Message-Id: <9306201839.AA10545@sc.zib-berlin.dbp.de>
Reply-To: Mailing list for the LaTeX3 project <LATEX-L@vm.urz.Uni-Heidelberg.de>
Date:         Sun, 20 Jun 93 20:33:55 +0200
From: Frank Mittelbach <MITTELBACH@MZDMZA.ZDV.UNI-MAINZ.DE>
Sender: Mailing list for the LaTeX3 project <LATEX-L@vm.urz.Uni-Heidelberg.de>
To: Multiple Recipients of <LATEX-L@vm.urz.Uni-Heidelberg.de>
Subject:      adding fake registers to ltx3
Status: R


Phil remarked about my comments on the limitations of internal TeX
registers and my proposal to extend them a bit:

> I am sure --- certain, even --- that you have a very good reason for
> providing a fixed number of fake counters, but may I urge you to
> re-consider this decision?  one of the fundamental problems of TeX,
> as you well know, is the small, fixed, number of certain entities
> which it provides: in the light of this experience, would it not make
> sense to allow an a TeX-extension module such as LaTeX-3 an infinite
> number of all entities, limited only by the available storage space,
> and interchangeable in that if, overall, an implementation allowed
> $n$ words of space, this $n$ could be partitioned among the fake counters,
> fake dimens, fake families, etc, in a totally flexible way?

In principle I'm in favor of this idea but we are not talking NTS here
but real life TeX programming. For example, there is *no* way to
extend the 16 families in the current TeX except by reimplementing
TeXs internal math handlings by hand within TeX.

Instead of keeping the values for fake registers in main memory which
would be slow and also needing a lot of it, I'm thinking of providing
global dimens, counters and skips (?) by using one or more dummy fonts
with no chars but a huge number of font dimens. This is very efficient
in space and fast. In principle you could even extend it to a general
allocation routine which loads the same font over and over again at
slightly different sizes, thus providing as many new "registers" as
necessary.

It is not clear to me whether or not a fixed upper boundary is enough
(this would be compile time constant of the format since you can say
something like \fontdimen10000\nullfont=0pt if you do it early
enough).

Of course, these registers are all global, and therefore
not usable in all situations. Another way of getting at local fake
registers at a reasonable speed, is to use a medium number of mu-skip
registers. This would be definitly limited to approx 200 but at least
they would be local.

I see no way to extend box registers or insertions and I do see no
need to have so many mu-skip registers.

Does this my remarks a bit better?

Frank