LATEX-L Archives

Mailing list for the LaTeX3 project


Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
"Michael J. Downes" <[log in to unmask]>
Reply To:
Mailing list for the LaTeX3 project <[log in to unmask]>
Wed, 27 Oct 1999 11:12:09 -0400
text/plain (239 lines)
Frank Mittelbach <[log in to unmask]> writes:

> questions:
>  a) anybody having any idea whatsoever to deal with this problem of robustness
>  other than LaTeX currently does? (conceptually i mean)

There are various reasons for the expansion done by LaTeX on
user-supplied text, but there is (I think) only one purpose for which
the expansion is really indispensable:

  Getting the current element number when preparing to make a running
  head or toc entry.

I.e., when copying material for multiple use.

[When the problem is posed in this way it can be seen that another way
around would be to write some kind of \ref-like string instead of the
actual number, assuming that the generation of labels could be suitably
automated---and that string-pool/hash-table limits were not a problem.]

The expansion done by \mark (for running heads) and \write (for table of
contents) does not need to be applied to the title of a chapter or
section, only to the section number---assuming that the user does not
use commands in the titles that change their meaning during the course
of the document. (If the chief reason for the current \protect mechanism
is only to allow this seldom-used possibility then the cost is too high.
There are other ways to reach the same ends. The switching from, e.g.,
\chaptername to \appendixname is the main counter-example here I think,
but that can, and probably should, be taken care of by a separate
mechanism. Some steps in this direction can already be seen in

There is a well-known difficulty with \write, however. Non-immediate
\write for writing .toc or index information does full expansion on its
argument *at the time the page is shipped out* in order to get the
correct page number. But no assignments can be done at this time, only
expansion. Therefore the standard token-register method for keeping text
unexpanded can't be applied to section titles. (For index terms the
\index command reads the data with verbatim-like catcodes since the data
does not need to be printed at its initial location, only written to the
.idx file.)

I guess what we really need is an e-TeX feature \unexpandedwrite that
writes material without expansion, then a section title could be sent to
the aux file by (a) pre-expanding the element number (b) applying
\unexpandedwrite to everything except for the page number. [Maybe this
is already in e-TeX? I don't remember.]

However in the absence of an \unexpandedwrite primitive we can put the
title in a macro and write the \meaning. I think this works for any
material that doesn't include actual # tokens (which would get doubled),
given that multiple-use text like section titles and figure captions
already have the restriction that they cannot have internal catcode

I did a test file for this, see below. The idea of the test file is that
that end-users never have to use \protect unless they put something
fragile into a numbering macro like \thesection.

I wrote "there is only one purpose for which the expansion is really
indispensable"; \MakeUppercase might be cited as another application,
but I think we really ought to look strongly into the possibility of
making .vf files for uppercase fonts and ensuring that they get widely
available, instead of continuing with the present edef mechanism (which
is mainly done to insure that \aa gets expanded to \r{a} so that the "a"
can be uppercased). Most applications really need \MakeTextUppercase, to
skip over math and \ref and stuff like that, where the difficulties of
the edef solution become more apparent.

Michael Downes

\title{Expansion tests}
\author{Michael Downes}

\makeatletter \catcode`\_=10 \catcode`\ =9 \catcode\endlinechar=9

% First let's fix up the uses of \mark, that is easier than the toc stuff.


  \xdef\@themark{\the\@currentmark}% For compatibility



% Need a better solution for \MakeUppercase.


% It would be better if \addcontentsline had a separate argument for the
% number. Lacking that, we have to scan "by hand" for \numberline.
% Typical usage:
% \addcontentsline{toc}{chapter}
%   {\numberline{chapnum}Title of this here chapter}

% Here is a better alternative for \addcontentsline:
% \WriteTocEntry{toc}{chapter}{\chaptername}{\thechapter}{Title of chapter ...}



% The \addtocontentswithpage function is similar to \addtocontents but
% provides more separation between the parts that need different
% expansion.

% For toc entries made with \WriteTocEntry we need something that
% handles more arguments than \contentsline:
  \csname l@#1\endcsname{\numberline{#2\enspace #3}#4}{#5}

%% % Simplified, for testing:
%% %
%% \renewcommand{\numberline}[1]{\begingroup \bfseries#1.\endgroup\quad }
%% \renewcommand*\l@chapter[2]{%
%%   \par
%%   \addpenalty{-\@highpenalty}%
%%   \addvspace{6pt plus1pt}%
%%   \begingroup \parfillskip0pt
%%   \noindent#1\nobreak\hfil \nobreak\hb@xt@\@pnumwidth{\hss #2}\par
%%   \endgroup
%%   \penalty\@highpenalty
%% }

\newcommand{\fragilecommand}{\relax\ifmmode Math\else Text\fi ?}

\makeatother \catcode`\_=8 \catcode\endlinechar=5 \catcode`\ =10


\chapter*{Preface \begin{math}\fragilecommand\end{math}}
Some prefatory text.

% Trying to send a fragile command to the toc without \protect:

\chapter{Title of a Chapter \begin{math}\fragilecommand\end{math}}

Some text.

\chapter{And Another Chapter \begin{math}\fragilecommand\end{math} Here}
\markboth{Other Chapter \begin{math}\fragilecommand\end{math}}
         {Other Chapter \begin{math}\fragilecommand\end{math}}

Some text.

\caption{Here is a figure caption containing some fragile
  stuff: \begin{math}\fragilecommand\end{math}.}

\bibitem{A} Joe Author, \textit{Some title}.
\bibitem{B} Terry Berry, \textit{Another different title}.