[Gcl-devel] Re: OpenBSD progress

[Magnus Henoch]

Camm Maguire <address@hidden> writes:

>> * W^X, "write xor exec", seems to cause problems with unexec.  The
>>   Emacs people haven't solved them, but just disabled it, so I did the
>>   same.
>> 
>
> How?

ld -Z, as below.

>>   The -Z flag needs to be added to all ld calls, including the one in
>>   configure where it tries to find the value of DBEGIN.  It is not
>>   immediately obvious to me how to do that - I would imagine that
>>   setting LDFLAGS would do it, but LDFLAGS is not mentioned at all in
>>   configure.in.
>> 
>
> We can also explicitly add in configure.  Is there a relevant test, or
> is the best just to add when building on openbsd?

I think just adding it for openbsd is simple and sufficient, which
I've done in my patch.

>> There are still some changes in my tree which I haven't verified that
>> they're really needed.  I'll sort that out in the next few days.
>> 
>
> Perhaps when I get back on wed. I can take a look at your patche and
> merge if not too difficult for 2.6.2, providing the windows problems
> are still outstanding by this date.

Everything applies cleanly to the Version_2_6_1 branch except for the
patch to main.c, which is trivially adaptable.  The branch compiles,
and had enough marbles to complete the ANSI tests; I've done more
tests on the head version.

txtUeSHCAIxTt.txt
Description: Text Data

configure.in: The second hunk is for dlopen - OpenBSD has it in libc,
while the current test assumes it is in libdl.  AC_SEARCH_LIBS does
the equivalent of the commented out part, except that it starts
checking without any libraries, and that it adds the needed libraries
to LIBS instead of TLIBS - I don't see how TLIBS would make a
difference.  As statsysbfd works, but dlopen doesn't (or maybe I just
don't understand how it works), the hunk can be dropped.

bsd.h, linux.h: I commented out definitions of HAVE_AOUT, as they only
confuse things.

alloc.c: As malloc now checks for recursion, it's OK to use error()
for allocation error reporting.  There are also some fixes to
baby_malloc - it's not normally used, but I used it for debugging, so
I needed it to be in shape.

main.c: Extend data rlimit, as in alloc.c.

unexelf.c: Not the same as what I posted the other day, as I
overlooked a change from Emacs.

others: OpenBSD has elf_abi.h instead of elf.h.  Some types and macros
are different.

With these modifications, GCL still compiles and runs on Debian/i386
testing.

> Would be nice to know if you can run the ansi-test suite, the
> random-tester, maxima, acl2 and axiom, as time permits of course.

The ansi-test suite completes with 1868 out of 16259 total tests
failed; is that good or bad?  Full output can be found at
http://w1.312.comhem.se/~u31227643/ansitestoutput.gz .

Maxima 5.9.0 compiles, passes testsuite modulo known bugs, and runs.
ACL2 works as well.

I ran the random-tester for almost 100000 iterations, with
loop-random-int-forms.  No output except for the progress numbers -
that's good, right?

That would be all... or did I forget something?

Regards,
Magnus