Message-Id: <9311152008.AA00526@sc.ZIB-Berlin.DE>
Reply-To: Mailing list for the LaTeX3 project <LATEX-L@vm.urz.Uni-Heidelberg.de>
Date:         Mon, 15 Nov 93 21:07:29 +0100
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:      NFSS math handling suggestion (was puzzle)
Status: R


> >  > The puzzle is: All the user intended to use was 7 + 8 + 8 + 8 =31
> >  > fonts. Why is it that NFSS assigns so many when nobody asked it to?
> >  > Wasn't the idea that all fonts were assigned "on demand"?
> >
> > They are. But all math fonts need to be loaded before math mode is
> > entered, remember? So, you are effectively asking for a much larger
> > set of fonts, namely (9 + 9) * (1 + number of math fonts per size).
>
> No Rainer, you haven't done your homework. Math fonts need not,
> repeat, NOT be assigned BEFORE you enter math mode.
>
> > Sorry, but we are using TeX here and have to be content with what it
> > offers.
>
> I am positive that this is a limitation of NFSS and not
> of TeX.

I wouldn't call it a limitation but a design decision of NFSS. This is
the old question of time versus space (mostly, but see below).

If you compare the number of fonts loaded in lfonts.tex with those
loaded in NFSS running on a document written for lfonts.tex you will
find that the fonts loaded are in fact smaller (provided you haven't
preloaded everything beforehand).

TeX is a turing machine and thus you don't even need the mathchardefs
but if you don't use them you have to program around that filling up
main memory quite fast. I assume that you don't disagree with me so
far.

This means that every symbol font needs to be loaded when you start
math mode since you can't predict if it is going to be used and in
what size unless you disable mathchardefs. However, you can of course
supress the setup until the point you actually typeset math (which, I
agree is not done).

As for the math alphabet, we decided not to preload them, however, if
a math alphabet is used it will then automatically be useable in other
sizes. This again is a time decision. If we start up the machinery of
turning an unassigned math alphabet into a font assigning math
alphabet everytime it is used the speed you go done very much. Keeping
tables of the sizes in which it is assigned and in which not is not
very promising either. Without changing that decision (which i think
is correct) you can still save by defering the setup until math is
actually used.

> the solution is plain and clear here: Use \everymath and \everydisplay
> I agree that about four math families need to be assigned
> (families 2 and 3 are a must) even if they are not used,
> but this is a far cry from what NFSS does, namely assign
> all fonts that are currently in the "math alphabet" every time
> one changes sizes, even when one does NOT enter math mode.

this is certainly a possibility. however, it can have strange side
effects some packages use \everymath for their own purposes. thus one
has to be a bit careful and needs to provide NFSS with a private
version of those tokenregs that can't be clobbered by other packages, eg
something like

\let\nfss@everymath\everymath
\newtoks\everymath

\nfss@everymath={<some code>\the\everymath}

this is certainly a possibility could be in fact produce better
results in terms of both in time and space. but i don't think that one
should alter the other design decisions of NFSS, ie

- load all symbol fonts declared
- load all alphabets used

as they would produce worse limitations.

> Well, that's like saying what do you care how big a program is.
> For starters, there are only 255 TFM files that any document van load,
> and for anybody who knows how to count, the number that NFSS actually
> uses comes dangerously close to the limit.

I'm a bit surprised that you claim to reach this limit (which does not
mean that the above suggestion isn't valid). The only type of
application where this is likely is the direct use of \fontsize for
arbitrary sizes using ps fonts or starting with a large number of
preloaded fonts in the first place.

>  In fact, with the
> "small" implementations of TeX, that test document cannot be
> processed, because the limit is 128. I'd say that's a problem
> but our friend Joachim doesn't think so.

well, my implementation runs this with

 35814 words of font info for 121 fonts, out of 90000 for 255

granted that this close to the limit of 128 (and worse than it could
be) it it is not above and it is certainly not near 255. but it is
already quite a heavy document as far as the number of math alphabets
used is concerned.  even by adding another six alphabets i don't get
to the limit (well, except the tex limit of 16 fams):

 72703 words of font info for 187 fonts, out of 90000 for 255

so, there is something wrong with your setup.

nevertheless, the suggestion is worth trying out, however, probably
not before next year, ie not for the first release of LaTeX2e. While I
think that this actually improve things in several directions I don't
consider the type of test document really a problem requiring
immediate action.

Frank Mittelbach