7:37 AM
@AlanMunn yay!

2 hours later…
9:22 AM
For once quite a good blog post

9:59 AM
When using primes in libertinus (lualatex, example below), I notice that the primes look larger than they look in the font file (U+2032). At what point/package is the scaling done?
\documentclass{article}
\usepackage{fontspec}
\usepackage{unicode-math}
\setmathfont{LibertinusMath-Regular.otf}
\begin{document}
$u'f'$
\end{document}

@mickep You are probably talking about the ssty font feature: unicode-math sets the prime in superscript, so the font uses the script style variant.

@MarcelKrüger Oh, that sounds promising, thanks!

@PauloCereda Isn't that price around €5000? It's expensive --- but here you can't buy a car (at least, a real one) with that. it's on the low side even for a (good) e-bike...

@Rmano + lots of taxes :) And don't forget about currency devaluation :)

10:14 AM
@JosephWright welcome to my world:-)

10:25 AM
@JosephWright not sure how many bosses read that.... :)

11:01 AM
@Skillmon Luckily there are no bosses involved in the programming I do

@PauloCereda FMi writes a decent amount of code

@JosephWright s/decent amount of/amount of decent/g
^^ that would be better :)

12:03 PM
@JosephWright You broke matplotlib stackoverflow.com/q/69645097/2777074 :)

Do you guys think I may be a good moderator for site? :))

@enthu This depends on one important question: do you like ducks?

@samcarter To eat or to talk to? ^__^ I like more to talk with them, prefer to be vegetarian!

@enthu In this case you are clearly qualified for the job :)

@samcarter So I feel so enthusiastic about it! :))

12:11 PM
@samcarter I'd need to see the LaTeX to work out what is up

12:25 PM
@JosephWright I've no idea how to intercept the tex code -- that's why I let the OP test my guess ...

@samcarter :)

12:38 PM
TeX Live 2021 available on Overleaf: overleaf.com/blog/tex-live-2021-now-available
4

@PauloCereda Woooo, almost looks like I didn't break lipsum

@PhelypeOleinik ooh a secret :)
@PhelypeOleinik What have the Romans ever done for us? :)

@PauloCereda :-)

12:52 PM
@PauloCereda And @JosephWright got a full chapter this time :)

@yo' To be expected

@JosephWright sure!

@yo' Hopefully I've picked up enough of the oversights that it's now working quite smoothly

@yo' ooh :)

@JosephWright our users will tell you us!

12:58 PM
@samcarter @JosephWright the latex code is in $HOME/.cache/matplotlib in my system, but: I can't reproduce the problem. That is, the two diagrams are ok for me. @Rmano Ah, cool The code for the text on the legend is, for example (I added a 2 to be sure to separate the two runs:) \documentclass{article} \usepackage{type1cm} \usepackage{textcomp} \usepackage[utf8]{inputenc} \usepackage{siunitx} \usepackage[papersize={72in,72in},body={70in,70in},margin={1in,1in}]{geometry} \pagestyle{empty} \begin{document} \fontsize{10.000000}{12.500000}{\sffamily some string 2} \end{document} And for the numbers on the left axis: \documentclass{article} \usepackage{type1cm} \usepackage{textcomp} \usepackage[utf8]{inputenc} \usepackage{siunitx} \usepackage[papersize={72in,72in},body={70in,70in},margin={1in,1in}]{geometry} \pagestyle{empty} \begin{document} \fontsize{10.000000}{12.500000}{\sffamily$-0.75\$}
\end{document}
...so basically siunitx is not used at all here

@Rmano So it could only be packages I load, which haven't changed between v2 and v3

@Rmano you could blame type1cm

1:15 PM
@DavidCarlisle So 'don't blame Joseph' - sounds good

@DavidCarlisle The thing is that it works ok for me. So either 3.0.32 -> 3.0.33 made it, or it is something that happened in the new versions of matplotlib (I have 3.1.2 and the OP has 3.3.2). It seems that the problem is a bounding box one...

@JosephWright the plan then of course would be to give @UlrikeFischer write access to the repo and blame her for any issues.
3

1:35 PM
@JosephWright if you add \showoutpout you can see a difference if you run with a new and and old.

@UlrikeFischer Well yes, if there was any LaTeX in original example ;)

@JosephWright sorry? What I meant is that with siunitx 3 etoolbox and color is loaded and one has an additional box and color in the output. Do you use some shipout hook?

@UlrikeFischer Oh, well yes, but as color is a required package, anything that doesn't work with it is broken

and not my fault

@UlrikeFischer I've switched from translator to translations: the latter uses some hooks but not a shipout one
@UlrikeFischer translations, rather than siunitx, loads etoolbox

1:48 PM
@JosephWright the additional box is not triggered by etoolbox but by color. But we can't know if it is the source of the displacement unless someone can test if it happens with color alone too.

@UlrikeFischer Like I said, if it doesn't work with color, something is broken

2:03 PM
Where can I find documentation for new@ifnextchar? It's not in source2e, not in TeXBook, not in interface3, not in amsmath, not in amsldoc (never mind, found in amsgen. DuckDuckGo search in PDF is just not great)

@UlrikeFischer Yes, color causes the same effect. — TomVincentUK 56 secs ago

@user202729 It's from amsmath: it doesn't skip spaces unlike \@ifnextchar
@samcarter Ah, right: it's not my fault, it's @DavidCarlisle's ;)

@JosephWright perfect! :)

@JosephWright or perhaps the pagecolor code? Does someone knows how they compile (which engine) and what they do then with this large page?

@UlrikeFischer the used engine is configurable via the matplotlib rc, iirc.
I used to have a plot that got slightly out of hand in size, and remember I had to switch to LuaTeX for plotting that one in matplotlib

2:18 PM

@yo' Cool
@yo' Wanted to go via the official channel
Overleaf with TL'21 working nicely for my test project

@JosephWright good idea as it allows to ping the right people :)
@JosephWright great to hear!

@JosephWright quack

@PauloCereda /quack
I hope the performance enhancements in siunitx help Overleaf

@Skillmon I would help, but I'm unable to reproduce it...

2:24 PM
@Rmano I'm quite busy today (and until Friday), so I can't help :(

@UlrikeFischer the cache directory (in my case) is full of .tex, .dvi and .png files. I can try to strace, but as I said I do not see any problem...

@UlrikeFischer It seems latex -interaction=nonstopmode filename.tex is the default for matplotlib

@MarcelKrüger I got this reply from Hans: mailman.ntg.nl/pipermail/ntg-context/2021/103899.html. So, it seems a bit hopeless at the moment to get nicer primes with this font in ConTeXt.

@JosephWright well, I'm currently very annoyed by a few things, code related.... (and I have to admit that one of which is partially my fault for being a lazy coder)
(instead of writing some multi threaded measurement program in which one thread fetches new data and the other writes existing data I took the easier q&d route and just pile up the ram to write the entire data at the end of a measurement. Who could've guessed that Windows would take a major slow down just because the memory is filled by >90%...)

2:42 PM
@samcarter well the question is how the dvi is then processed. I guess the dvi-driver gets confused by the color code, but if one can reproduce it, it is difficult to debug.

2:57 PM
@UlrikeFischer dvipng being passed dvips color specials?

matplotlib question now on the main site: tex.stackexchange.com/questions/619652/…

@DavidCarlisle could be. But I tried dvipng on the dvi and didn't see a problem.

I just looked at the clock of the windows pc in front of me and asked myself how long it could be 17:02. Turns out the UI is a bit frozen...

@Skillmon it's clearly broken as it's 16:07

@Skillmon Do you get an hazard bonus for having to use windows?

3:08 PM
@samcarter Unfortunately not :(
@samcarter A few weeks ago I even had to use LabView...
Oh, the clock now says 17:08

@Skillmon Everyone uses Labview ;)

@JosephWright I hate every single second dealing with it.

