LATEX-L Archives

Mailing list for the LaTeX3 project

LATEX-L@LISTSERV.UNI-HEIDELBERG.DE

Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Condense Mail Headers

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

Print Reply
Sender:
Mailing list for the LaTeX3 project <[log in to unmask]>
Subject:
From:
Frank Mittelbach <[log in to unmask]>
Date:
Thu, 21 Oct 1999 21:01:30 +0200
In-Reply-To:
Reply-To:
Mailing list for the LaTeX3 project <[log in to unmask]>
Parts/Attachments:
text/plain (144 lines)
David Carlisle wrote in reply to James Kilfiger:

 > > but I get errors when I put instances in an \edef.
 >
 > The number of commands that work in an edef is vanishingly small;-)
 > Probably (instances of) templates could be (or perhaps already are!)
 > robust via the normal latex \protect mechanism, in which case they
 > should survive the latex equivalent of \edef which is
 > \[log in to unmask]

instances of templates should be robust if they are not we should make them or
at least their call through a standard mechanism like \UseInstance should be
robust.

this needs checking but is not significantly difficult to provide. more
difficult might be to make a the use of \UseTemplate robust since that might
contain arbitary code in its key/value segment.

 > There is no way to make an instance of a template expandable in general
 > as the key setting in the template call are typically low level register
 > assignments in TeX, and they are not expandable.

this answer that template instances can't be expandable (with reasonable
effort --- disregarding that (i think) Alan Jeffrey once proved that one can
implement everything in TeX's mouth if one is prepared to wait long enough for
the answer (this is really the turing machine argument to please don't argue
that machines get faster :-)

but since the question of why does a command that is claimed to be "robust"
fails in an edef and we had a similar "bug report (pr/3084)" just recently i
appended my reply back then below since i think this is of some interest.

questions:

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

 b) if the current scheme is basically sound (and i think it is) would be
 trapping \edef a sensible approach, ie a plain edef would effectively run
 \protected@edef and a code writer who really would like to use a vanilla
 \edef for speed reasons would have to use something like \unprotected@edef?
 This would clearly help in cases like 3084 or in James' usage. but it would
 slow down processing a lot if packages and the latex kernel are not adjusted
 accordingly (the kernel currently contains 175 \[xe]def's in places where we
 thought we know what the input is going to be --- all such places would
 suddenly no longer run a primitive but a macro which in turn does some
 additional processing)

 c) there are some theoretical possibilities for a command to find out on its
 own whether or not it is in a dangerous context (such as \edef) but i haven't
 figured out a single solution so far that would run in acceptable time
 (assuming that about every command would need to make this check over an over
 again this is a significant overhead compared to the current approach where
 the meaning of \protect is externally changed so that there is no or nearly
 no penalty for testing the current state for the individual command)

 d) is the current concept okay except for the problem that one needs to
 educate people better about the use and misuse of \edef compared to
 \protected@edef ? this is the current approach taken and somewhat the
 opposite to b)

if somebody would have ideas for a) or more specifically c) which is a
subclass of a) i guess this would be great.

if somebody would have a heart and would try to test b) on a large number of
documents that would be great too and report about timing differences and
failures found; essentially it means running the document with a format that
contains additionally something like

\let\@@edef\edef
\def\protected@edef{ <as before but with \@@edef instead of \edef> }
\let\edef\protected@edef

and ditto for \xdef and \protected@xdef and perhaps a few other commands in
that part of the kernel


if somebody would feel like writing a tugboat article on the current \protect
i guess the TUB editor would be glad and so would i and probably a lot of
people since this is a poorly understood concept (there is a bit of intro docu
written long time ago in ltdefns.dtx but i unfortunately didn't got very far)

have a go ---  ( not just silently listen to this list )

frank


---- from pr/3084; see www.latex-project.org for the whole pr


Subject: Re: latex/3084: New add@accent definition prohibits \edefs
Date: Fri, 17 Sep 1999 11:31:38 +0200

Bernard,

 >  >The new definition of add@accent is supposed to keep correct
 >  >spacefactor's after accented uppercase letters. The cost is
 >  >that a command like \edef\agrave{\`a} will not work as
 >  >expected (with respect to spacefactors).
 >  >More importantly, this makes the babel package fail,
 >  >and possibly others.

you are right that the new definition of the internal accent handling made
babel fail or rather made its use inside a straight \edef fail.

problem is that LaTeX *never* supported arbitrary input inside straight
\edef's.

LaTeX goes a long way to make most commands robust (and if fragile provides
\protect to protect them). However to make this work, LaTeX commands who
accepts arbitrary input in their arguments are never allowed to pass them
straight to an \edef. Instead they have to pass them to \protected@edef which
defines \protect suitably to make everything work as expected. (note that in a
straight \edef \protect does not help you as its default definition is
\relax!)

within standard LaTeX the new defnition of \` and friends still work perfectly
in all circumstances. the fact that inside babel \edef was used in a place
where \protected@edef should have been used was an oversight and is corrected.

of course that doesn't help if other packages have commands that receive user
input and pass it to \edef. but there is nothing that one can do about this
other than fixing those packages. the fact that the accent commands are now
failing inside such commands is a pity, but i don't think the conclusion can
be to keep them broken in other respects just because there are potentially
some broken packages that do not implement the protect mechanism of LaTeX
correctly. it is simply a fact that to get the space factor right in case of
\`A one has to make an assignment and assignments are not expandable in
TeX. so to make them work you have to avoid putting them into a primitive
\edef.

and even if they would not have that restriction and would work inside \edef
there are plenty of other commands that have the same kind of problem inside
an \edef. so the conclusion can only be:

    DONT USE \edef unless you absolutely control its content.

as i said, babel is now fixed again in this regard.

hope this explains the general problem behind the anomaly you detected.

best wishes
frank

ATOM RSS1 RSS2