@DavidCarlisle thanks. I've never used latex-dev. I think I should better wait ;).

what does ^^M mean in logs?

@UlrikeFischer and @DavidCarlisle -- (sorry for late report; I've been occupied with some urgent matters) That visible text is pretty not good in some quite consistent ways, e.g., "a line" instead of "align"? The audio is better. (I know the audio is what is really the purpose, so that's a positive.) I'll try to take another look later.

@egreg -- this looks like one for you: tex.stackexchange.com/q/737701

@cfr we really do ask all package authors to use latex-dev, that way they find if their package breaks 6 months before a latex update not the day after. You don't need to do any special install, just use pdflatex-dev file instead of pdflatex file

@cfr same as if you use it in a document, it's control m, character 13, newline

@barbarabeeton well the input examples are ours why don't you like them? We can adjust of course. The tex file is ww.github.com/latex3/tagging-project/project-examples/align/…

@barbarabeeton oh sorry by visible text you meant the closed captions not the typeset text. The closed captions are nonsense, I didn't add any so YouTube ai is just auto transcribing the audio, I'll try to disable that next time, the captions are not in the file they are being added as you view,

@DavidCarlisle I might do that. I just don't want to make a mess, which I will if I try to test putting things in the right places. I'd rather keep them in the wrong places ....

@DavidCarlisle well, it isn't a newline-type newline or I assume it would be a newline.

@barbarabeeton I simply can't understand how having \acs{i} in a formula could help readers.

This is LuaHBTeX, Version 1.21.0 (TeX Live 2025)
 restricted system commands enabled.
(./test.tex
LaTeX2e <2025-06-01> pre-release-2 (develop 2025-2-23 branch)
L3 programming layer <2025-01-18>

@cfr l3build has a --dev option to run tests with the dev version of the formats - so you can quickly 'peak'

@cfr I thought the issue was about getting or not getting the right TS1 subset when you ran a document so it wouldn't be putting any files anywhere?

@cfr well it is a normal newline but the question is how did you get a normal newline into the data being logged as normally they are parsed as space or \par or some such not as a character token for character 13.

@DavidCarlisle ctan.org/incoming

@JosephWright but not yet in texlive 2024, or did you get there already?

@UlrikeFischer incoming empty, but I assume that was tagpdf? :-)

@DavidCarlisle yes, I just uploaded (I need to figure out how to make the test better, currently it fails on github as the date of the embedded file changes ...).

@UlrikeFischer moan at @JosephWright that l3build isn't normalising the date?

@UlrikeFischer Not yet for my mirror, no

@DavidCarlisle Sounds reasonable :)

@UlrikeFischer Example line?

@JosephWright I basically forgot to set the ModDate to a fix value, but that is perhaps something we could do in regression-test?

\ExplSyntaxOn
\pdfdict_put:nne  {l_pdffile/Params} {ModDate}{(D:20200422135851Z)}
\ExplSyntaxOff

Reminds me that @cfr asked about normalising lines ...

@UlrikeFischer I think ATM that would go in a local config - feels like we should be refining the interface a little to me

@JosephWright well actually it should perhaps go into l3pdfmeta, I have a regression-test command there..

@UlrikeFischer if you set the EPOCH settings for reproducible builds does the moddate default to whatever you set or does it still get the real date?

@UlrikeFischer Wouyld be good

@DavidCarlisle epoch settings works fine for the main document, but when I embed a file I get its date, so I should overwrite that.

@UlrikeFischer oh it's a file you checked in, you are not writing it from within the epoch setting, OK

@DavidCarlisle and now the date is fine but the checksum is still wrong grumpf ;-(

@UlrikeFischer dos line ends?

