[Axiom-developer] Re: CCL development

[Mike Dewar]

Hi Tim,

> First, I had no documentation of the language and it is clearly not
> common lisp. System commands, such as "save-system" and memory management
> commands (like setting heap and stack sizes) are the kind of things I
> NEED to know. I know these commands with GCL as I helped define them.
Its pretty close to Common Lisp but not a full implementation.  From
memory (I'm at a meeting in Italy on the end of a slow connection so
can't experiment) the heap size is a command line option, probably "-K".
However this turns off the default behaviour which is to grow the heap
as necessary so shouldn't be used except in special circumstances.  nAll
the command line options are handled in src-cslbase-csl.c.  The stack
size is fixed and I think depends on the system stack size (the single
most common Axiom bug report used to be that the stack blew out on some
machines like HPs where the system stack size was low - a problem we
never had with CCL).  

I apprecaite that you know AKCL inside out - we didn't and found CCL
more intuitive and also found Arthur very helpful.

> Second, I couldn't figure out how to debug code. The "break loop"
> commands were not obvious to me. There is no documentation on the
> structure of the runtime image or of the byte codes. These don't
> exist either for GCL but I spent years with it and know the GCL
> code well.
Debugging in CCL is poor, you basically have a back trace and thats it.
In Axiom commands can be found by going ")set break break <cr>1/0" in
Axiom and following the instructions.

> Third, there didn't seem to be a way to generate "intermediate"
> code.o files from the build process. This lack breaks the whole
> makefile tree and means I need to deeply reengineer the process.
In CCL code.o files make no sense so you can't build them.  You can
build "uxx.c" files which are C versions of lisp functions which you
link into the kernel.  This is documented (I think) in either the
caxbase or axbase src directories.
 
> Fourth, Several of the shipped algebra files will not compile in the
> NAG CCL image and I couldn't figure out how to find the failure.
I find this very surprising since they were rebuilt as part of our
product build cycle - this is probably a Linux problem.  Do you have an
example - if I can reproduce it in Oxford then I could try and talk you
through what I'd do to fix the problem?

> Fifth, the CCL image hangs if you run it long enough and I couldn't
> figure out why. It isn't swapping and it uses 99% of the cpu. (The
> easiest way to see this is to fire up the NAG axiom image and compile
> the algebra libraries). I tried to debug this also without success.
Again, I've never seen this and no user has reported such a bug.  It
sounds like a Linux problem again - bear in mind that we never ported
Axiom to anything newer than RedHat 6 and Linux releases tend to throw
up portability issues.
 
> Sixth, Size isn't the issue (anymore). Actually, it never was. I had
> Axiom running on a 16Meg laptop years ago. Portability is an issue and
Actually it is an issue - real-world computer algebra applications in
robotics and molecular modelling are too big to run in any existing
system.  

> I expect that CCL is going to be very useful there. Speed also is not
> an issue. Schelter built a special mechanism for Axiom that deeply
> optimizes function calling (the so-called .fn files). We had many
> discussions about it. Camm has also extended GCL with a function that
> allows me to see the assembler code generated by pieces of lisp so I
> can see exactly what is expensive (a more useful disassemble output).
Speed is an issue for some people - look at Faugere's work for example.
There are profiling facilities in CCL which you should know about - I'll
try and send you details when I'm back in Oxford.  
 
Somehow we need to ensure that all the useful but undocumented
information about Axiom and CCL stored in Themos and my brains is made
explicit, so please let me know about any problems you encounter.
Although I can't promise to be able to help I will certainly try.

Cheers, Mike.


________________________________________________________________________
This e-mail has been scanned for all viruses by Star Internet. The
service is powered by MessageLabs. For more information on a proactive
anti-virus service working around the clock, around the globe, visit:
http://www.star.net.uk
________________________________________________________________________