XPost: linux.debian.maint.tetex
From: [email protected]

On 08.08.04 Benjamin BAYART ([email protected]) wrote:

Hi,

On a document of mine, too complex to be sent as a minimal example,
e-TeX seems broken with registers after 255. The following input:

minimal exmaple (just took your one and reduced it):

\documentclass{minimal}
\begin{document}
\makeatletter
\dimendef\z@=263\relax
\z@=0pt\relax
\show\z@
\end{document}

root@drachi:~# latex etex1.tex
This is e-TeX, Version 3.14159-2.1 (Web2C 7.4.5)
entering extended mode
(./etex1.tex
LaTeX2e <2001/06/01>
Babel <v3.7h> and hyphenation patterns for american, french, german, ngerman, russian, nohyphenation, loaded.
(/usr/share/texmf/tex/latex/base/minimal.cls
Document Class: minimal 2001/05/25 Standard LaTeX minimal class
) (./etex1.aux)

\z@=\dimen263.

l.7 \show\z@

?

same bug with Knuth TeX. So why do you think this is an e-TeX
problem?

My guess is that one of the "magical stuff" done to \z@ by amsmath
(and seems to be by other packages, and perhaps even by the LaTeX
kernel itself) makes e-TeX wrong.

Of course, the same example just runs fine in lambda (omega)
which also allows registers >255 (but have other bugs elsewhere).

Hmm: same code as above.

root@drachi:~# lambda etex1.tex
This is Omega, Version 3.14159--1.23.2.1 (Web2C 7.4.5)
Copyright (c) 1994--2000 John Plaice and Yannis Haralambous
(etex1.tex
LaTeX2e <2001/06/01> patch level 1
Babel <v3.7h> and hyphenation patterns for english, spanish, russian, greek, inuit, dumylang, nohyphenation, loaded.
(/usr/share/texmf/tex/latex/base/minimal.cls
Document Class: minimal 2001/05/25 Standard LaTeX minimal class
) (etex1.aux)

\z@=\dimen263.

l.7 \show\z@

?

unfortunately I don't understand so much about (La)TeX internal
registers and their limitations...

H.
--
sigmentation fault


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

Hilmar Preusse writes:

On 08.08.04 Benjamin BAYART ([email protected]) wrote:

On a document of mine, too complex to be sent as a minimal example,
e-TeX seems broken with registers after 255. The following input:

minimal exmaple (just took your one and reduced it):

\documentclass{minimal}
\begin{document}
\makeatletter
\dimendef\z@=263\relax
\z@=0pt\relax
\show\z@
\end{document}

root@drachi:~# latex etex1.tex
This is e-TeX, Version 3.14159-2.1 (Web2C 7.4.5)
entering extended mode
(./etex1.tex
LaTeX2e <2001/06/01>
Babel <v3.7h> and hyphenation patterns for american, french, german, ngerman, russian, nohyphenation, loaded.
(/usr/share/texmf/tex/latex/base/minimal.cls
Document Class: minimal 2001/05/25 Standard LaTeX minimal class
) (./etex1.aux)

\z@=\dimen263.

l.7 \show\z@

?

You misunderstood the nature of the problem. Using a variation of the
original example:

$ cat etex-bug.tex
\documentclass{article}
\usepackage{amsmath}
\begin{document}
\makeatletter
\dimendef\z@=256\relax
\z@=0pt\relax
\[aa\]
\show\z@
\end{document}
$ latex etex-bug.tex
This is pdfeTeX, Version 3.141592-1.20a-rc7.1-2.1 (Web2C 7.5.3)
output format initialized to DVI
entering extended mode
(./etex-bug.tex
LaTeX2e <2003/12/01>
Babel <v3.8c> and hyphenation patterns for english, french, german, ngerman, ca talan, croatian, czech, danish, dutch, estonian, finnish, greek, hungarian, ita lian, latin, mongolian, norwegian, polish, portuguese, romanian, russian, slove ne, slovak, spanish, swedish, ukenglish, ukrainian, welsh, dumylang, nohyphenat ion, loaded.
(/usr/local/share/texmf-dist/tex/latex/base/article.cls
Document Class: article 2004/02/16 v1.4f Standard LaTeX document class (/usr/local/share/texmf-dist/tex/latex/base/size10.clo)) (/usr/local/share/texmf-dist/tex/latex/amslatex/amsmath.sty
For additional information on amsmath, use the `?' option. (/usr/local/share/texmf-dist/tex/latex/amslatex/amstext.sty (/usr/local/share/texmf-dist/tex/latex/amslatex/amsgen.sty)) (/usr/local/share/texmf-dist/tex/latex/amslatex/amsbsy.sty) (/usr/local/share/texmf-dist/tex/latex/amslatex/amsopn.sty)) (./etex-bug.aux)

\z@=\count652.

l.8 \show\z@

?

The significant bit is that it prints \z@=\count652 instead of

1) Complaining about a "Bad register code" (which plain TeX would have
done.
2) Printing \z@=\dimen256 like omega or aleph does.

This is evidence that e-TeX's internal datastructures are messed up.

In e-TeX registers 0..255 are handled as in TeX, but the
larger-numbered registers are allocated in a sparce array, and
deallocated when there is no longer any reference to them. What
appears to happen is that the reference counting doesn't take into
account that references already exist to \z@ because it is being
redefined instead of seeing a pristine definition. So the register
gets deallocated and the memory reused. This is something I'll have
to punt to the e-TeX maintainer (who's on vacation at the moment).

The amsmath code is needed just to provide enough "churn" to actually
make this problem manifest itself in the example.

--
Olaf Weber

(This space left blank for technical reasons.)



--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

tags 264394 + upstream
stop

On 11.08.04 Olaf Weber ([email protected]) wrote:

Hilmar Preusse writes:

Hi all,

root@drachi:~# latex etex1.tex
This is e-TeX, Version 3.14159-2.1 (Web2C 7.4.5)
entering extended mode
(./etex1.tex
LaTeX2e <2001/06/01>
Babel <v3.7h> and hyphenation patterns for american, french, german, ngerman, russian, nohyphenation, loaded.
(/usr/share/texmf/tex/latex/base/minimal.cls
Document Class: minimal 2001/05/25 Standard LaTeX minimal class
) (./etex1.aux)

\z@=\dimen263.

l.7 \show\z@

?

You misunderstood the nature of the problem. Using a variation of
the original example:

I was afraid of this. Thanks for clarification. Is this bug that
serious, that we have to assume e-TeX to be useless and to use better
Knuth TeX instead?

The significant bit is that it prints \z@=\count652 instead of

1) Complaining about a "Bad register code" (which plain TeX would have
done.
2) Printing \z@=\dimen256 like omega or aleph does.

This is evidence that e-TeX's internal datastructures are messed up.

OK thanks.

So the register gets deallocated and the memory reused. This is
something I'll have to punt to the e-TeX maintainer (who's on
vacation at the moment).

So can we assume, that you forward the problem to the e-TeX team?

Many thanks for you efforts. I'll tag that bug upstream for now.

Regards,
Hilmar
--
sigmentation fault


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

Hilmar Preusse writes:

I was afraid of this. Thanks for clarification. Is this bug that
serious, that we have to assume e-TeX to be useless and to use better
Knuth TeX instead?

This has been in e-TeX for a _long_ time. I believe the idiom needed
to trigger to be rarely used (and only e-TeX-specific code can use it
at all).

So can we assume, that you forward the problem to the e-TeX team?

Yes.

--
Olaf Weber

(This space left blank for technical reasons.)



--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

Le Wed, Aug 11, 2004 at 04:14:38PM +0200, Hilmar Preusse:

I was afraid of this. Thanks for clarification. Is this bug that
serious, that we have to assume e-TeX to be useless and to use better
Knuth TeX instead?

upstream maintainers will probably deal with it: the LaTeX-project team
plan to use e-TeX for future LaTeX upstream releases, so if it is
broken, there will be work to fix it before releasing a new LaTeX
version.

Olaf, about your comments on the bug itself: in my minimal example, \z@
was re-defined, so your explaination might be possible. But in the full
initial example, \z@ was initially (as far as I can see) defined as
\dimen263: I'm working on a very large document, which loads dozens of packages, and thus need more than the 256 registers in TeX, so I did a
small patch on the LaTeX kernel to make it use register <255 for inserts
and >255 for everything else. It works smoothly with Omega (did not yet
try aleph) but triggered this bug in e-TeX.

So can we assume, that you forward the problem to the e-TeX team?

Many thanks for you efforts. I'll tag that bug upstream for now.

Thanks for your efforts both of you.

Benjamin.



--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

On 14.08.04 Benjamin BAYART ([email protected]) wrote:

Le Wed, Aug 11, 2004 at 04:14:38PM +0200, Hilmar Preusse:

Hi,

I was afraid of this. Thanks for clarification. Is this bug that
serious, that we have to assume e-TeX to be useless and to use
better Knuth TeX instead?

upstream maintainers will probably deal with it: the LaTeX-project
team plan to use e-TeX for future LaTeX upstream releases, so if it
is broken, there will be work to fix it before releasing a new
LaTeX version.

For the time being you may switch back to Knuth TeX. Read NEWS.Debian
how to do that.

H.
--
sigmentation fault


--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

Benjamin BAYART writes:

Olaf, about your comments on the bug itself: in my minimal example, \z@
was re-defined, so your explaination might be possible. But in the full initial example, \z@ was initially (as far as I can see) defined as \dimen263: I'm working on a very large document, which loads dozens of packages, and thus need more than the 256 registers in TeX, so I did a
small patch on the LaTeX kernel to make it use register <255 for inserts
and >255 for everything else. It works smoothly with Omega (did not yet
try aleph) but triggered this bug in e-TeX.

Can you supply the patch and an example that shows the bug in
combination with it? If nothing else, this can be a sanity check that
whatever fix is made actually fixes the underlying issue.

--
Olaf Weber

(This space left blank for technical reasons.)



--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)

XPost: linux.debian.maint.tetex
From: [email protected]

FYI, I believe that the problem can be worked around (at the expense
of leaking some memory) by changing .../texk/web2c/etexdir/etex.fix to
read as follows:

---------------8<---------------8<---------------
% Bugfixes for etex.ch version 2.1
% Public Domain.

@x l.1163 - |int_val| is a level, not a value.
begin cur_val:=int_val;
@y
begin cur_val_level:=int_val;
@z

@x l.2737 - typo
else q:=sa_ptr(cur_ptr)
@y
else q:=sa_ptr(cur_chr)
@z

@x l.5503 - comparison with 0 instead of |null|.
if cur_ptr<>0 then
@y
if cur_ptr<>null then
@z

@x l.5514 - do not deallocate even when reference count goes to 0.
begin decr(sa_ref(q));
@y
begin decr(sa_ref(q));
return;
@z
---------------8<---------------8<---------------

--
Olaf Weber

(This space left blank for technical reasons.)



--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]

--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)