Arch Linux

is this still an issue?

Re-using this issue because:
With `pgkctl repo clone texlive-texmf` as of 2023-06-15,
the fmtutil script `/usr/share/libalpm/scripts/texlive-fmtutil` fails.

Warnings output begins with:

```
(5/7) Updating TeXLive format files...
fmtutil [ERROR]: no (or empty) hitex.fmt made by: hitex -ini -jobname=hitex -progname=hitex -etex -ltx hitex.ini </dev/null
```

To investigate, I ran the reported hitex command manually.
Turns out that etex expects hyphenation files. Lots of them.
I had to install all of the texlive-lang* packages except *arabic, *japanese, *korean to get the hitex format built again.

The reason for those language requirements seems to be that the two (initially identical) files

/etc/texmf/tex/generic/config/language.def
/usr/share/texmf-dist/tex/generic/config/language.def

from package `texlive-basic` includes all those hyphenation files.

That seems to call for a pacman hook to autogenerate the language.def similar to the fmtutil hook.

After installing all the *lang packages except the above-mentioned three, one warning by fmtutil-sys remains:

```
fmtutil [WARNING]: inifile fmttriggers=cm,hyphen-base,knuth-lib,plain for aleph/aleph not found.
```

and installing the remaining three texlive-lang* packs does not help to get rid of that warning.

The aleph warning seems to be due to a wrong entry in
/var/lib/texmf/arch/installedpkgs/formatsextra.fmts
owned by the package texlive-formatsextra.

Partial solution: Let a hook do

```
perl -pe 's{^\\addlanguage\{[^{}]+\}\{([^{}]+)\}}
{(!`kpsewhich $1` && "%").$&}e' \
/usr/share/texmf-dist/tex/generic/config/language.def \
> /etc/texmf/tex/generic/config/language.def
```

This reads the /usr/.../language.def, comments-out the commands with nonexisting hyphenation files. and writes the result to the /etc/.../language.def.

This seems to work for plain hitex formats.

Next stop: hilatex formats. Those use a `language.dat` instead.
Those are updated by `tlmgr install hyphen-xxx` and `tlmgr remove `hyphen-xxx`.
We do not use `tlmgr`, I suppose?

For the language.dat, the corresponding hook command is:

```
perl -pe 's{^\w\S*\s+(\S+)|^=}
{($1 ? ($c = !`kpsewhich $1` && "%") : $c).$&}e' \
/usr/share/texmf-dist/tex/generic/config/language.dat \
> /etc/texmf/tex/generic/config/language.dat
```

This reads the /usr/.../language.dat, comments-out the languages with nonexisting hyphenation files, also comments out the subsequent alias definitions, and writes the result to the /etc/.../language.dat.

This works. I have been able to uninstall the unneeded texlive-lang* packages.

The commands for adapting language.def and language.dat have to run AFTER mktexlsr and BEFORE fmtutil-sys.
Since the hyphenation patterns mostly affect formats, it would make sense to incorporate these steps directly into `/usr/share/libalpm/scripts/texlive-fmtutil`.

Oh, and the hook trigger patterns should be updated to rerun those scripts when language packs have been installed or removed.
(Some language packs do not trigger the existing fmtutil hook.)

I have attached a patch suitable for `git am`.

FIXME: Although the language.def and language.dat in /etc/... have
a comment that says not to edit by hand and use tlmgr instead,
this could previously be ignored because pacman would leave those
files untouched. Now the hook scripts overwrite them.

A better solution would be to mimic what tlmgr would do to those files.
Presumably tlmgr allows local tweaks somehow.

Thanks. I have implemented a different fix in 2023.66594-10 based on the existing formats/maps splitting mechanism.

Closing this old issue, please post any follow up comments in the current bug report  FS#78799