Version 17 (modified by shepheb, 9 years ago) (diff)



Everything from here to the "Original Page" heading is for the revival in July 2008.


A Hackathon in the San Diego area for sometime in 2008 was suggested. Thoughts?

The Influence of GHC 6.10

Any notable pending changes we need to wait for? Do we need to wait for the final 6.10 release?

Owned Devices

All developers, what ARM-based devices do you have available for testing?

  • shepheb has a Nokia N810.

First Unregisterised Hack Notes

This documents shepheb's first attempt at a crude, unregisterised build for the Nokia N810 using Maemo 4 Diablo and the Maemo SDK (scratchbox-based)

I'm following the Building/Porting guide's section on porting to a new platform.

Target-side Changes

utils/pwd/pwd is an x86 binary, and there's no GHC to recompile it

The guide's suggestion to run

./configure --enable-hc-boot --enable-hc-boot-unregisterised

fails because utils/pwd/pwd is a Haskell binary that I can't run, since this is the ARM-based emulated target system and not the x86 it was compiled for. I can't reach the host's GHC to recompile it, so I just replaced it with the hacky

echo /home/braden/ghc/ghc-6.9.20080614

and the above ./configure worked fine.



cd includes

step in the porting guide fails with an error message implying that libgmp headers were missing. Well clearly, they aren't included in the Maemo SDK. I grabbed the GMP sources inside the SDK scratchbox environment, and did

make check
make install

and all was well. Since make check passed, I have to hope it (cross-)compiled properly and I have a working libgmp now.

After that the includes/ make worked fine. That completes the target-side part for now.

Host-side Changes

utils/pwd/pwd again

utils/pwd/pwd failed again, not sure why this time (I run an i686 Arch Linux and custom kernel, and I'm using 6.8.2. was it built for i386? using 6.9.*?). Anyway, running

cd utils/pwd
ghc -o pwd pwd.hs

fixed it, so that's that.


As the porting guide says, I had to change the TARGETPLATFORM to arm-unknown-linux. The LeadingUnderscore setting can be left at NO, which is the target-side setting.

Need libffi

GHC 6.9 now uses libffi, and so I needed to install it too. I Googled libffi, installed the newest version (it was released in April 2008 so I assume it's old enough to be the one used in 6.9), built and installed it on the host machine.

Its header files are not installed into a system-wide include directory, so I symlinked them into /usr/include. Then the cd compiler && make boot && make step ran fine (though it took long enough running natively on my Core 2 Duo 2.8 GHz that I fear how long the emulated stage 2 will take).

make in /rts

This fails saying

ghc-6.9.20080614: could not execute: /home/braden/src/ghc-6.9.20080614/driver/mangler/ghc-asm

There is no such file, though there is the script driver/mangler/ghc-asm.lprl. I tried making it executable with no results, but I'm not really sure what to do next. Any pointers are greatly appreciated!

Of possible note is that I re-ran ./configure a while back while trying to get /compiler to make boot && make in an effort to make it find libffi. I doubt that would matter, but I'll record it anyway.

Original Page

GHC port for arm-unknown-linux-gnu

My goal is to create a registerised port of GHC to the nokia 770.


There is currently an unregisterised build available for Maemo 1.x. This project is temporarily on hold while two big transitions take place:

  1. Maemo 2.0 supports the new EABI standard which affects a bunch of things (done) (for more info see )
  2. GHC 6.6 release (soon)

Setting up the build environment

I have been using the standard maemo cross-development environment. Instructions for setting up this environment can be found here:

Changes to standard procedure

The updated instructions on this page should now work:

With two small changes:

(1) I had to add --srcdir=.

Anyplace configure is called I get this error:

This configuration does not support the `--srcdir' option..

Adding --srcdir=. makes the error go away.

(2) ghc/Makefile SUBDIRS ordering

This has been fixed in head, but if you download the 6.4.2 release you will need to edit ghc/Makefile and change the ordering of the SUBDIRS so that lib comes before compiler.

