[darcs-users] wishlist for darcs 2.1 -- green buildbots

[Karel Gardas]

zooko wrote:
> Next, karel's solaris box fails to compile, with this error message:
> 
> http://buildbot.darcs.net/builders/karel%20solaris%20Solaris-5.11% 
> 20nevada-79a%20x86_64/builds/43/steps/compile/logs/stdio
> 
> [ghc] src/OldFastPackedString.o
> 
> src/OldFastPackedString.hs:98:0:
>      Bad interface file: /usr/local/ghc-2008-05-04/lib/ 
> ghc-6.8.2.20080430/lib/base-3.0.1.0/System/Mem.hi
>          magic number mismatch: old/corrupt interface file?
> 
> This sounds like the kind of thing that happens when there are stale  
> files from a failed previous build, or something like that, but  
> buildbot is configured to start fresh everytime.  In particular, you  
> can see from the earlier "checkout" step in the same build that it  
> ran "rm -rf /buildbot/workspace/karel solaris/build".
> 
> So this suggests that there really is something wrong with ghc on  
> that system.  Any ideas, karel?

Not at all! I'm thinking if this might be hardware issues, but well, the
machine is running ZFS and there is no errors on zfs scrub. The machine
is running with ECC RAM enabled and there is no log about uncorrected
error. Also if you force build again (which I did in those two cases so
far noted) the new build runs fine. The only think might be kind of
local over-heating but only for ghc process, but I'm still not sure it's
possible. So if you find any real reason why it might happen I would be
glad to know, otherwise I will need to consider making cooling better or
to stop some other CPU hungry application. FYI: the machine is running 2
folding at home on 2 vbox linux guests and once darcs change is pushed
upstream it also runs 1 solaris native build and Qemu-based OpenBSD
build so at this time it's loaded 100%.

Karel
-- 
Karel Gardas                  kgardas at objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com