AKCL and varargs...

[newton@vlsi.caltech.edu (Mike Newton)]

In an earlier message, Pat Gatewood writes :
>>	Subject: Can't make on sgi4d
	
>>I am having some troubles making akcl with the latest versions of kcl and
>>version 3.2 of the Irix operating system.  I get fatal errors in eval.c,
>>line 1125, format.c, line 2047, unixsave_sgi.c, lines 302:312, and in list.d,
>>line 249.
>>For a while I thought maybe I needed to get the GNU C compiler.  I did that
>>but it didn't fix the problems.
>> ...	
>>caused by COERCE_VA_LIST, called in eval.c, format.c, funlink.c and list.d,
>>...

As you noted, there are several problems with the current versions
(470,473) of akcl on any machine that is 'unusual' (ie: not array based)
w.r.t. varargs.  I've patched these all up and submitted the changes to
Bill Schelter, but he's not had time to inlcude them yet.

You will also probably have problems with {unixsave,sfasl}.c .  I
ended up doing pretty major surgery on this, but it was worth it,
as the sfasl loader is very fast.  Though it is was done
for the m88000, the changes will be useful for many machines.  You'll
have to do work -- segment alignments, relocation code..., but there
are a lot more comments than there used to be.

All these patches are on csvax.caltech.edu in ~ftp/DG .  I'm going
to put slightly newer versions there sometime soon (a day or two?),
but these changes will be mostly for the 88k.

- mike

ps: i finally got around to doing timings on this port.  If you
preallocate enough area [ ie : (allocate ...) in the core image ]
(and have core to match to prevent paging death), the timings
compared _very_ favorably with Lucid's and Franz's on such machines
as the HP850's (?), the SS1, ... .  Final results will be posted
(i'm building a new version now...).