[Gcl-devel] 2.6.2 on amd64

[Camm Maguire]

Greetings!  Our release testing process has uncovered a few minor
issues on this platform which I've fixed in CVS head and 2.6.2pre:

1) One true bug in which the c stack could be overwritten in
   quick_call_sfun (eval.c) if the arguments passed were numerous
   enough.  This actually occurs in the maxima 5.9.0 regression suite,
   overstepping the bounds on temp_ar by 1 element.  I'm guessing this
   hasn't manifested itself elsewhere due to the odd stack alignment
   structure on this machine (more below)

2) pointers are 8 bytes long on amd64, but the stack alignment is only
   4 by default.  I've therefore added an explicit (long) cast in
   gcl_cmploc.lsp to precede inlined arguments which are fixnum
   constants, as they need to be extractable in call_proc_new via
   va_arg(,fixnum).  without this, only 4 bytes containing the int
   representation of the constant are written to the stack, leaving
   the other 4 uninitialized and garbage.

3( removed -m64 from cflags on amd64 leaving this up to the user.
   Modern distros set it by default when they want to.

4) reenabled SA_SIGINFO which had given problems before to enable SGC
   optimized garbage collection on this platform.

I don't feel this is worth a 2.6.2c3 tag at this point, but intend
instead to complete the tests result summaries/release notes, add to
2.6.2pre and then tag as 2.6.2.  If any of you have strange segfaults
you cannot track down and with to try the patch to eval.c, please feel
free. 

Take care,

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