@Skillmon Had to use this during studying for one electronics course - already the attempt to install it crashed my computer :)

@samcarter Well, I tried to do a quick measurement. Just a single DMM from National Instruments. Turns out LabView doesn't recognize the device. Installed the correct driver versions -> still no luck. Now I'm using a Python script I've hacked together in roughly 20 minutes or so (see above how "well" it is hacked together)

@Skillmon :D

3:14 PM
I'm totally blaming the PC. Who's mental enough installing Windows 10 on 8GB ram

@Skillmon I don't think the ram size is relevant in this sentence :)

@UlrikeFischer I see a "displaced" png but for both codes (with v2 or without). If I remove the \hbox{} hack I see a correct bounding box. (The plot thickens)

@samcarter no, but a very important factor. Windows isn't particularly good at swapping...

with v3 and v2 I have the same. If I remove the line with the \hbox{} hack, which is not generated by my older version of matplotlib, I have the following:

@Rmano well this doesn't help. If both have the same margin it doesn't explain why only one is displaced in the final picture. One need a difference between both.

3:19 PM
Yes, that is quite puzzling. It explains why I do not have any problem with my older matplotlib, but not why the OP has a problem with v3 and not with v2.

3:32 PM
@JosephWright /ping

3:48 PM
@PauloCereda /pong

4:07 PM
@samcarter twitter.com/hyxpk/status/1450854335219335175 I love the "I am a wall" bit

@PauloCereda Ohh, and all their friends are in the audience :)

@samcarter Yes!

1 hour later…
5:35 PM
@Rmano I can't reproduce the behaviour as well.

5:55 PM
@Skillmon yes, this is puzzling. But it's true that the line \ifdefined\psfrag\else\hbox{}\fi% changes the output. And I am not sure why.
There is a spurious newline there...

6:12 PM
@Rmano well it produces exactly the expected output adding an empty box before the first line will displace the first line by \baselineskip

6:27 PM
ooh fmitex-unicodefonttable

@PauloCereda :)

@PauloCereda better than fmitex-uncomfortable

@DavidCarlisle ooh
@DavidCarlisle that would be fmitex-unbequem :)

@DavidCarlisle Ah, yes. Mhhh then probably it's always wrong under the matplotlib point of view. I would tend to blame @matplotlib for breaking it between 3.1 and 3.3 ...

@PauloCereda fmitex-unbreqn

6:33 PM
@DavidCarlisle ooh

6:51 PM
@MarcelKrüger Do you know if unicode-math is coded to use the ssty alternative for this particular font (libertinus)? It looks like that for other fonts (say latin modern and lucida), the prime glyphs are larger, and that the "original" glyph is used(?).

@mickep ooh Lucida <3

@PauloCereda <3 indeed, although I am currently cheating on lucida, so...

@mickep oh no

@mickep ssty activates special forms for script and is not active in general. What it does depends on the font, unicode-math always activates is for scriptsize and scriptscriptsize fonts.
@mickep But since Latin Modern Math at least has a \prime which is already small-ish and in the superscript position and unicode-math places the prime as a superscript in script-size font (which would normally therefore be higher and smaller), I'm pretty sure that Latin Modern Math does some substitution too.

@MarcelKrüger Yes, but it seems to be done differently in different fonts. Would have been nice with some kind of standard.
(thanks for explaining!)

7:05 PM
If people could agree on a standard, they would either agree to always use the glyph from the base font (which then always has to be superscripted already) or always the glyph from the script font (and then have a huge prime at baseline in the glyph slot). The current situation is basically a hack to work more or less correctly with both types of renderers.

7:36 PM
@MarcelKrüger It sounds to me as if the superscripted variant would be a nice way to go. But I probably miss something.

8:05 PM
@mickep There are placement issues with that: If the font is responsible for superscripting the prime, it is at a fixed height and doesn't adapt to the previous expression and it doesn't compose properly with subscripts. (These can be worked around, but that gets messy) Anyway, you have to convince all font designers first (and then build a time machine and convince all font designers before they design our current fonts).
There is a reason why the OpenType spec is a mess: There are a lot of weird fonts out there and most of them never get updated.