Re: [Gcl-devel] The state of gcl/maxima



Greetings!  Thought I'd give a brief update here.  gcl/maxima is now
very close to running without issue on all 11 Debian architectures!  

When this round of porting is wrapped up, hopefully in a few days,
I'll post another update, which should then be somewhat constant as we
turn attention to missing features.

P.S. -- it sure would be nice to get rid of those pejorative comments
	regarding gcl in maxima's README.lisps :-)

Camm Maguire <camm@enhanced.com> writes:

> cpu    OS       BFD    SGC   compile  max5.6 5.6test    Note:
>                     (kernels)
> 		   	     
> x86    Linux    yes    all   pass     pass   pass
> ppc    Linux    yes    2.4x  pass     pass   pass       (1)
> sparc  Linux    yes    2.4x  pass     pass   pass       (2)
> m68k   Linux    yes    all   no       no     no         (2.5)
> arm    Linux    yes     no   pass     pass   pass       (3)
> s390   Linux    yes    2.4x  pass     pass   pass       (4)
>          	   	     
> mips   Linux    no     all?  pass     pass   pass       (5)
> mipsel Linux    no     all?  pass     pass   pass       (5)
> 		   	     
> alpha  Linux    no      no   pass     pass   pass       (6)
> ia64   Linux    no      no   pass     pass   pass       (6)
> hppa   Linux    no      no   no       no     no         (6)
> 		   	     
> sparc  solaris  yes    yes   pass     pass   two errors (7)
> 		   	     
> x86    w32      no     yes   pass     pass              (8)
> 
> ???    xBSD     ???     ???  ????     ????   ????       (9)
> ???    osX      ???     ???  ????     ????   ????       (9)
> 
> 
> 
> 
> (1)  at least with kernels 2.4.x, SGC relies on siginfo_t here
> (2)  sparc seems to honor siginfo_t in newer kernels, but apparently
> 	minor issue remains in sgbc.c with PTR_ALIGN being set to 8.
	UPDATE:  Issue now resolved, sparc running SGC.
  (2.5) UPDATE:  In cleaning up -Wall, I broke this build, as it
        apparently treats functions returning pointers differently 
        from those returning integers.  A straightforward hack should
        restore this build, but I'd like to do something cleaner.  It
        would require changing the compiler to declare all functions
        returning object, with a cast of any long integers into an
	object at function return.  The issue is one of performance --
	gcl avoids unnecessary make_fixnum accesses to the memory
	manager when a temporary integer is simply briefly sought.
> (3)   UPDATE: sigill resolved.
> (4)  should work with 2.4.x, cannot yet test.  Can you test for me,
> 	Gerhard? 
       UPDATE: can run SGC, if work out a way to detect kernel 2.2.x.
> (5)  mips/mipsel should be fully functional once the relocation
> 	support in the bfd library is finished, or I get around to
> 	writing a workaround.  MIPS_GOT16 relocs are not currently
> 	handled by the bfd.  Any knowledgeable mips people interested?
       UPDATE: mips(el) now use dlopen for object loading.  This is
       hopefully temporary, but must wait on a patch to bfd.
> (6)  all these 64bit ports compile, but likely still have a few issues
> 	with sizeof(int)
       UPDATE: all (known) 64bit issues resolved.  Alpha and ia64 work
       without problem.  hppa still has a gc issue.  Same bfd
       situation as mips(el)
> (7)  Dan, can you chase these down in a debugger?  What about 5.9 on
> 	solaris? 
	UPDATE: Dan you were waiting on a clean -Wall build before
	looking into this.  We have this now.  Can you please report?
> (8)  Mike, would like to know if bfd and/or sgc are working for you.
> (9)  Anyone looking at these?
> 
> Take care,

-- 
Camm Maguire			     			camm@enhanced.com
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah