Bug#225833: 225833: letter vs A4 in TeX
Hi Norbert,
I'm not sure what your opinion is: I read the beginning of your post as
"everything is fine already", but in the end you seem to agree that
changing the system-wide geometry.cfg is good?
Norbert Preining <preining@logic.at> wrote:
>> > OTOH, if you add the geometry package it selects automatically the pdftex
>> > driver and the a5paper option from the documentclass is executed, so:
>> > \documentclass[a5paper]{article}
>> > \usepackage{geometry}
>> > and pdfinfo gave me A5 paper as intended. As it should be.
>>
>> Yes, but why can we get this magic with PDF, but not with DVI?
>
> Because geometry.sty *automatically* slects the pdftex driver if it is
> run via pdf(la)tex. This is the point, it *DETECTS* that it is running
> in pdf mode and uses the pdf "specials".
Well, what's the alternative to PDF? It is DVI mode. Which specials
should geometry use for DVI mode? Theoretically, the answer is
"geometry cannot know this without a driver option". However, in
practice, the answer might be "dvips specials, because there's no driver
in use that uses other types of papersize specials".
I'm not 100% sure about the part after "because". But even if 10% of
our users who use DVI mode would use something else, doesn't it make
sense to give the 90% a sane default? And I very much doubt that dvips,
dvipdfm and dvipdfmx together make less than 90% of the use cases.
>> > \usepackage[driver=dvips]{geometry}
>>
>> And I would actually *not* like to encourage people to hardcode this
>> driver selection in their files. Too likely they also do it for
>
> Or educate them to use the *right* driver.
IMHO, it's better to not have driver specifications in the LaTeX source.
At least for graphicx and hyperref and if you do not use dvipdfm this is
the preferred way in the newsgroups. geometry is particular in doing
*nothing* when it doesn't get a driver in DVI mode, whereas graphix (I
think) and hyperref (I know) choose *something*: This gives the users
of dvips working files, and users of dvipdfm(x) or anything else at
least get decent (albeit verbose) error messages about non-interpretable
specials. With geometry, you just get no errors or warnings, but wrong
paper.
>> > Finally, for pdflatex, use the option driver=pdftex
>>
>> That's a superfluous option, IMO it should be deleted from geometry,
>> since it does not make a difference whether it is specified or not.
>
> I don't think so! It is only that geometry.sty detects pdftex and
> activates this feature. Well, in this sense it is useless ;-)
What do you think is the use of pdftex? What would be different without
it?
> So the only question is: DO we do it as distributors or do we put a
> comment in the README file.
If we agree that dvips is a sane default for >90% of the users (and for
sure it is, if we add those who never use DVI mode), then I think we
should set this default, and not just document it.
> Furthermore, if we do it as distributers, do we change the file
> /usr/share/texmf-texlive/....
> or do we put a file into
> /etc/texmf/...
> I would prefer the FIRST option to change the system wide geometry.cfg
> and inform the users that they can override these defaults ...
The first option IMHO violates the Debian policy. If we use
geometry.cfg for a site-wide configuration (and what else is this?),
then geometry.cfg is a configuration file which needs to be in /etc and
get dpkg's conffile protection, or similar maintainer script handling.
Regards, Frank
--
Dr. Frank Küster
Single Molecule Spectroscopy, Protein Folding @ Inst. f. Biochemie, Univ. Zürich
Debian Developer (teTeX/TeXLive)
Reply to: