-wrights-powerbook-g4-17 crashdump[4830]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ dsrun008.crash.log Aug 14 13:37:32 gregory-wrights-powerbook-g4-17 crashdump[17478]: ghc-6.5 crashed Aug 14 13:37:33 gregory-wrights-powerbook-g4-17 crashdump[17478]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ ghc-6.5.crash.log Aug 14 13:49:45 gregory-wrights-powerbook-g4-17 crashdump[19142]: enum02 crashed Aug 14 13:49:46 gregory-wrights-powerbook-g4-17 crashdump[19142]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum02.crash.log Aug 14 13:50:29 gregory-wrights-powerbook-g4-17 crashdump[19167]: enum02 crashed Aug 14 13:50:29 gregory-wrights-powerbook-g4-17 crashdump[19167]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum02.crash.log Aug 14 13:50:40 gregory-wrights-powerbook-g4-17 crashdump[19186]: enum02 crashed Aug 14 13:50:41 gregory-wrights-powerbook-g4-17 crashdump[19186]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum02.crash.log Aug 14 13:51:44 gregory-wrights-powerbook-g4-17 crashdump[19255]: enum03 crashed Aug 14 13:51:44 gregory-wrights-powerbook-g4-17 crashdump[19255]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum03.crash.log Aug 14 13:52:25 gregory-wrights-powerbook-g4-17 crashdump[19279]: enum03 crashed Aug 14 13:52:25 gregory-wrights-powerbook-g4-17 crashdump[19279]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum03.crash.log Aug 14 13:52:37 gregory-wrights-powerbook-g4-17 crashdump[19299]: enum03 crashed Aug 14 13:52:37 gregory-wrights-powerbook-g4-17 crashdump[19299]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ enum03.crash.log Aug 14 13:54:16 gregory-wrights-powerbook-g4-17 crashdump[19460]: exceptions001 crashed Aug 14 13:54:16 gregory-wrights-powerbook-g4-17 crashdump[19460]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ exceptions001.crash.log Aug 14 13:54:21 gregory-wrights-powerbook-g4-17 crashdump[19480]: exceptions001 crashed Aug 14 13:54:21 gregory-wrights-powerbook-g4-17 crashdump[19480]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ exceptions001.crash.log Aug 14 13:54:24 gregory-wrights-powerbook-g4-17 crashdump[19496]: exceptions001 crashed Aug 14 13:54:24 gregory-wrights-powerbook-g4-17 crashdump[19496]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ exceptions001.crash.log Aug 14 15:00:44 gregory-wrights-powerbook-g4-17 crashdump[28757]: maessen_hashtab crashed Aug 14 15:00:44 gregory-wrights-powerbook-g4-17 crashdump[28757]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ maessen_hashtab.crash.log Aug 14 15:00:57 gregory-wrights-powerbook-g4-17 crashdump[28785]: maessen_hashtab crashed Aug 14 15:00:57 gregory-wrights-powerbook-g4-17 crashdump[28785]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ maessen_hashtab.crash.log Aug 14 15:23:13 gregory-wrights-powerbook-g4-17 crashdump[2235]: strun002 crashed Aug 14 15:23:13 gregory-wrights-powerbook-g4-17 crashdump[2235]: crash report written to: /Users/gwright/Library/Logs/CrashReporter/ strun002.crash.log

Could someone on MacOS X try the 6.4.x branch again? I just committed a fix that makes the threaded RTS more stable on Solaris, and I'm hoping it clears up the problems on MacOS X too. Remember to re-enable -threaded in ghc/compiler/Makefile if you previously disabled it.

The problem seems to be solved now, great work! Cheers, Wolfgang