@Rmano Thanks. I'll look into it. I suppose I could just try to figure out Wayland, but I suspect it isn't going to let me have different configs for different input devices.
I think the logic in pdftex.def can be improved adding something like the block here
\documentclass{article}
\usepackage{graphicx}
\makeatletter
\ifnum\pdfshellescape=0
\expanded{\def\noexpand\Gin@epstopdf@aux#1.\detokenize{eps}}.#2\relax{#1-eps-converted-to.pdf}
\def\Ginclude@eps#1{%
\...
@samcarter well no train was actually cancelled, only the train to frankfurt was delayed because of a known construction site. Happily the next train was delayed because of a technical problem and so I catched it and in the end got all the trains and seats from my reservation ;-). The only problem was that DB didn't know that everything worked fine and so didn't let me check-in automatically ;-)
@DavidCarlisle too many \noexpand ;-). I couldn't test yet, but shouldn't it error if conversion fails? Or is that error only in epstopdf?
@DavidCarlisle your fix doesn't work for me. If the -converted-to exists there is no error but also no picture, only a blank place.
@DavidCarlisle and if I don't use the extension .eps then it simply errors and doesn't try to look for the converted file (not sure if one should handle that case).
@UlrikeFischer yes I think that's Ok and consistent with if eps had never been defined at all for pdf (We could always add .eps to the default extension list but I think just doing the explicit eps is better here)
@DavidCarlisle and pdftex.def also mentions that one can suppress epstopdf with \newcommand{\DoNotLoadEpstopdf}{} and that should then probably be handled too as it switches silently to draft also with restricted shell-escape.
@JosephWright have meetings now but with active spaces couldn't you avoid the tl_replace_all and simply make them active on the way in, like accented characters
@JosephWright I tested it now with pdfresources, tagpdf and all of latex-lab and it seems to work fine. I only realized again that latex-dvips-tests are a pain -- I catched one error only by chance ;-).
@JosephWright well you could extend l3build with an option that outputs a list of tests which ended with an error ... (not a test failure, but a tex error) ;-) Then I would catch missing definitions faster ...
@JosephWright If I remember correctly it doesn't work on windows. But I need to check.
@JosephWright I pushed the current state of the pdfmanagement into develop, so you could update your version in latex3 and check if your tests against it.
Just wondering, why is this so bad:
\documentclass[a4paper]{article}
\usepackage{xcolor}
\begin{document}
\[
\left[\frac12\normalcolor\right]^2
\]
\end{document}
the ^2 is not placed correctly
I was using this inside a macro what might use \middle so I I cannot use grouping so was using \color{red} something \normalcolor and noticed that the placement of scripts in the \left/right was wrong
Hmm, publisher classes and their subtle coding mistakes. That one took a while to find. No idea what they thought effectively \leavevmode\mbox{\itshape} would end up doing....
@JasperHabicht str tokens are all catcode 12 or 10, and accented utf-8 in pdflatex requires catcode 13. So basically you can not prevent it, or to put it another way, do not use str for typesetting, use them for filenames or compuational string functions where you want the bytes directly
