[Top][All Lists]

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

[Gcl-devel] Re: [Axiom-developer] gcl 2.7.0 and "ready"

From: Camm Maguire
Subject: [Gcl-devel] Re: [Axiom-developer] gcl 2.7.0 and "ready"
Date: 20 Jun 2007 11:47:11 -0400
User-agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2

Greetings, and thanks for the interest and the kind words.

As far as 2.6.8pre goes, this will be released when 

1) We can satisfy Vadim/read-char-nohang on windows:
2) We port to intel mac

I'd like to refrain from ansi work on 2.6.x if possible, though I will
put in ensure-directories-exist for the people who've requested it.

2.7.0 has much, much more work behind it, and is where the bulk of my
    effort on GCL is spent.  The basic changes:

1) two word cons -- 50% more cons space at the expense of extra
   overhead in type_of
2) immediate fixnums and vastly accelerated fixnum arithmetic in a
   range determined at compile time ((INTEGER -268435456 268435455) on
   32bit x86 Linux), again at the expense of more type_of overhead.
3) gmp random numbers, gcd functions, and many more bignum
4) src inlining.  GCL keeps a compesssed src string of every loaded
   compiled function, and uses source inlining to improve
   type-propagation and branch-elimination, as well as simplifying the
   compiler significantly.
5) auto proclamation/recompilation -- no more need for sys-proclaims
6) mutual recursion detection and tail-recursion implementation
7) many, many ansi fixes
8) apply optimization
9) unboxed complex arithmetic
10) dynamically-generated persistent 1 instruction calls to external
   shared C library functions 
11) libm function integration
12) faster safety mode
13) fork-based parallel processing, p-and, p-or and p-let

The changelog in debian/changelog contains the details.

In general, I'd like to release 2.7 when (hopefully by September):

1) Paul's ansi suite passes without error.  There are many more tests
   in the 2.7.0 branch.  This suite effectively defines ansi
   compliance for me.
2) performance is at or better than 2.6.x for all applications
3) The extra compile time induced by autorecompilation is under

Things I would like to do:

0) centralize on Schelter's funcall descriptor
0a)call closures with the same semantics as normal functions
0b)decide on a complex thread-safe call or a simple call using globals  
1) support slime
2) improve native debugging, including Schelter's dbl.el emacs mode
3) complex short-float and complex double-float array types
4) optional blas interface
5) robustify parallel processing
6) build the mpi extension by default in the Debian package
7) support apply/&rest -> funcall list elimination
8) extend the src-inlining coverage to car et.al. and aref et.al. (One
   can define safe source which when inlined at safety 0 has the
   checks automatically removed.  Ask me if interested in the details.)
9) provide lisp implementations for as many functions now written in C
   as is practical.
10) clean up dead C code written into .c source files
11) refine source inlining algorithm
12) compile top-level closures
13) get a simple useable bechmark suite in place to quickly spot
    performance degradation.
14) re-liberate the dpans info documentation

As you are interested in ansi there really are only a few fundamental
issues left:

1) (setf foo) function descriptors need implementing as
2) symbols in common-lisp package
3) print/format
4) special declarations

Paul had summarized a todo in the BUGS file.  His contributions have
been immeasurable.  I haven't heard from him in some time, and hope he
is OK.

Take care,

Stephen Wilson <address@hidden> writes:

> Camm Maguire <address@hidden> writes:
> > I'm trying to release in September.  What is really needed right now
> > are testers which can understand and deal with some transient
> > breakage, but still provide useful feedback as to the general
> > direction and needs of the release.
> Hello Camm,
> I have been working as of late with the ansi build of gcl-2.6.8pre.
> Without reservation, I must say I am impressed.  This is truly
> excellent work!
> I would be very happy to contribute to GCL, and in particular to its
> advance towards ANSI compliance.
> I have noticed several issues which I feel I could address, but do not
> understand the priorities of the GCL team for purposes of releasing
> 2.7.0.  Could you provide a brief description of what you mean by
> `general direction and needs of the release'?
> Thanks,
> Steve

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]