Hi Gabor,
thanks a lot for your effort on GHC cross-compilation. You are talking
about cross-compiling to ARM too. I would like to warn you here as GHC
HEAD build for ARM is broken since February due to LLVM backend
miscompiling stage2 -- at least that's my analysis of this so far. See
http://hackage.haskell.org/trac/ghc/ticket/7049
Hopefully once GHC/LLVM unbreaks I will be more than happy to test your
stuff for {solaris|linux}/x86->linux/arm cross-compilation.
Thanks!
Karel
On 07/11/12 12:19 PM, Gabor Greif wrote:
Hi all,
coming back on the cross compilers issue. As you know, I had a working
PPC cross compiler in a github fork that did not require any alien
tools, just a cross-gcc to distill all characteristics of the platform
that are needed by GHC.
Unfortunately this fork has bitrotten somewhat (mostly because the
alien-cross merge). Now I have some energy to port my solution to
HEAD. The question is how you feel about my suggested approach:
1) get all non-conflicting changes on HEAD (i.e. put them into the
official repo). These are:
* the 'awk' scripts (*) and makefile changes to create the target
constants (sizes, offsets).
2) make sure there are no regressions for some nights, so we can be
confident that I did not break anything.
3) The currently conflicting (though not invasively) modifications.
These include:
* the 2-stage build changes (**) still needed in makefiles and
configury (if still necessary). These include "--with-cross-gcc"
detection etc.
* the modifications to the GHC wrapper script. I have something
small working, but an experienced GHC implementor surely can point me
to a more general approach.
* modifications to the building of the libraries, so that the
correct cross-compiling options are passed to the package builds.
4) again, observe the bot for some time whether stuff borked, get some
beta testers into the boat, implement feedback.
5) Add some tests that exercise cross compilation (e.g. x86_64 host,
i386 target or x64 target with 32-bit pointer API on Linux). For
setting up these I'd ask for some help from the more experienced of
you.
It remains to clarify whether I can obtain commit privileges for the
above steps w.r.t. the HEAD branch of the official repo. I believe
that these privileges would speed up the porting a lot. For 1) I would
ask for post-commit review and for 3) I'd suggest pre-integration on a
branch followed by a pre-merge review, then merge to HEAD.
Alternatively, I'd just try to pull this off in a new github fork and
ask you for pulls to the official (if you feel hesitant about granting
me the privileges).
A bit of background... I got official permission to evaluate Haskell
in an embedded programming context for research projects from my
employer (Alcatel-Lucent). I have to present concrete evidence that
this direction of attack results in usable artifacts for our research
this summer, as the proposal for continuation of said activity will be
concluded in September.
Thanks for your attention and in hope for cooperation,
cheers,
Gabor
(*) I promise to convert all 'awk' scripts to 'runhaskell' scripts as
soon as the cross-compilation works from vanilla HEAD for PowerPC and
ARM.
(**) ultimately we'll need a 3-stage build for cross compilation,
because I consider Template Haskell as essential.
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
_______________________________________________
Cvs-ghc mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/cvs-ghc
