Okay, thanks for the explanation.
But that does not completely answer the original question: now, using branch 7 of GHC what can you do to get a haskell program compiled on/for an ARM platform without using Debian? You have to use LLVM? So you have to compile your program on a regular x86/x64 PC for LLVM backend and then use that bytecode on your ARM platform, is that it? Is LLVM bytecode that portable? I don't much about LLVM, so sorry if those questions feel a bit dumb ;)
ghcarm speaks about the Pandaboard and LLVM in a post: http://ghcarm.wordpress.com/2011/07/03/llvm-on-arm-testing

And thanks for the information about git-annex, I'm checking that out, it looks interesting ;)
Actually, bottomline I would be interested in running a web app (preferably using Yesod) on a Raspberry Pi (or similar, but more expensive), but this use case is cool too.

Le 10 avril 2012 13:13, Joachim Breitner <mail@joachim-breitner.de> a écrit :
Hi,

Am Dienstag, den 10.04.2012, 13:04 +0200 schrieb Yves Parès:
> > All these are not cross-compiled, but natively
> > compiled on the repective architecture, and I don’t think it is
> easily
> > possible to cross-compile GHC itself even today.
>
> So how did they get compiled the first time? How do you get a GHC
> working on or for an ARM platform if you don't use Debian?
> And why was Joey Hess talking about performance issues?
> (I'll be eventually interested, as Graham Klyne suggested earlier, in
> compiling for Raspberry Pi, if the hardware suits).

well, GHC was more portable in version 6.8 and before (this is not
cross-compiling, at least not really:
http://hackage.haskell.org/trac/ghc/wiki/Building/Porting

When I ported GHC to s390x half a year ago, I think I started with
porting 6.8 and then kept building the next released version with the
previous. It would be great, though, if porting current versions
directly would become possible again.

Most of these architectures do not have a native code generator (so they
are compiled via C) and are unregisterized, i.e. GHC knows nothing about
their registers. Both cause a performance penalty; I don’t know numbers.
I assume this is what Joey refers to. But maybe also that ARM machines
tend to be slower :-)

I’m happily running git-annex on a NSLU2 (266MHz/23MB RAM ARM NAS
device) and have done so before it was registerized, so it is definitely
a useful target for Haskell.

Greetings,
Joachim

--
Joachim Breitner
 e-Mail: mail@joachim-breitner.de
 Homepage: http://www.joachim-breitner.de
 Jabber-ID: nomeata@joachim-breitner.de

_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe