path: root/TODO

$Id$
Public domain.  TeX Live TODO:

0. Remaining todo.
1. Package updates
   a. fix ctan2tds and friends
   b. outstanding package updates
2. Source updates
3. Infrastructure and installer
   a. tlmgr
   b. more
4. Random stuff
5. Won't do

0.* Forever.
==========================================
*** work on the items in the TODO file!
Consider items in the tracker: https://puszcza.gnu.org.ua/bugs/?group=texlive

0.1 After release.
===========================================
* retry untar if it failed after 3sec (configurable timeout) to make sure
  other programs have passed by

* Rewrite installer option_ handling that *always* the option name in
the tlpdb and in the $vars{'option_*'} agree.
Make initialization of those options from the tlpdb using
	$tlpdb->option('-installer' => 1)
not complain on the tlpsrc-set (installer only option) not complain


1a. fix ctan2tds et al.
-----------------------
(Aside from the whole script being a horrible kludge that needs
rewriting from scratch.)

tlprm and place do not know how to delete all emptied parent directories.

csplain cstex cs cslatex (and others) -- tlpkginfo should create the
directories, so that tlpkg-ctan-check can check them, instead of doing
some of the weird unpacking in ctan2tds.pl.

if ask for susy and have Susy, or reverse, complain.
In general, the package name field should match the tl directory.

place should run t1lint on pfb's, etc. (a la check-files-by-format).

place should complain about inconsistent line endings and
	case-insensitive clashes instead of leaving it for the nightly run.

1b. outstanding package updates
-------------------------------
check all mentioned in staw msg on tldev, 23 Apr 2008 23:56:50.

beebe et al. needs tons of work (see "sorry" in ctan2tds)
ethiop madness: combines ps-type1/ethiop and other packages.
eulervm vs. euler vs. new euler ... no eulervm.pdf
logic/gn-logic14 conflicts with logic font


2. Sources updates
==================
kpse: texmfdbs should not have to exclude non-!! dirs, should be able to
  go back to just TEXMF.  kpathsea should search the disk even if ls-R is
  present, without !!. (koch 29dec06)

dvips and all: snprintf, not sprintf, if any question



3. Infrastructure and installer
===============================
3a. tlmgr
* improve info <pkg> output for multiple repositories

* xetex conf.d postcode:
1. I'm sorry, but can you remind me why we have xetex/conf/conf.d/*.conf
   and why we copy them in our installation?  Neither XeTeX or LuaTeX
   do anything with these files, right?  Some of them refer to fonts
   and/or software that are not in TL.

2. According to my current understanding, we should actually be creating
   texlive-fontconfig.conf when luatex is installed, not just xetex.
   Although luatex does not use -lfontconfig like xetex does, it still
   reads the fontconfig configuration (specifically,
   /etc/fonts/fonts.conf) and thus to use system font name lookups in
   luatex on Unix, fontconfig stuff must be done.  Am I missing
   something?

* pinning
should we add a way to allow for numerical comparison of revision?


3b. More
- keep list of forcibly removed packages in tlpdb, so package-to-package
  dependencies can really be followed.

- _install_package in TLMedia and install_package in TLUtils appear to be
  doing basically the same job. Bad. Rewrite installer to use
  TLMedia instead of the TLUtils hacks.

- tlpobj vs actual files vs other packages installation, aka safer^2
  currently we just unpack what is in the tar file, but never check that
  the contents actually agrees with what is listed in the tlpdb. We not
  even check that the tlpobj contained in the container agrees with the
  tlpobj taken from the tlpdb. That allows to ship practically everything
  in the container, overwritting unrelated files.
  Also not checked is that unpacking the container does not overwrite
  files from a different package.

- extend revisions (Master/tlpkg/etc/generalize-revisions.diff)
  (implemented and working)
  we want: NNN[.RRR]
  - NNN as normal revision number
  - .RRR to provide various upgrades (think of context providing
    packages with everyday addint +1 to the revision number they could
    easily overtake upstream texlive)

- change tl-update-tlnet and tlpkg-by-size to implicitly get the network
  release directory (tldev, 2008) instead of hardwiring.  Then update
  doc/packages.txt.

infra: texmf/xindy not reported by check-file-coverage though not in tlp's

MAYBE TODO (needs discussion)
- ordering or collections removal (see below for details)
- (?) removal does not remove created format files from TEXMFSYSVAR
- after install/update show the number of bytes transfered 
  (email Rolf Niepraschk)