@JosephWright does windows have the equivalent of touch so l3build could offer a platform neutral way of setting the mod date of test data ` touch -d 19700101 p.log` makes the file have

$ ls -l p.log
-rwxrw-r--+ 1 davidc davidc 7734 Jan  1  1970 p.log

@DavidCarlisle Sort-of: copy <filename>+ would update the modified date

But to 'now'

@JosephWright yes but "now" is what you want to avoid in reproducible builds:-)

@JosephWright hmm google says to use this but I don't think we want to that:-) You can manually change the Last Modified Date/Time for a file using a free software called Attribute Changer from petges.lu.

oh you can do it via powershell with admin perms (Get-Item hello.txt).lastwritetime=$(Get-Date "8/28/2018 2:35 am") but that still seems more than we'd want, maybe @UlrikeFischer's right and just forcing the metadata to not match the file is simpler

@JosephWright oh lfs provides touch:

@UlrikeFischer

$ git commit -a -m "root-supplemental-file=latex-align-css.html"
[main 2624dd9] root-supplemental-file=latex-align-css.html
 3 files changed, 45 insertions(+), 41 deletions(-)
 create mode 100644 project-examples/align/latex-align-css.html

@DavidCarlisle that would work :)

@UlrikeFischer @barbarabeeton updated video without the auto-generated closed captions and with continued-row for wrapped lines youtu.be/YyOI4MyJzNg

@DavidCarlisle Oh no, now it is not accessible for deaf people.

@mickep I did actually start to hand correct the closed caption text but it's a lot of work as it's almost all wrong to start with. Possibly it should be possible to automate re-adding the actual text that mathcat generated for soeech as closed captions but I'm not sure how to re-sync the timings. The nvda speech viewer would giv eme all the text I think.

@DavidCarlisle Probably possible, but sounds like head ache.

@mickep also the videos aren't really the point just a demo of the screen reader experience and there the closed captions aren't an issue

@DavidCarlisle I understand that. (I was not too serious)

@mickep yes i guessed your initial comment was at least half in jest but as I had spent some time wondering about that this morning I decided to answer it at face value

@DavidCarlisle You should spend more time finding bad ic examples in luatex. :(

Hi, a quick question. I’d like to see HarfBuzz the default in luaotfload, is there a possibility of this ever happening. Lets say I have a time to work on the hyphenation issue (I don’t currently) and I mange to get it in an acceptable state, would there be any other blockers to making harf mode the default?

Also, would there be any interest in allowing harf mode for math fonts? HarfBuzz would be used only to load the font and get data from the MATH table, but math typesetting would still be done by the engine as usual. I’m hoping using harf mode would allow me to avoid issues like github.com/latex3/luaotfload/issues/278

@KhaledHosny I think anything's possible modulo compatibility issues (but latex has some standard ways of mitigating those) but you'd be best to discuss it with Marcel who's not usually here, and an issue at the github luaotfload repo would reach everyone that needed reaching

@DavidCarlisle Makes sense, thanks

@KhaledHosny Isn't one issue it's not as good for western European scripts as the standard Lua setup?

@JosephWright That is why I’m asking. I’m aware of the inadequate handling of hyphenation, but I’m not aware of other issue with these scripts.

@KhaledHosny Well there's the whole business that ConTeXt is never going to use HarfBuzz, so the standard setting as-is means that LaTeX and ConTeXt do the same thing ...

@KhaledHosny Could just be that - I only know that currently one can't say that HarfBuzz is always better

@JosephWright I was not aware that matching ConTeXt behavior is a concern, but if it is then that is one answer to my question.

@JosephWright I guess then asking on luaotfload issue tracker is where one gets a definitive answer as David suggested above.

@KhaledHosny I'm not sure it is, but I can see a change leading to queries from the interested users (people like @mickep who use ConTeXt)

@KhaledHosny Yes, best place to discuss I agree

@KhaledHosny I'll look out for the issue :)

@KhaledHosny and @JosephWright I maybe read too quickly, but I do not see it being a problem for ConTeXt if LaTeX defaults to HarfBuzz.

@JosephWright github.com/latex3/luaotfload/issues/300

Is is possible to use latex-dev with a GitHub workflow?

@JasperHabicht yes, I'm doing that e.g. in tagpdf.

@UlrikeFischer Nice, I’ll check

So essentially it is checkformat ="latex-dev", right?

I‘ll try this. Thanks!

@JasperHabicht you perhaps need to set that up, see the build.lua, but yes.

@JasperHabicht Or l3build check --dev

@JosephWright Ah, that’s probably even better as I might be able to set up a second workflow

Looks like it works

@JosephWright ah, thanks. I could test the current code by 'peaking', then. but I still would have to change the code to test the TS1 thing.

