Re: [Axiom-developer] Re: [Axiom-commit] SF.net SVN: axiom: [508]branches/wh-sandbox

[Waldek Hebisch]

Bill Page wrote:
> On April 27, 2007 6:32 PM Waldek Hebisch wrote:
> > ... 
> > svn diff between wh-sandbox and build-improvements is probably
> > quite large, mostly due to file removals.  If we do not take
> > into account files which are entirely (or almost entirely)
> > removed and algebra files (where wh-sandbox removed bootstrap
> > Lisp) we get a diff of 780 kb (20456 lines).  The bulk of this
> > is in interp directory (about 11000 lines).  The second large
> > piece is graph subdirectory (about 3000 lines).  The next is
> > new asq program.  Other changes are relatively small.
> >
> 
> Although everything else is very nice, I think the single
> must significant achievement is wh-sandbox is the new algebra
> bootstrap method and the removal of the Lisp from the algebra
> files. In my opinion this is of great importance to the future
> of Axiom.
> 

I think content of algebra should go almost as-is to build-improvements.
ATM wh-sandbox uses different paths, but that is very easy to adjust.
I plan to add some explanations to Makefile.pamphlet (old
explanations are _very_ out of date).  Besides new bootstrap
procedure (which is contained in Makefile.pamphlet) algebra
has a namber of bug fixes and Martin's guessing package.  However,
there is some (weak) dependence on changes to interp. 

> > ... 
> > Both graph and asq should be independent of other changes.
> > 
> > Well, enough for today: I will summarize other changes later.
> > 
> 
> What about the changes to HyperDoc?
> 

HyperDoc changes are actually a few small fixes, same in interp,
some in lib, some in hyper.  In hyper Makefile.pamphlet may be
controversial:  IMO  '${HYPER}/pages/ht.db' rule in wh-sandbox
is more readible.

Let me repeat (and update) summary of main changes in wh-sandbox:

1) removal of unused code from compiler/interpreter
2) translating depsys using Shoe and removal of depsys bootstrap Lisp
3) new algebra bootstrap and removal of algebra bootstrap Lisp
4) mechanical generation of (most of) .pht pages
5) hyperdoc and algebra fixes
6) removal of useless files
7) pretty-printing of intermediate files and asq rewrite
8) adding guessing package written by Martin Rubey
9) prototype conversion in graph
10) safety/portability fixes in interp
11) making variable initialization explicit in interp

I think that main problem is merge of interp -- this is bulk of
changes and they are not entirely mechanical.  There are some
easy ones: bootstrap Lisp in c-util.boot.pamphlet and in 
interop.boot.pamphlet is unused, so this code in removed from
wh-sandbox.  OTOH Gaby converted to Shoe a bunch of files that
were earlier converted in wh-sandbox -- this created a conflict.
There were few fixes to interp files in build-improvements
revision 485 I think that I incorporated them all.  But recent
mass edits for Shoe covertion create conflicts in some files.

So there is some work to do in interp, especially due to Shoe
covertion:  my experience is that such changes while seemingly
trivial must be carefully validated, and this takes time.

There is some potential for controversion in interp/Makefile.pamphlet.
The version in wh-sandbox IMHO is simpler and has more functionality
than version in build-improvements.

I think that we can quickly agree about trivial or uncontroversial
changes (graph, algebra, some interp cleanups including removal
of bootstrap Lisp from c-util and interop).  If we agree on
design decision in interp then we will have to resolve conflicts
in recent work.

Gaby, now you should say how you imagine merge, in particular
if you object to some changes made to wh-sandbox (it looks
that you attach special importance to AX_FLAGS)?


-- 
                              Waldek Hebisch
address@hidden