@egreg by happy accident @cfr was not here. but for what?

@AlanMunn I really have no idea. it is not their usual list. I'm not even sure exactly what they want to axe in languages ... especially since we're supposed to have this fantastic 'languages for all' thing.

@mickep I see ;).

is it expected that extracting a page from a pdf created with luatex and a unicode font will fail if gs is used for the extraction? more specifically, it does not like the annotations and discards the ToUnicode mappings completely. a type1 font or a unicode font with xetex works fine.

@mickep thank you for checking. I guess @egreg already knows.

@cfr Oh no, is he reading my emails? :)

@mickep that was my assumption.

@mickep everything I have tried with luatex over the past few days has involved it misbehaving (from somebody's perspective or other). at least this is consistent regardless of font. all the others involve a difference when using otf versus type1, but I think these are definitely bugs, even if I don't know in what.

@cfr Well, I don't know much about luatex, but if you encounter problems that you suspect are bugs, I think it is good if you report them (maybe on luatex list). You do not need to debug yourself, to come up with a patch or so. But a small example of what you do and what you expect and what you get. Either you do something wrong and then someone will likely point you in the right direction, or you found a problem, and then it might get fixed.

After all, the luatex developers are active...

@mickep true, but it is more interesting to try to figure out what is happening. and if half a dozen different things might be at fault, nobody else has much reason to diagnose anything. at least one of them is almost certainly not in luatex or luaotfload.

the other one, though. I have no idea. (but I'd be inclined to blame gs first and something else only later, even if it is luatex/otf specific.)

@cfr Post deleted?

@cfr I think it is the one I am replying to, @JosephWright.

@cfr Well, if you play and find something that is a problem and could be useful for others to get resolved, it is nice to report...

@cfr You don't follow rugby, do you? 🤭🏉

@mickep Bad decision, IMHO. Maybe you can propose that \tracinglostchars=4 would detect missing characters when a box is being built?

@mickep sounds quite odd. Where does the width and height of the box then comes from?

@UlrikeFischer it uses 0 size as far as I can see

@egreg Before you judge, consider that there might have been a reason for this. (I am not going to push for something in luatex, that is better sent to one of the mailing lists.)

The fact that this question/issue comes up after 15 years also says something about how important it can be...

@DavidCarlisle It uses the .notdef width, height and depth are zero.

@UlrikeFischer It is odd. If one doesn't use the box, they'll never know about a missing character in it and the data obtained are random.

@egreg I suspect Hans' thinking is that if one doesn't use the box, the fact glyphs are not there is really not important

@JosephWright Building a box without using it is done for measuring things…

@egreg Sure, but typically there will also be a use of the same glyphs - it's odd to not need to typeset the result at all

@egreg ah ok

is there a script to check/set the correct file permissions in the texlive installation folder?

日本語TeXレッスン...

@egreg Maybe Hans can actually implement it. But I think only if the LaTeX team promises to enable it by default. :) So, @DavidCarlisle, @UlrikeFischer, @JosephWright, what do you say?

@mickep I would assume if it was implemented we'd use it.

@mickep @JosephWright sometimes you measure without using the box, eg \begin{enumerate}[i)] sets a label box the width of vii) (that being the widest small roman number) although the chance that you have a list of less than 5 entries with a font that has i but not v is ... small

@mickep I think I did know about this once, but just put it down like hyphenation happening at a different time as just one of the differences.

btw, does tlmgr update --self automatically jump from tl2024 to tl2025 or does it stick with its year?

@Lupino no you need to manually move years, i always do a fresh install but there is a more or less documented path to upgrading an existing installation from one year to the next.

@Lupino it's advised to do a fresh install generally.

@Skillmon you mean even intra-year updates?

@Lupino no, just from TLyyyy to TLzzzz

This is LuaHBTeX, Version 1.21.0 (TeX Live 2025) (format=lualatex-dev 2025.2.10)

@UlrikeFischer Oooh

@UlrikeFischer Installing here

@DavidCarlisle I guess it will take a few hours, so to motivate, it is good if it will surely be enabled ...

@mickep Definitely can use if available - @DavidCarlisle or I will make the necessary adjustment

@mickep I would probably be good if you could still choose between warning and error. So =4: warning in box =5 error in box.

