Barak,
This is interesting, and it works in my simple tests. I'll get back to you
later with further comments. How much tweaking of the code did you have to do
after the github copilot produced its results?
Meanwhile, here are some small points:
1. The -parallel=num option isn't parsed correctly. Currently, line 2306 is
if ( $1 =~ /^\d+$/ ) { $parallel_jobs = $1; }
The problem is that the pattern matching operation resets $1 to undef, so the
assignment to $parallel_jobs is not what is intended; I was getting warning
messages about an uninitialized value. Just change that line to
if ( $1 =~ /^(\d+)$/ ) { $parallel_jobs = $1; }
and everything will be fine. (There are now parentheses around the matched
pattern, so that $1 is reinitialized, correctly.)
2. Stdout and stderr are redirected in the forked processes so that the outputs
doesn't get mixed up. But the *latex programs also sometimes get input from
stdin, particularly under error conditions (unless -interaction=batchmode is
used). Potentially the same could happen with other invoked programs I think
that can be solved by redirecting stdin to be from /dev/null.
3. The methods used don't work on Windows. So under Windows, I would give a
warning if $parallel_jobs isn't 0 or 1 after the command line is parsed, and
then set it back to zero.
4. Would you mind changing the version identification and keeping the date
updated. I'm calling my test version 4.88.parallel. It makes it a lot easier
to identify what's being run, especially when comparing different versions or
getting bug reports. (Note that by default, latexmk outputs the version
information near the beginning of its work, which makes the output self
documenting as to the version in use.)
Best,
John
On 3/23/26 4:24 PM, Barak A. Pearlmutter wrote:
I asked github copilot to add a latexmk -parallel[=n] option and ate
dinner and came back and tested it and it seems to work fine. First it
just allowed various top-level .tex files to be processed in parallel.
Then I asked for other stuff to be concurrent too, and also to stop
mixing up outputs and buffer them instead, and it did that. Then I
told it I was worried about errors from interaction of multiple
subjobs generating the same intermediate file, so it added some
flock'ing. And I tested it some more and it seems to work fine.
https://github.com/barak/latexmk branch copilot/add-parallel-option-to-latexmk
"If you can dream it...!"
--Barak.
