🚀 go-pugleaf

Meissner@org.osf writes:

[ The building of emacs on a Sequent (i386) was failing, when built with
  ld from binutils-1.9, gcc-1.40 & gas-1.38.1 ]

| Simon Marshall writes:

| |         ./temacs -batch -l inc-vers
| | Loading inc-vers...
| | New Emacs version will be 18.57.21
| |         ./temacs -batch -l loadup.el dump
| | Loading loadup.el...
| | Loading subr...
| | *** Illegal instruction - core dumped
| | `all' not remade because of errors

| | 	I traced the "error" to mark_object() which was calling abort(),
| |      this is some of the code starting at 1015 in that function (file
| |      alloc.c), the switch is on (XGCTYPE (obj)), obj being the Lisp_Object
| |      mark_object's single arg points to:

[ The cause was obj being odd, when it was assumed by emacs to be even.
  It appeared that the problem was with GNU-ld, since linking with the host's
  ld resulted in emacs-18.57 building without a hitch. ]

| I have seen this before, on machines like the 386 which do not align
| bss items to word boundaries.  Emacs GC assumes that all objects are
| at least 16-bit aligned.

| I use the following shell script, hacking the Makefile to call it
| instead of 'ld' (you may have to tweak it to suit):

	Can anyone confirm this?  I was using all GNU stuff to build emacs
     (so it's not really our 386 machine at fault?), but GNU-ld was not
     putting BSS items on word boundaries, which GNU-emacs expects?  Is
     there a fix to GNU-ld for this, if not (IMHO) maybe there should be!

	Is this the problem?  Simon. 
_______________________________________________________________________________
Simon Marshall, Dept. of Computer Science, University of Hull, Hull HU6 7RX, UK
	      "``La la la la la la la la la'' means I love you."
 Email: S.Marshall@Hull.ac.uk    Phone: +44 482 465951 (office)    Fax: 466666