powerpc-linux port

Wolfgang Thaller wolfgang.thaller@gmx.net
Sun, 28 Apr 2002 19:10:35 +0200

>I see there is already working support for Darwin. What will have to be
>adjusted to get from powerpc-apple-darwin to powerpc-unknown-linux?

Well, the first step should be to build an "unregistered" version - 
if there has ever been a working LinuxPPC port (I don't know about 
any), this is easy. If not, you will need the intermediate .hc files 
generated by a working version of GHC on some other platform. I don't 
know how to do this, but there are people on this list who know more. 
Basic unregistered support should be "relatively easy" to get to 
work, as LinuxPPC should not be too different from Linux/i386.
Other things that [might] require porting are:

createAdjustor: the implementation for foreign export dynamic. The 
Darwin/MacOS X implementation might work unchanged, that is, if 
LinuxPPC uses the same calling conventions.

At this point we would have a fully functional haskell compiler that 
generates slow-running binaries.

ghc-mangle: A hideous Perl script that does really bad things to 
assembly code in order to make it run faster :-). While I don't think 
that we can use the same code for Darwin and LinuxPPC, a lot of what 
I found out for the Darwin port can also be applied to Linux, so I'll 
be happy to share information.

ghc-split: A smaller, but still evil Perl script that splits one 
assembly language file into many. This is not required, but it can 
make compiled programs smaller (unless I've misunderstood something). 
I haven't yet ported this for Darwin, as Apple's strange dynamic 
Linker makes things difficult. A port for LinuxPPC might be easier.

ghci: I think this requires a) a dynamic loader, and b) some haskell 
code to generate a few instructions of PPC machine language.
There is already an ELF loader for x86, and b) is easy compared to a), IMHO.

native code generator: not necessary except for improving compilation 
speed. A lot of work, but the same code can be used for Linux and 
Darwin. However, I don't think it is really worth the time...

These are all parts that I know about. Building an unregistered 
version is most important, but I don't know enough to help with that.