Matthias Kilian wrote:
On Wed, Nov 19, 2008 at 01:18:39PM +0000, Simon Marlow wrote:
My only idea how to fix this it to add -pthread to cc-options and
ld-options of unix.buildinfo if SplitObjs=NO, but this sounds like
a crude hack to me.
If anyone has a better solution, please let me know.
SplitObjs is a red herring: it only hides the real problem, which is that
the unix package depends on the pthread library on OpenBSD, and doesn't
express this dependency in its .cabal file. The fix is surely to add
-lpthread to extra-libraries in unix.cabal. Right?
Yeah, or better add -pthread to ld-options and cc-options, at least
in theory; on OpenBSD, -pthread affects both the preprocessor (at
compile time) as well as the linker. On the other hand, the Semaphore
stuff are more or less wrappers, to -lpthread may do as well.
I'll make a diff, and if it works (i.e. no regressions), I'll send
it here in a few days (my machine is too slow, and i'm short of
time).
However, what about Linux? Doesn't it show the same problem? IIRC,
sem_wait(3) and friends are in libpthread on Linux, too.
We depend on librt, which apparently depends on libpthread, so we get away
with it. It wouldn't hurt to add pthread to extra-libraries for Linux too,
I imagine.
Cheers,
Simon
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
