[Gcl-devel] strange tests with gcl 2.4.0

[Vadim V. Zhytnikov]

I've promised to send more precise information about some
bad test result with gcl 2.4. Today I've managed to reproduce
them.
So I have gcl 2.4.0 (official 2.4.0 tarball not later cvs
versions)
and Maxima 5.6. Gcl is compiled on Linux with maxpages=64*1024.
Test is performed with the following init.lsp

(progn
(si::allocate 'cons 20000 t)
(si::allocate 'fixnum 4000 t)
(si::allocate-relocatable-pages 10000 t)
(si::allocate 'cfun 5000 t)
(si::set-gmp-allocate-relocatable t)
(setq  si::*notify-gbc*  t))

which pre allocates lots of RAM.
I evaluate
    factor(expand((x+y)^n)+1)$
sequentially for n=30, 60, 100.
For n=30 and 60 I've got some reasonable
timings: 0.1 and 1.7 seconds respectively.
But for 100 gcl goes into infinite loop with
lots of cfun garbage collections.
I've killed Maxima after 10 minutes.
Notice that nothing like this is happening
to gcl 2.3.8. I get nice time for all
n=30, 60, 100, 105. Similar strange things
are happening with standard Maxima test
(tests.lisp). Some tests fails, some becomes
very slow. How many and which one precisely
depends on the amount of preallocated RAM.
Once again nothing like this is observed with
gcl 2.3.8.

An now comes the good news. Finally I repeated all
tests with latest gcl from Savannah. Now it works
fine! It seems that all problems are fixed. Actually maybe
not _all_ since I've noticed some strange things
in situation when I allocate some RAM and later allocate
some more RAM. But it have to do some more
thorough tests.

Vadim


--

[ Vadim V. Zhytnikov  <address@hidden>  <address@hidden> ]