[Openmcl-devel] Bootstrapping problem 0.14.2-dev

[Gary Byers]

On Fri, 1 Oct 2004, Peter Bengtson wrote:

>  From reading the archives, it seems that sometimes the CVS
> bleeding-edge version is able to build itself, and sometimes not. Does
> anybody know if it indeed _is_ capable of building itself at this time?
>
> I have downloaded all the most recent stuff through CVS, installed the
> latest kernel from /testing, added the most recent headers, but when I
> do
>
> 	cd lisp-kernel/darwing
> 	make clean
> 	make
> 	openmcl

What does the "Welcome ..." banner say ? (or, if you'd prefer, what
does LISP-IMPLEMENTATION-VERSION say) ?

> 	(ccl:x-load-level-0 :force)
> 	(ccl:compile-ccl t)
>
> I get
>
> 	;Compiling "/Users/pjotr/lisp/ccl/lib/eval.lisp"...
> 	;Loading #P"/Users/pjotr/lisp/ccl/bindarwin/eval.dfsl"...
> 	> Error in process listener(1): Wrong PFSL version.
> 	> While executing: %FASLOAD
>
> Which seems to indicate that the fasl files OpenMCL produces can't be
> read by the very same version of OpenMCL.

That's certainly a mystery.  I'm mostly wondering why COMPILE-CCL thinks
that EVAL needs to be compiled (and why "ccl:lib;eval.lisp" hasn't been
removed from CVS, but that's less of a mystery.)

The full-blown evaluator - which existed primarily to support STEP, which
wasn't as useful as it could have been - got removed in 0.14.2, which
was released in early May.  The current version is 0.14.2-p1 (released
about a week later), and I believe that it -should- be able to compile
the bleeding-edge sources without major problems, but the warnings may
be starting to get scary-looking.

If you are using an image that's a version or two older, the first
mystery has a plausible explanation: if the file that was compiled and
loaded just before EVAL was "ccl:lib;nfcomp.lisp" (COMPILE-FILE),
you've loaded a current version of COMPILE-FILE into an older lisp.
(If the fasloader said "not only does this file have the wrong version
information in it, it seems to have come FROM THE FUTURE!!!", that
might - or might not - have made things clearer.

>
> Have I misunderstood something/made a horrible mistake somewhere?
>
> Thanks for any pointers/ideas/advice.
>

I -think- that an 0.14.2-p1 image should compile and bootstrap the
current sources, but I haven't tried in a while.  I did put a more
recent image in the testing directory a moment ago:

<ftp://clozure.com/pub/testing/dppccl-image-040925.tar.gz>

I don't think that things are as deeply out-of-synch between the
released binaries and the bleeding-edge sources as they sometimes
have been, but I think that the image above should compile those
sources without scary-looking warnings.

> 	/ Peter Bengtson