Open Problems

Joachim Schrod
Tue, 19 Sep 1995 12:32:56 +0200 (MESZ)

Paul wrote:
> > The latter. The draft gives the impression as if texmf/tex// might be
> > a good (i.e., valid) TEXINPUTS path. That's not true, filenames are
> > only unique over the trees tex/{<format>,generic}/. Come to think of
> > it, perhaps this should be added as an explicit requirement.
> On page 8 it says, ``a recursive search beginning at \path|texmf/tex|
> is a correct path for {\TeX} inputs in a \abbr{TDS} tree.''

Ooh, even worse.

> Are you saying
> that this is incorrect? 

Yes. And that's independent of TDS. There have been cases in the
past, where two different files with the same name are part of two
different macro packages. The most prominent example is LaTeX 2.09
vs. LaTeX, but there are other ones, too. (E.g., there were *.sty
files for Vanilla TeX in PC-TeX distribution that were in conflict
with the LaTeX files.)

I.e., *for TeX*, it is not valid to search all possible locations of
macro files for any format, as there may be (will be) name conflicts.

If one assumes sensible names for EPS files, there should not be any
name conflicts -- except if two lusers introduce a logo.eps. ;-)
Therefore I would assume that texmf/tex// is a valid (and good)
search path for DVI drivers if they need access to such files. (Of
course, slow without caching.) If programs need to locate macro
files, they must now the format for it.


PS: More and more I come to the point that the discussion of `there
are no unique macro file names' should be part of the rationale.

Joachim Schrod			Email:
Computer Science Department
Technical University of Darmstadt, Germany