gcl-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gcl-devel] Regarding the release ...


From: Camm Maguire
Subject: [Gcl-devel] Regarding the release ...
Date: 24 Feb 2003 10:44:56 -0500
User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2

Greetings!  I wanted to touch base regarding the release.  I have one
final commit which I would like to make, but it has turned up a few
difficulties which I can not yet resolve.  

Briefly, I've reworked the ansi image build procedure to follow that
of the standard image -- i.e. with shipped .c .h and .data files that
are rebuilt from lisp source when needed, and which compile to .o
files which are linked directly into raw_ansi_gcl.  The goal is to
enable the shipping of an ansi image on systems which still link with
dlopen -- ia64, hppa, mips, mipsel, and alpha.  (When using dlopen, it
is impossible to load .o files into the lisp core, save the system,
and find them loaded on restart.  With this working, the 2.5.0 release
would transparently provide the option of ansi vs traditional image to
all GCL users, including the compilation of maxima and acl2 on all
platforms. 

OK, so what I have builds an image which performs identically on
Paul's tests as does the current ansi image.  Maxima is built without
error, but an error (fortuitously, it turns out) showed up in the
maxima regression tests.  On investigation, I discovered that the ansi
image, built as we do now or in my prototype version, compiles maxima
significantly differently than the standard image.  Try it!  Just set
the compiler::*default-c-file* (h data) variables to t and look at the
source produced.  Inline function call optimization appears to be
totally turned off, accounting for the performance degradation I had
observed earlier but erroneously attributed to the image size.  I have
been unable as yet to figure out what is causing this.

As I only have about 1 week left, I'm trying to decide whether to fish
or cut bait.  Suggestions welcome.  The issue is probably simple, but
until it is sorted out I would feel very uncomfortable advertising the
ansi image in a stable release.

So we have the following decisions:

a) do we want preliminary ansi support in this next release?

b) if so, should I commit what I have now so that others can help
pinpoint what's going on, within this week?

Take care,

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




reply via email to

[Prev in Thread] Current Thread [Next in Thread]