Re: [Axiom-developer] Re: ASDF and ./configure

[Gabriel Dos Reis]

David MENTRE <address@hidden> writes:

| Hello Tim,
| 
| I looked at the ./configure script so I can reply more specifically
| now. 
| 
| root <address@hidden> writes:
| 
| > The issue, as i understand it, is not that the configure script
| > doesn't work on the systems where it was tested. The issue is that
| > the configure script is not robust. Certain problems have to be
| > decided.
| 
| I understand, this is the issue faced by all ./configure scripts.
| 
| You have two options to make it more robust:
| 
|  1. use the ./configure generated by autoconf tools. There is a lot of
|     knowledge in them, but most of this knowledge is not needed for
|     Axiom:
| 
|     - who needs to build Axiom on outdated systems like Ultrix?

The value of using Autoconf for Axiom is not in building it on or for
Ultrix, but in reusing packaged knowledge about various regular systems,
taken care of by system packagers.  Axiom really should be developing
its own idea of config.guess or system triplet.  The last thing I
would hate Axiom to exhibit is the the NIH symptom as far as the build
and packaging mechanism is concerned.

|     - most of autoconf knowledge is related to C issues, while most
|       Axiom issues are Lisp ones (once GCL is built, and GCL is using
|       autoconf),

this is true, as long as, one ignores that a configure script
generated from Autoconf can contains *custom* variation either
directly in configure.ac or in the macro aclocal.m4.  I contend that
if you can write the configure script by hand, then you should
take advantage of Autoconf by letting it generate configure
automatically and customizing the generation through aclocal.m4.  I
also claim that is you write configure by hand, then you will end up
reinventing Autoconf functionalities with more bugs -- put in here
usual wisdom about lisp functionalities and programs written in other
programming languages.  

|     - when the generated configure script fails, it is a nightmare to
|       understand *what* is going wrong;

I have used, build, and maintained "large" systems with Autoconf and
the above matches the opposite of my experience.  Quite often, I've
found that when the build, installation fails or is brittle, it is
because someone has tried to avoid Autconf and ended up with
reinventing the wheel with more bugs.

[...]

|  - the equality of strings is "==", not "=" (as far as I know);

ahem, in which shell?  Even configure generated by Autoconf
exclusively uses "=".  Autoconf has been widely test in more wild
environment than anything variation you'll come up here.  I would
suggest that we avoid NIH and reuse the wisdom of the Autoconf guys
who have faced and experienced wild shell variation issues.

-- Gaby