@UlrikeFischer so slow:

I should get the pretest though....

@DavidCarlisle Seems to need about 1h from the sever in .cz

@DavidCarlisle but I have version 1.21 ;-)

@UlrikeFischer Two new options? Time to find funding...

@JosephWright yes fetching now, in between improving my Japanese

@UlrikeFischer Have you checked the test suite?

@JosephWright Can use or will use? Again, if it will not be used, how should I convince Hans that it is worth to spend the afternoon on?

@mickep Sorry, was ambiguous - if the change is made, I will make a PR and I strongly suspect it will be accepted by the team

@JosephWright No, but I can try to run a bit while making lunch.

Installing [0946/4812, time/total: 14:43/01:05:14]: context-vim [21k]

all my favourite things

@DavidCarlisle Installing [4935/5044, time/total: 58:22/59:10]: xii [2k]

@JosephWright far more useful, and in good time for the 2025 season

@DavidCarlisle :)

@JosephWright trying to understand my own javascript :(

@DavidCarlisle :)

@UlrikeFischer I'm set up now - starting with L3

@DavidCarlisle, @UlrikeFischer Indents change in various LuaTeX outputs ...

E.g.

*** ../build/test/m3token002.luatex.tlg Mon Feb 10 12:54:35 2025
--- ../build/test/m3token002.luatex.log Mon Feb 10 12:54:36 2025
***************
*** 158,164 ****
  ============================================================
  ! Argument of \use_i:nn has an extra }.
  <inserted text>
! \par
  l. ...}
  I've run across a `}' that doesn't seem to match anything.
  For example, `\def\a#1{...}' and `\a}' would produce
--- 158,164 ----
  ============================================================
  ! Argument of \use_i:nn has an extra }.

@JosephWright oh updating the tlg is going to be fun :(

@DavidCarlisle Yup

@DavidCarlisle Will apply everywhere: I have the same in siunitx

Installing [3524/4812, time/total: 01:00:28/01:17:51]: pst-vehicle [2635k]

a bt more to go ^^^

@JosephWright I guess we had better check it's not a mistake and it'll change back

@JosephWright not again!

@DavidCarlisle Indeed

@Skillmon Telling me!

@JosephWright @DavidCarlisle all pdf-based tests will fail. luatex because of this change:***** ..\..\BUILD\TEST-CONFIG-MATH-LUATEX\structelem-text.tpf
  681:  endcodespacerange
  682:  0 beginbfrange
  683:  endbfrange
  684:  3 beginbfchar
***** ..\..\BUILD\TEST-CONFIG-MATH-LUATEX\STRUCTELEM-TEXT.LUATEX.PDF
  681:  endcodespacerange
  682:  3 beginbfchar
*****

and pdftex as it inserts more space-chars now ...

@UlrikeFischer I could try to normalise - but seems like more something we can take as a one-off hit

@UlrikeFischer Oh goody

@egreg only enough for my question to have been disingenuous.

@JosephWright which?

@cfr Floating point one

@JosephWright normalisation is not easy, as this naturally effects also Length keys and the xref table. We will to do it by rebuilding the tests imho. What I would like to have is that l3build could produce a "rebuild-all" script (including the one which needs --engine XY). We have a number of "update-all.sh" around but they are typically incomplete.

@mickep ??

@UlrikeFischer We've discussed 'rebuild all' before ... it's really just the same as running a check, but more importantly there's the whole business that any changes need to be signed off

@JosephWright it got an official answer ...

running check in latex2e/develop base...

@cfr Sorry for pinging you, I meant to point Joseph.

@mickep I think it will get reported after the finder has time to look into it. gs says it is LuaTeX's fault, so that is probably where it will get reported.

@mickep ah. that is why it made no sense to me :-)

@JosephWright Yes I know. But currently one has to run l3build check -S, then you can save and then you have to rerun l3build check -S to get the engine specific failures and then you save again.