4. Random stuff
===============
stop using system(foo <bar >baz) (in tlmgr, installer, everywhere)
to avoid shell quoting problems; see dev/dev.sys_redir.pl.
Use list form of system if at all possible.

rewrite fmtutil, texconfig
	. in perl
	. work with multiple fmtutil.cnf files independently, maybe.

resolve the psnfss/fonts/metrics chaos
	see http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=433137
	(aren't there other more general debian bugs? utopia and fpl
	aren't the problem.)
With the metric files, the problem isn't so much the packaging as the
provenance.  We do not know where many of them came from, some
correspond to nonfree fonts, others are obsolete, etc., etc.  And yet,
removing any tfm at all will probably break documents and/or packages.
frank@kuesterei.ch's msgs to tex-live of 23 Oct 2006 19:19:28,
11 Oct 2006 13:32:20, and related begin to get at the problem.
... meanwhile, we could at least merge all the "font" packages
(avantar..zapfding) into psnfss, where really they come from on CTAN?
rk mail to tldev, 19 Sep 2009 01:06:34.

repo: make special repo backups
repo: memman.pdf, etoolbox.pdf, hyperref README.pdf -- svn guesses
      wrongly that it is text, resulting pdf is junk, pdffonts can't
      read it for example.  original on ctan is fine.
      1) check for other broken pdf's (done by check-files-by-format)
      2) change place to 
        svn propset svn:mime-type application/octet-stream
         for all *.pdf.
        (done in r11793 but comment says 'needs testing')
      3) report heuristic bug to svn.

doc: utf-8 fails accent chars with texlive-ge

tex: --no-mktex=fmt  should be supported
pdfopen: fixed http://magic.aladdin.cs.cmu.edu/wp-uploads/mavpdfopensrc.zip
dvips: comment line break http://groups.google.com/group/comp.text.tex/browse_thread/thread/b8ed25cc6153d5c
dvips: \special{map: ...} a la \pdfmapline (reinhardk 30dec06)
dvips: maybe stuff from dvipsfnt.txt from yy

iso: optimize image, http://www.tina-vision.net/tina-knoppix/remastering.html

sigh: lm vs. ec, 3jun06/stubner.gz msg

announce: slashdot.org, freshmeat.net, prweb.com, prnewswire.com ?

Things to do that karl noticed in TL'05, many still relevant :( --
      dvips psfonts.map != pfb /Fontname => courier
      qa: online mf
      qa: pdfetex -ini \&etex story sparc crash (sparc)
      qa: wrong /Length embedding, http://www.tug.org/pipermail/tex-live/2004-September/006755.html
      qa: wncyr10
      qa: english hyphenation in context!
      cdbuild/dupl.pl, 8-bit pseudoprint from olaf
      eitan real source for .tex (seb 10oct04)
      check that all pfb's are in psfonts.map
      kpsewhich mention which env names in debug output/if not found
      autoupdate .bib files
      fr hyphen test (from eurotex03 abstracts + de from hufflen?)
      output format initialized in log file, 16 Nov 2004 08:09:16 +0100 gaulle
      ps2ps/bbox for pdftricks, 15 Nov 2004 15:07:26 +0100 gerben
      dvips mapline support a la pdftex
      dvips +ufoobar.map -> no error?
      doc: bio of contributors, with pictures
      doc: tex-live.sty no include lmodern, for German
      special.map: handle like main maps (wl mail, 19 Jun 2005 01:21:15 +0200)

5. WON'T DO:
============
should only find package names at the appropriate levels, so that
graphics.tlpsrc does not have to include lines like
docpattern +!d texmf-dist/doc/latex/tufte-latex/graphics
-- No, the special cases involved are worse than the disease.

tex: implement fancier behavior at prompt file name, see knuth update notes
and thread at http://tug.org/mailman/private/tex-implementors/2008/001925.html
-- No, current behavior seems good enough, interactive use is minor.

use unzip 6.0?
-- Not until we see that there's some gain.

Create a "single user" installation option such that
TEXMFSYSCONFIG == TEXMFCONFIG and TEXMFSYSVAR == TEXMFVAR
so that updmap and updmap-sys would do exactly the same thing,
hence never conflict.
-- this would just add another layer of confusion and source of error.

Have BibTeX not require .bib extension, similarly not other extensions.
(suggestion from Bob Morris to karl, 10 Oct 2015).
-- no, because if "bibtex foo" found "foo" when there was no foo.bib,
it could easily be a binary. This has been a constant problem,
introduced from day one by the concept of "default extension". I fear it
is too deeply embedded in all TeX software by now to change.