[tex-k] Problem with latxmk

[John Collins]

Hi Karl,

> mktexpk runs mf in the cwd (I believe), but that's not writable when
> running with TEXMF_OUTPUT_DIRECTORY. So it needs to cd to the output-dir
> first. Also the "computation" of $PKDESTDIR needs to take the output-dir
> into account. Probably it should just leave everything in the output-dir
> itself; trying to construct a texmf tree there doesn't seem sensible >
> That's my guess. I haven't tested it. Likely I got it wrong.

Actually mktexpk (and presumably the other mktex scripts) changes directory to 
a process-specific temporary directory.  See texmf-dist/web2c/mktex.opt lines 
38-44 and lines 62, 63.  (That file is sourced by mktexpk.)


>      I'm able to restore correct functioning by inserting
>          unset TEXMF_OUTPUT_DIRECTORY
> 
> I would argue that if --output-directory=foo is specified, the correct
> behavior is for mktexpk+mf to write under foo, not to the normal
> $TEXMFVAR / whatever location. --thanks, karl.

I don't think so.  The output directory is something that the user has 
specified to pdflatex (etc) for the run-specific output files.  As a user, I 
think I expect automatically generated font files to be placed in their 
standard locations (under TEXMFVAR) independently of the setting of the output 
directory.  I think that is especially true if the fonts themselves (or their 
source files) are provided as part of the TL 2024 distribution, as with those 
in the ifsym package.

If I understand the motivation correctly, TEXMF_OUTPUT_DIRECTORY is set so as 
to communicate the output directory to programs that are called by the \write18 
mechanism.  I think we are just seeing an unintended side effect of that.

Maybe the appropriate place to unset TEXMF_OUTPUT_DIRECTORY is in 
texmf-dist/web2c/mktex.opt, since that is common to all the mktex scripts, and 
is setting directory names.

John