I'm back (I think :-)...
Wed, 13 Sep 95 23:00:34 PDT
I am still quite interested in the idea of a Registry for TDS names, but
I never got much (read, ANY) reaction to my specific proposals from the
TeX wizards out there. SPQR has asked me to jump back into the fray,
however, so I will.
As I see it, there are two problems a Registry can and should solve:
1. It should document TeX-related packages in sufficient detail
that a prospective user can find out (a) what the package is
good for, (b) where it can be obtained, (c) who maintains it,
et copious cetera.
This problem has been addressed in at least three organized ways:
a. The PTF "0.doc" files and HTML pages. (The latter may be
seen on www.ptf.com, and will appear on the next issue of
"Prime Time TeXcetera".)
This format is not optimized for machine-readability, but
we might be able to translate it into another which is...
b. The DRAFT specification of the IETF IAFA working group,
which defines a format for indexing information found on
Internet FTP archives. This is used by the ALIWEB server.
c. The "LSM" files used by the Linux community.
The latter two are essentially equivalent, for our purposes.
The IAFA may be more tightly specified; the LSM is certainly
more widely used. I have no strong preference in the matter.
2. It should unambiguously define the portion of the TDS name
space occupied by the package.
My initial notions (see my earlier postings) started with the
idea of specifying an explicit set of directory paths. I also
suggested that some form of wild-carding (regular expressions)
could be used to shorten the description. Finally, I said that
it might be possible to give (a) a name and (b) a package "type"
and let default rules establish which directories would be used.
In retrospect, I think I prefer a merger of these three ideas.
The package author defines the type (and paths) by listing the
top-level directories across which the name is claimed. (The
doc directories corresponding to the mainline directories would
also be reserved.)
I think this has the advantages of safety, simplicity, and
generality. If I lay claim to fonts/morin/martian/, I
should reserve any directories my font could reasonably use.
Here is a simplistic set of examples:
/ dvips, web2c
I suspect that a TeXnician could boil these down further. It might
be, for instance, that "plain" should be unique within "/". Let me
know about stuff like this, please... Also, would someone who has
built a reasonable-size TDS tree be willing to look into it and tell
us how much chaos this scheme would cause?