[Gcl-devel] Re: problem with si::save-system on Mac

[Camm Maguire]

Greetings!  Great to hear from you Aurelien!

Aurelien Chanudet <address@hidden> writes:

> Hi,
> 
> > --disable-statsysbfd --disable-locbfd --enable-custreloc
> > I am going to default to custreloc for the Mac in CVS head.
> 
> I feel unconfortable with that. The custreloc code will result in huge
> dumps. Besides, I'm not sure this has ever worked with ACL2.
> 

OK, I'll be most happy to defer to your judgement, most especially if
you have the time to keep this feature up to date.  I could certainly
do it if pressed, but my time is to limited to place learning mach-o
high on the list.  

> > Aurelien, something is broken in the mach-o bfd support.  Briefly, the
> > reexec in init_darwin_done_compat fires up an executable which dies
> > with a memory access permission fault:
> >
> > Breakpoint 1, build_symbol_table_bfd () at sfasli.c:36
> > 36        bfd_init();
> > (gdb) n
> > 37        if (!(bself=bfd_openr(kcl_self,0)))
> > (gdb)
> > 39        if (!bfd_check_format(bself,bfd_object))
> > (gdb)
> > 43        if (!(link_info.hash = bfd_link_hash_table_create (bself)))
> > (gdb)
> >
> > Program received signal EXC_BAD_ACCESS, Could not access memory.
> > 0x00000000 in ?? ()
> > (gdb) r
> 
> How can I reproduce this bug ? What version of Mac OS X (10.3 or 10.4)
> ? What compiler version (gcc 3 or gcc 4) ? What version of gcl : CVS
> head ?
> 

To reproduce, just try saving an image and executing it.

Darwin hons.local 7.9.0 Darwin Kernel Version 7.9.0: Wed Mar 30
20:11:17 PST 2005; root:xnu/xnu-517.12.7.obj~1/RELEASE_PPC Power
Macintosh powerpc

gcc 3.3

gcl 2.6.7.

> > I've noticed that there are mach-o files in upstream binutils, at
> > least recently.  I've moved CVS head to binutils 2.16.  Did your stuff
> > get accepted into binutils proper, or are these files something else?
> 
> The files in binutils 2.16 are those of Apple. No relocation support
> at all in those files at all. I haven't gone through the whole process
> of submitting my files to the binutils people.
> 

OK, please note that HEAD and 2.6.8pre have moved to binutils 2.16, as
the older version no longer compiles reliably in Debian unstable.  Can
we migrate your code to this binutils version?  

If you do submit your code and it gets accepted, it might make
maintenance much easier.  I can help you if you are interested here. 


> > mach-o-reloc.c appears gone or unbuilt, leaving the two mach_o
> > routines we need in sfaslbfd.c (bfd_mach_o_craft_fp_branch_islands and
> > bfd_mach_o_inject_fp_branch_islands) undefined.  Can you update your
> > mach-o support to work with the binutils in CVS head if it is not
> > already supported in some other way, in which case can you please
> > advise what to do about the missing functions?
> >
> > Also, we still have an
> > outstanding issue when trying to configure with expanded maxpage
> > settings, as well as the SGC support.  Any progress here?
> 
> What's the issue exactly ? How can I trigger the bugs ? Please, feel
> free to provide me with a detailed description as I have some free
> time to devote to gcl now. Thanks !

Great!  Why don't we start with bfd, but as for the others

1) try building acl2 on top of, say, --enable-maxpage='512*1024'.  As
   big as you can.  My understanding of your implementation is that
   there is a hardoded heap max somewhere which does not scale with
   the configuration settings.

2) The SGC problem is the same you were diagnosing when we last
   corresponded.  The records are all in gcl-devel.  If memory serves,
   you found that the sgc_type_map values were not being predictably
   set and maintained.

Take care,

> 
> Aurelien
> 
> 
> 

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