@UlrikeFischer yes it's often easier to use l3build check and copy the logs than to use l3build save

@JosephWright @UlrikeFischer but only diff so far (after 373/565) is unrelated:

*** ../build/test/github-1336.luatex.tlg	Mon Feb 10 13:50:41 2025
--- ../build/test/github-1336.luatex.log	Mon Feb 10 13:50:41 2025
***************
*** 12,20 ****
  \macro@cnt=\count...
  \MacroTopsep=\skip...
  \index@excludelist=\toks...
  \IndexMin=\dimen...
  \c@IndexColumns=\count...
- Can't find multicol.sty -- will use normal index layout if necessary.
  \GlossaryMin=\dimen...
  \c@GlossaryColumns=\count...
  \check@sum=\count...
--- 12,71 ----
  \macro@cnt=\count...
  \MacroTopsep=\skip...

@UlrikeFischer @JosephWright oops forget that. The cygwin tl2025 has tl2024 binaries still :(

New primitives:

***** ..\BUILD\TEST\tlb-extraprimitives-001.luatex.tlg
  234:  ignoreligaturesinfont
  235:  immediateassigned
***** ..\BUILD\TEST\TLB-EXTRAPRIMITIVES-001.LUATEX.LOG
  234:  ignoreligaturesinfont
  235:  ignoreprimitiveerror
  236:  immediateassigned
*****

***** ..\BUILD\TEST\tlb-extraprimitives-001.luatex.tlg
  277:  mathdisplayskipmode
  278:  matheqdirmode
***** ..\BUILD\TEST\TLB-EXTRAPRIMITIVES-001.LUATEX.LOG
  278:  mathdisplayskipmode
  279:  mathemptydisplaymode
  280:  matheqdirmode

Does TeX perceive any vskip as pagebreakable or is it possible to make it even a little bit easier to have a pagebreak.

@MaestroGlanz a skip is a penalty 0 break but you don't need a skip at all you can add a negative penalty with or without a following skip to encourage a break. see \pagebreak[3] or \goodbreak or \filbreak

@UlrikeFischer Some funding were found, this is now added. I've heard...

@JosephWright @DavidCarlisle @egreg ^ I assume now 4 or 5 will be used as default.

(And tested, of course!)

@mickep yes I guess so, but the extra space in the "to be read again" logs doesn't look right..

*** ../build/test/github-0251.luatex.tlg	2025-02-10 15:37:44.984499000 +0000
--- ../build/test/github-0251.luatex.log	2025-02-10 15:38:30.908374400 +0000
***************
*** 83,87 ****
  .\glue(\spaceskip) 3.33333 plus 1.66666 minus 1.11111
  ! OK.
  <to be read again>
! }
  l. ...{\showoutput \showbox0}
--- 83,87 ----
  .\glue(\spaceskip) 3.33333 plus 1.66666 minus 1.11111
  ! OK.
  <to be read again>
!                    }
  l. ...{\showoutput \showbox0}

@DavidCarlisle That is unrelated to the missing char thingy, right?

@mickep yes the missing char behaviour has always been in luatex, but that log difference is new today from texlive 2025

@DavidCarlisle there are also quite a lot failures like this:

@DavidCarlisle Yes, but what I meant was: That space was not added by the missing char "patch", right? Something else?

@mickep no that are changes in texlive 2025. Indentation and length of error messages are different.

@mickep yes completely unrelated to the char patch these differences are from the public tl2025 prelease

OK, "great".

@mickep log changes are really painful for the l3build regression testing as we have thousands of logs and the basic idea is that they don't change if the output hasn't changed so cosmetic log changes that are only affecting the log break all the tests....

@DavidCarlisle Oh, so you need to find what change did implement them?

@mickep well we need to find out if that change is going to be reverted as we don't want to spend a day checking and updating diffs if they are going to be undone:-) But I guess @JosephWright @UlrikeFischer and I are all running incomplete tests at present we'll probably collate results and raise on the luatex-list

@DavidCarlisle But you need to know what change implemented the indentation change. Or do you already know?

@mickep No I haven't looked at the source. But as I say, apart from personal interest we don't really need to know. If it's an intentional change, we'll update the test suite, if it's not, and the change gets reverted, the test suite will start passing again. But we can probably make some smaller plain tex test files rather than just running the full latex test suite

@UlrikeFischer more arbitrary differences between wtpdf and 32005 texlive.net/rnc/rncview.html?rnc=wtpdf

@DavidCarlisle Thanks! works for gag purposes!

