Cannot decode byte (Data.Text.Encoding: invalid UTF-8 stream) when using "--log" option

[HiS-Brahms]

Explain the problem.
Pandoc fails with pandoc: Cannot decode byte '\xfc': Data.Text.Encoding: Invalid UTF-8 stream when using the "--log" option. This only happened on one specific file, based on which I produced an example (see files). This also only fails when using the "--log" option, if we dont use it, it works and produces a PDF.

example-broken.log
example-broken.md

We use the eisvogel.tex.txt template. (github doesnt like *.tex files apparently ...)

Command to reproduce:

pandoc example-broken.md -o example.pdf --from markdown --template eisvogel.tex --log=example-broken.log

Strangely, splitting the sentences up into own lines produces no error:

example-working.log
example-working.md
example-working.pdf

I also made sure that the file was indeed utf-8 (based on the documentation) with iconv -t utf-8 but it produced the same output.

It also makes no difference if the locale is en.DE-UTF-8 or en.US-UTF-8.

Pandoc version?
We can reproduce it both in

• fedora 42 toolbox (running in latest kinoite), pandoc version 3.1.11.1
• pandoc docker image (docker.io/pandoc/extra:latest sha256:25e2c7bdde7a637bfcc7d5f3e58bc7e65567afe46275b16f8f347df325a59a80), pandoc version 3.7.0.2

[jgm]

It's extremely strange that adding the paragraph breaks makes this go away.
I note that I can also reproduce it by using --verbose instead of --log.

[INFO] [makePDF] LaTeX run number 1
[INFO] pandoc: Cannot decode byte '\xfc': Data.Text.Encoding: Invalid UTF-8 stream

This suggests that the problem comes when pandoc tries to read the log output produced by the first LaTeX run.

[jgm]

Probably coming from the conversion at https://github.com/jgm/pandoc/blob/main/src/Text/Pandoc/PDF.hs#L322-L324