Re: [Axiom-developer] Axiom silver branch

[root]

> Well Axiom is doing similar things with GCL. Even more, GCL is part of 
> the Axiom archive. Hmmm...
> 
> 
> 
> I would feel better if if Axiom becomes something like a shell that 
> depends on several prerequisites. That would make Camm's work much 
> easier, and maybe we should simply learn from him. Yes, that would mean 
> for each version we have to say what is the minimum version of the 
> dependency object.

Check the email archive. Camm has explicitly said that it is ok to 
package GCL with axiom. And Richard Stallman said that it is ok to
package GPL code with Axiom given our distribution system. I don't
want to reopen these license discussions as they went on for months.

Camm occasionally gets bug fixes and patches from us. Why does it make
life easier to force him to upgrade when we want to change things?

> Well, but not all people use DEBIAN. How would other people be easy with 
> not having gcl and noweb in their Axiom distribution?
> 
> There are maybe two options.
> 1) We have two archives, one is the prerequisite thing and one is the 
> Axiom stuff we are actually focused on. Looks a bit similar to what we 
> have now, but Axiom would NOT have any Makefile target that deals with 
> building the prerequisites. Those Makefile things should be in the other 
> archive. So this would be different from the current situation.
> 
> 2) There is just one archive and this is Axiom without noweb, gcl and 
> any other additional useful package. AND there is an installation script 
> (Makefile or whatever) which is responsible for setting the stage so 
> that Axiom will compile without complaining. That script checks if gcl 
> is there in the right version, if not it would get it via "wget" from 
> the internet and "make" it. Same with "noweb" etc.

Please use the 'silver' package to experiment with these options.
Remember that your results have to work everywhere. The sourceforge
people maintain a compile farm which sourceforge developers can access
to ensure that their packages work on other systems. Please experiment
with this resource.

> 
> I feel it is somehow a disadvantage to have a need for an internet 
> connection for installation, but in some sense it is a bit like Debian 
> or other GNU/Linux distributions.

Try finding the 'right' libXpm.a file for FC5. Even if you find it
you might find that it is not installed where you want it and that
other systems install things in odd places. The BSD systems are
wildly different from linux. Windows has no hope of conforming
to your expectations. 

> 
> The real disadvantage is when you want to press Axiom on a CD. Of course 
> you must also provide the most important prerequisites like gcl and 
> noweb on the CD since you cannot expect internet connection then.

In fact, the Doyen project (sourceforge.net/projects/doyencd) which is
trying to make Axiom into a 'science platform' on a live CD has exactly
this problem.

> We clearly already have some prerequisite. However, could somebody say 
> what the prerequisites are apart from gcl and noweb? We rely on some 
> "standard linux distribution". But what is "standard"? Debian 
> dependencies show "libc6 (>= 2.3.2.ds1-4)" and "libgmp3". Interesting... 
> I don't find them in our "zips" directory. ;-)

We only package things that break. If you look at the zips directory
history you'll find that we no longer distribute the X library include
files. This used to be a problem historically but has been fixed in
all distributions and the file is no longer shipped. 

> 
> And what about the Axiom developers? Well, if they see that they are 
> missing gcl (./configure would tell them), they look at an AxiomWiki 
> page and follow the steps manually (until we have the script mentioned 
> above).

Even the developers post 'it didn't work for me' messages.

We put up a new debian server (thanks to Ralf) at Risc Linz last week
and Axiom will not build there. Apparently the problem is that emacs
is not installed but I always use emacs and I got the error message
in an emacs shell buffer so either that's not the problem or the
Debian depends: stuff is not configured properly.

I think you'll discover that almost every system is broken in some
obscure way and if you distribute code that requires the developers
to 'figure it out' every time you'll lose developers.

> 
> Ah, yes, wasn't there someone who wanted to write some autoconf/automake 
> stuff for Axiom? I would really appreciate that. It cannot hurt if we 
> learn about such standard tools. Am I wrong?

Check the email archive. The issue is that autoconf/automake assume
that you're building a C program. ASDF is the lispish equivalent.
There is not much use for a 'standard screwdriver' in a project that
uses nails.

t