@dedded you can edit here:-) (<- like this)

Thanks, still learning my way around

@DavidCarlisle wtpdf + pdf/ua are imho more restrictive than 32005.

I wouldn't have believed that this were ever possible for this many people to be interested in LaTeX (yet alone Overleaf) in the world, but when we hit 10M users in late 2022, it became a pretty realistic scenario: Overleaf surpassed 20 million users!

@yo' Congratulations!

@UlrikeFischer yes but why do for example the Ruby elements in 32005 have special markers saying "read the spec for extra conditions" but others like H you have to read a thousand pages over 5 separate specs from two organisations to note a one line comment "H should not be used" ?

@mickep Thanks!

@DavidCarlisle 32005 gives me chills... :)

@mickep an interesting way of specifying a grammar

@DavidCarlisle they should write "use H only if you understand how it should be used" (has the same effect ...).

@UlrikeFischer I think I like this constraint the best. (from WTPDF) "A Span structure element may be used anywhere it is permitted"

@DavidCarlisle Haha

@UlrikeFischer anyway all the latex pdf2.0 examples validate against the new wtpdf.rnc schema with the extra constraints (basically, no H, no Note and every Figure has alt or actualtext) we could start adding NoteType attributes to FENote though?

@DavidCarlisle yes, but Frank is just reviewing the OR-stuff so I would keep out of footnotes for now.

@DavidCarlisle could you check some examples with the luamml from my branch but without my latex2e code? Just in case there is some dump error if we sent luamml first to ctan.

@UlrikeFischer ok

@UlrikeFischer mathsockets ?

@UlrikeFischer died on first attempt, which version of latex-lab ?

@UlrikeFischer anyway I just uninstalled my local one and it's running now with tl2025 + your luamml

@DavidCarlisle better use math socks :) nerdysocks.nl/en/product/math-socks

@JosephWright Can you look at this? I can open an issue, maybe.

\documentclass{article}
\usepackage{etoolbox}
\NewEnvironmentCopy{minipage*}{minipage}

@JosephWright The error is ! Paragraph ended before \etb@carsquare was complete.

@egreg Hmm

! Paragraph ended before \etb@carsquare was complete.
<to be read again>
                   \par
\endvarwidth ->\par

@JosephWright It seems something is not \long as it should be.

@JosephWright The issue was brought up by a user of memoir that loads etoolbox.

@egreg Indeed: \long\def\etb@carsquare#1#2#3\@nil{#1#2} seems to fix

@JosephWright Same error with \newcommand{\foo}{x\par}\NewCommandCopy{\baz}{\foo}

A hot fix seems needed

@egreg Report to team list?

@JosephWright Done

@egreg Thanks

@DavidCarlisle [luatex] A bug in show_context second line indentation

@mickep aha thanks.

I realise today is not a good day to do this but ... I broke cfr-lm. the package cannot be loaded without error. but the package built on github: all the tests passed and the docs built. and they still do. even if the tests are completely useless, that shouldn't be possible, so I am doing something idiotic but I have no idea what.

(I'm not puzzled why the package doesn't work. that's obvious.)

@cfr check the logs in your tests, is it using the texlive version rather than the version extracted from the git sources?

@DavidCarlisle it isn't even installed on github action thing. also, I double-checked the doc build log on github

@cfr where is the log?

Is there also a LaTeX donut challenge? :D

@mickep :)

@cfr the check failed but the failure was masked and the pipeline built the release anyway, or the check didn't fail?

@cis topanswers.xyz/tex?q=1197

@samcarter Uii... very nice.

@cis I don't know if it counts as challenge, but it has doughnuts and latex :)

@cis ^^^ also cupcakes :)

@JosephWright I guess someone will take that thread up again...

ctan.org/incoming

@DavidCarlisle such elegance!

@yo' WOW

@Skillmon the best bit is the new \thanks as I have someone to blame for any inelegance

@AlanMunn yes it's usually possible to juggle downvotes to stay on track, I guess I'll need to move from 2-7 to 3-8 soon:-)