TeX, LaTeX and Friends

cfr

05:38

how can I figure out from a test log whether I've forgotten a \fi or to close a group? I can tell from the pre-normalised log, but if compilation fails with an incomplete \ifx, I can't figure out how to make a test which reliably fails.

(actually, I can't even make one which unreliably fails.)

@mickep 'The two first born' means something very different from 'the first two born'. :-)

mickep

@cfr Yes, thank you, indeed!

cfr

@mickep I'm sure it's helpful in discussing matrices ;).

mickep

06:08

@cfr In Swedish both sounds about right, that was why my colleague was not sure, I guess.

cfr

06:52

@mickep I find it very easy to believe the difference would be not at all obvious. just if you ever get a time machine and go back, remember that injudicious use of the first might start a war or get you beheaded or something.

or just stick to matrices.

or go far enough back they don't speak english.

or stay in sweden.

or just don't get a time machine.

Joseph Wright

07:23

@cfr Put a test at the end for the current if level (using the eTeX primitive) - we do the same for group levels

cfr

@JosephWright thanks. can I have a hint where i should look that up?

Joseph Wright

@cfr \TYPEOUT { The~current~conditional~level~is~\number\currentiflevel } or similar

Joseph Wright

07:41

@cfr Perhaps I should add to the core of l3build - I can see an argument for this

cfr

08:15

@JosephWright I meant the primitive ;).

@JosephWright ??

Joseph Wright

08:38

@cfr As in, we could check at the end of every run

David Carlisle

@cfr texdoc etex

Joseph Wright

09:01

Suggestions for a better command name for github.com/josephwright/siunitx/issues/684? I don't like \hms, really, and \time is taken :)

Ulrike Fischer

@JosephWright that would quite useful imho if the log would typeout something if the result is different to zero.

Joseph Wright

09:38

@UlrikeFischer Indeed

daleif

09:49

Someone sent me this
\documentclass{article}
\usepackage[T1]{fontenc}
\usepackage{mathtools}
\mathtoolsset{mathic}
\usepackage[left=6.6cm, right=6.6cm]{geometry}
\begin{document}
%\showoutput
\textit{A connected graph~\(G\) on \(n\) vertices has no less than \(n-1\) edges.}
\par
\textit{A connected graph \(G\) on \(n\) vertices has no less than \(n-1\) edges.}
\end{document}
It looks like this

I'm trying to understand why the difference in breaking due to with or without ~.

I looked at \showoutput after the word graph and got

....\kern 0.76714
....\penalty 10000
....\glue 3.57774 plus 1.53333 minus 1.0222
....\mathon

....\kern 0.76714
....\penalty 92
....\glue 3.57774 plus 1.53333 minus 1.0222
....\mathon

So the kern and glues are the same, the penalties are not. Can those explain the diferences in line breaks?

mickep

Where does the penalty 92 come from?

Joseph Wright

@egreg Did you see my musing on github.com/josephwright/siunitx/issues/732?

daleif

@mickep not sure. Some 10 years ago we added some code to the mathic part of mathtools to fix an issues with preceeding ~\(...\) nullifying the italic correction. Not 100% sure if that is adding the 92. Probably need to output totally without mathic. I'll eat my lunch first

daleif

10:17

@mickep You're right, that 92 is a bit fishy.

mickep

@daleif But I do not understand why it has the result it has.

Clearly the first line is overful, so it is a "fail". But why that does not happen to the second one, I do not get.

daleif

@mickep It is down to the ~ and the code that \( executes when ~ was detected before it.

For reverence this is the relevant code, can be inserted before activating mathic

\MHInternalSyntaxOn
\def\MT_mathic_true: {
  \MH_if_boolean:nF {math_italic_corr}{
    \MH_set_boolean_T:n {math_italic_corr}
    \MH_if_boolean:nTF {robustify}{
      \MH_let:NwN \MT_mathic_redeffer: \DeclareRobustCommand
    }{
      \MH_let:NwN \MT_mathic_redeffer: \renewcommand
    }
    \MH_let:NwN \MT_begin_inlinemath: \(
    \MT_mathic_redeffer:*\({
      \relax\ifmmode\@badmath\else
      \ifhmode
        \MH_if_dim:w \fontdimen\@ne\font>\z@
          \MH_if_dim:w \lastskip>\z@
            \skip@\lastskip\unskip

Rmano

@JosephWright or going full \datetime{2024-10-03 12:28:33}? (/me ducks) ;-)

mickep

12:04

@daleif Did you solve it yet?

daleif

12:52

@mickep Haven't had time yet. I'm guessing the code some from the chat many years ago. Not sure if I understand the code. The penalty comes from \penalty\count@, but I do not know where the 92 comes from.