@DavidCarlisle it's about getting the right subset if the subset is declared in the .fd but not the .sty. so I can test cfr-lm with -dev but it doesn't test that because I put the declarations back into the .sty. I wrote something to test the TS1 thing but it isn't part of any package. I just wrote it to figure out what was going on and file the issue on github.

@DavidCarlisle yes. I don't know where it is coming from, only that it is engine-dependent. luatex doesn't do it - only pdftex and xetex.

@cfr ah yes, Ok

@JosephWright public things and hooks ...

@cfr the output form of non ascii characters is setable try pdftex --8bit for example then i think you'll find it logs ^^M as a literal newline or you can use --translate-file and specify a mapping explicitly

@cfr

\documentclass{article}

\begin{document}
\typeout{====}
{\catcode13=12 \typeout{aaaa^^Mbbbb}}
\end{document}

@cfr pdflatex gives aaaa^^Mbbbb but pdflatex --8bit the ^^M gets printed as a newline (expected) and aaa vanishes (less expected)

@DavidCarlisle what does \SHOWFILE do?

@cfr it copies the file to the log so it shows up in the test (see texdoc l3build) which has all the regression-test test commands

@DavidCarlisle sorry. the rest got lost because my system froze.

\IfFileExists{#1}%
  {\begingroup %
     \catcode`\^^M\active %
     \edef^^M{^\string^M^^J}%
     \everyeof{\noexpand}%
     \obeyspaces %
     \@sanitize %
     \message{\@@input #1 }%
   \endgroup }%

@DavidCarlisle so are they expected for pdf/xetex if I use \SHOWFILE?

@cfr if there are MSDOS line endings in the original then possibly yes, they are ^^M^^J pairs in the original but tex normalises all line endings on input so depending how you copy it ^^J is treated as the line end and ^^M is left hanging.

@DavidCarlisle that seems odd. the files I'm inputting are just written on a previous run. so the test compiles 3 times (rather than 1) and reads files generated on the second run into the log for comparison. there shouldn't be any msdos?

\RequirePackage{svn-prov}
\ProvidesFileSVN{$Id: prooftrees-memoize-t1.lvt 10540 2024-10-29 04:41:14Z cfrees $}[v0.0 \revinfo][\filebase: memoization test]
\input regression-test.tex\relax
\documentclass{article}
\START
\TEST{prooftrees-memoize-t1-mmz}{%
  \IfFileExists{prooftrees-memoize-t1.mmz}{%
    \TYPE{prooftrees-memoize-t1.mmz:}%
    \NEWLINE
    \SHOWFILE{prooftrees-memoize-t1.mmz}%
  }{%
    \TYPE{CANNOT FIND prooftrees-memoize-t1.mmz}%
  }%
}
\OMIT
\usepackage[extract=python]{memoize}
\mmzset{%

but I don't understand at all what is happening. I don't even understand where the luatex.tlg comes from.

oh, l3build-check creates it. this was not obvious to me from the doc. I read that as saying the log output was normalised and not that a standard .tlg was also transformed for luatex ...

but I still don't understand why I get ^^M explicitly in the logs for pdftex/xetex.

@cfr what os are you on?

@DavidCarlisle linux

@cfr hmm.

@cfr vvv

the file written out doesn't have dos line ends but \SHOWFILE shows as ^^M anyway, I suspect @JosephWright is to blame, but it's been a long day fighting with css and I'm not sure I want to trace this

@DavidCarlisle it is not expected?

@cfr oh sorry yes it's exactly the code that you posted earlier, it's showing each newline with an explicit ^^M

@cfr the code is explicitly typing ^^M it isn't a newline being shown as ^^M by TeX's built in mechanism for showing control characters, it really is the three characters ^ ^ and M so it's expected by "somebody"

@DavidCarlisle I more meant: this is not something \SHOWFILE is expected to do as a feature?

@cfr yes that's what I tried to answer. Whether it is "expected" depends who is doing the expecting. Looking at the code it's clearly intentional, but I wasn't expecting it.

@DavidCarlisle ah, ok. thanks. I wasn't sure. but it then gets normalised away for luatex, I think.