I've removed the error logs, but if anybody else is interested in
patching ECL for OS X Intel, here is at least a first attempt that
looks like it works. Instructions for compiling are below.
---------- Forwarded message ----------
From: Brad Beveridge <brad.beveridge@...>
Date: Mar 30, 2006 6:59 PM
Subject: Re: [Ecls-list] Re: Build failure on OS X 10.4.5 x86
To: Juan Jose Garcia Ripoll <lisp@...>
I've run the tests and found 192 occurances of the string ERROR!!, I
don't know how this stacks up to regular ECL builds. Please find the
output files and patch against cvs attached.
The steps I had to go through were very few, the first problem I had
was in gcconfig.h, the I386 define block didn't have any entries for
DARWIN. After fixing that up I ran into an error in ffi_x86.c, a
generated file. The assembler code at ffi_x86.d:77 had "rep movsb",
Apple's gcc seams to require this to be on two lines. Infact I've
seen docs that suggest GCC ought to require it be split also. Anyhow,
the change should be benign for regular GCC too I think.
And that's about it :)
If this patch is accepted, then OS X Intel users will need to
1) Have DarwinPorts installed
2) Have gmp installed (sudo port install gmp)
3) configure ECL with LDFLAGS=3D-L/opt/local/lib ./configure
If you can tell me which tests are failing unexpectedly, I can take a
look at them.
On Thu, 2006-03-30 at 19:04 -0800, Brad Beveridge wrote:
> I've removed the error logs, but if anybody else is interested in
> patching ECL for OS X Intel, here is at least a first attempt that
> looks like it works. Instructions for compiling are below.
I have committed your patches. Could you now try building ECL from
scratch as follows
./configure --prefix=3D$HOME --with-tcp
The flag '--host=3D...' should be automatically passed to the GMP library
and it is precisely the one that Darwinports uses, deactivating the low
level optimizations and going for a more portable library.
The problem seems to be that Apple uses a completely different assembler
syntax on OSX/Intel and that thus the GMP guys cannot reuse the code
they have for other platforms. Not even GMP 4.2 seems to build on
BTW, non of the tests that failed on your machine are unexpected. Most
of them are related to either MOP issues or accuracy of the signal
handler for floating point exceptions.
Max-Planck-Institut f=C3=BCr Quantenoptik
Hans-Kopfermann-Str. 1, Garching, D-85748, Germany
Phone: +49 89 32905 345 Fax: +49 89 32905 336