David Carlisle

13:14

@daleif @mickep looks like a bug in mathtools

            \MH_if_num:w \lastpenalty>\z@
              \count@\lastpenalty\unpenalty
            \MH_fi:
            \@@italiccorr
            \MH_if_num:w \count@>\z@
\show \b            \showthe\count@
              \penalty\count@

You conditionally save the last penalty in \count@ but you unconditionally use it if it's greater than 0 so you pick up whatever random value it has at that point

92 is the character code of \ and gets set in \newif at the start of article.cls

\NeedsTeXFormat{LaTeX2e}[1995/12/01]
\ProvidesClass{article}
              [2024/02/08 v1.4n
 Standard LaTeX document class]
\newcommand\@ptsize{}
\newif\if@restonecol

@daleif ^^ @mickep I'm sure such leakage never happens in context....

daleif

@DavidCarlisle I knew something was wrong in that code, just not use to the count syntax.

What exactly is the correct solution here? Use a dedicated counter? Run \lastpenalty > 0 again (I haven't checked how \/ is implemented yet)

Joseph Wright

@daleif Dedicated counter

David Carlisle

@daleif the problem isn't really using the scratch counter it is using the value when you haven't set it. Even an mtools-specific counter would be wrong here

daleif

@JosephWright does \/ set any penalties?

@DavidCarlisle Since it is a sratch cntr, counldn't we just set it to zero in the missing else case

David Carlisle

@daleif also as etex is assumed it would be better to replace v which is the classic tex approximation to testing if the previous item is a penalty with \ifnum\lastnodetype=13

@daleif yes

@daleif no, but the \maybe@ic command taht adds italic correction in \textxx does

daleif

13:28

For now I just add the missing set to zero case and make a note that \ifnum\lastnodetype=13 is s better test and then leave that for the Expl3 conversion

David Carlisle

\def \fix@penalty {%
  \ifnum \lastpenalty=\z@
    \@@italiccorr
  \else
    \count@ \lastpenalty
    \unpenalty
    \@@italiccorr
    \penalty \count@
  \fi
}

@daleif the kernel could use the same improvement ^^

daleif

@DavidCarlisle the lastnumtype?

David Carlisle

@daleif yes latex2e.ltx is full of \ifnum\last[skip,penalty]=\z@ tests that would be better as \lastnodetype tests (at some small compatibility risk)

daleif

13:47

Marked as mathtools issue #62. I'll update the dtx latex, just wanted to write it down fresh in memory.

Should say "will update the dtx later"

mickep

14:59

It is still a bit curious that the line break much later es influenced by the penalty before (or inside) the first formula.

David Carlisle

15:11

@mickep after the ~ \count@ gets set to 10000 and that polutes the remaining cases just as the initial 92 did.

mickep

@DavidCarlisle Ouch, so somehow it gets set before the last formula? Or, it looked like what happened.

David Carlisle

@mickep so you get nobreak penaly inserted before \(n\) and before \(n-1\)

mickep

@DavidCarlisle Ah, so, a nice trick not to have to type ~ at all places.

David Carlisle

@mickep the code says "if there is no explicit penalty before inline math, insert a penalty with whatever number happens to be in \count@

mickep

@DavidCarlisle Oh, OK

David Carlisle

15:15

@mickep much more efficient than using context

mickep

@DavidCarlisle Well, ConTeXt has nice mechanisms so that one does not have to write the ~ before short formulas.

David Carlisle

@mickep sigh, clearly just copying latex again, it seems

mickep

@DavidCarlisle Yes, but honestly we did not know that you had this 92/10000 feature there :(

Otherwise we would of course had used the same trick.

yo'

Btw, Overleaf is looking for people to join the support team! LaTeX expertise welcome but not required. Let me know if you were interested or if you wanted some more details about the position! digitalscience.pinpointhq.com/postings/…

5

mickep

Oh, yet another unit question. Now with a \text answer. I guess they get what they deserve.

cfr

17:58

@JosephWright that would definitely be useful - if it would check ifs and grouping. and it would have saved ctan processing my package twice. I think I really should have noticed. but I'm not confident I will, which is why I wanted to test. but I realise I will have to add the test for every package.

Joseph Wright

18:14

@cfr I'm look at it - probably next week

@mickep :)

cfr

@DavidCarlisle thanks.

@yo' possibly, but I doubt you would want me.

barbara beeton

18:56

@JosephWright -- Maybe some shortening of "duration".

Joseph Wright

19:53

@barbarabeeton Oooh, that's quite good

@barbarabeeton \durn/\drn?

barbara beeton

@daleif -- What's your default paper size, A4 or lettersize? I'm trying to replicate this with plain TeX, and may calculations don't produce a sensible result.

@JosephWright -- Oh, I rather like \durn, but it sounds like one is somewhat exasperated.

Ulrike Fischer

@JosephWright Gert suggests \tardis

6

Paulo Cereda

20:58

@UlrikeFischer ooh

David Carlisle

21:49

@barbarabeeton to replicate with plain, put \nobreak before every inline math

Transcript for

$TeX - LaTeX$ TeX, LaTeX and Friends