This is the default ordering:

ifeq "$(BootingFromHc)" "YES"
SUBDIRS = includes rts docs compiler lib utils driver

and you want

ifeq "$(BootingFromHc)" "YES"
SUBDIRS = includes rts docs lib compiler utils driver

That should get to the point of having a ghc-inplace built.

NOTE: if you try to move the directory to a new location or name, the inplace compiler will stop working because it has absolute paths hard coded to the current location.

Build ghc using ghc-inplace

(1) unroll the ghc source tarball into a new directory.

(2) ./configure --srcdir=. --with-ghc=/abs/path/to/ghc-inplace

(3) create a mk/ with these two lines:

GhcUnregisterised = YES
GhcWithNativeCodeGen = NO

(4) make

(5) make install

I think there may have been one other step in there somewhere...

This should build and install ghc. Unfortunately, the floating point code will be broken.

Run the test suite

(1) get the testsuite that corresponds to your release, for example:

(2) untar it in the ghc-6.4.2 directory.

(3) edit mk/ and change the -e config.time_prog line to:

        -e config.timeout_prog=\"\" \

I had to do this because the timeout program interacted badly with the scratchbox shell causing all the tests to timeout and fail.

(4) cd to test/ghc-regress

(5) make TEST_HC=ghc fast


(5) make TEST_HC=ghc # for a longer test

Step By Step Porting to Maemo 2.0

T & H

tar -xvjf ghc-6.6-src.tar.bz2
cd ghc-6.6

$ ./configure --enable-hc-boot --enable-hc-boot-unregisterised --srcdir=.
$ cd includes
$ make

$ ./configure --srcdir=.

Create H/mk/, with the following contents:


GhcUnregisterised = YES
GhcLibHcOpts = -O -fvia-C -keep-hc-files
GhcRtsHcOpts = -keep-hc-files
GhcLibWays =
SplitObjs = NO
GhcWithNativeCodeGen = NO
GhcWithInterpreter = NO
GhcStage1HcOpts = -O
GhcStage2HcOpts = -O -fvia-C -keep-hc-files
SRC_HC_OPTS += -H32m
GhcBootLibs = YES

Change Target* and TARGET* variables in H/mk/


TARGETPLATFORM			= arm-unknown-linux

TargetPlatform_CPP		= arm_unknown_linux
TargetArch_CPP			= arm

arm_unknown_linux_TARGET       = 1
arm_TARGET_ARCH      = 1

Copy T/ghc/includes/ghcautoconf.h, T/ghc/includes/DerivedConstants.h, and T/ghc/includes/GHCConstants.h to H/ghc/includes. Note that we are building on the host machine, using the target machine's configuration files. This is so that the intermediate C files generated here will be suitable for compiling on the target system.

Touch the generated configuration files, just to make sure they don't get replaced during the build:


$ cd H/ghc/includes
$ touch ghcautoconf.h DerivedConstants.h GHCConstants.h mkDerivedConstants.c
$ touch mkDerivedConstantsHdr mkDerivedConstants.o mkGHCConstants mkGHCConstants.o

I just followed the guide upto making the hc bundle. I had to comment out this line in H/Makefile:

#	echo ghc-$(ProjectVersion)/libraries/haskell-src/Language/Haskell/Parser.hs >> hc-files-to-go

because that file does not seem to exist anymore.

Wrong Stuff

Oops, I mis-read the directions, so this next section is junk.

Build the compiler on the host. There seems to be a circular depends between utils and compat so I had to hack it a bit. First edit H/utils/Makefile and remove ghc-pkg from the SUBDIRS list in the else clause.


SUBDIRS = mkdependC mkdirhier runstdtest hasktags hp2ps hsc2hs \
	  parallel prof unlit genprimopcode genapply runghc

Then run 'make boot' in the utils directory


$ cd H/ghc-6.6/utils
$ make boot

Now restore ghc-pkg to the SUBDIRS line:


SUBDIRS = mkdependC mkdirhier runstdtest ghc-pkg hasktags hp2ps hsc2hs \
	  parallel prof unlit genprimopcode genapply runghc

And build H/ghc-6.6/compat and then utils:


$ cd H/ghc-6.6/compat
$ make boot && make
$ cd H/ghc-6.6/utils
$ make boot && make