LaTeX: MathTeXEngine 0.5.0, Luxor version 3.5.0

[cormullion]

MathTeXEngine version 0.5 changed their LaTeX code, so please could Luxor-LaTeXers check and report any issues with Luxor 3.5.0? Thanks!

[davibarreira]

Ok, now I've found this issue. I'll try to check it during lunch. Have you already found any problems?

[cormullion]

I think the font selection code doesn’t find the NewComputerModern Italic easily, and defaults to Regular.

[Kolaru]

Each TeXChar carry its own (FreeType) font, so it should be possible to add a binding in Cairo.jl that can access it direclty using https://www.cairographics.org/manual/cairo-FreeType-Fonts.html

Similarly, now that MathTeXEngine use glyph indices directly, you would need https://www.cairographics.org/manual/cairo-text.html#cairo-show-glyphs that doesn't seem to be exported by Cairo.jl either right now. CairoMakie call into the C library directly so it doesn't have those problems.

I could restore TeXChar.represented_char as a quick hotfix (i.e. a unicode char instead of a glyph index in a specific font), it would give the wrong glyph for greek letters, integral and sum, but should otherwise work nicely.

[cormullion]

Thanks. Ideally better font routines would be added to Cairo rather than to all the packages that depend on Cairo. I think more investigations are needed. Thanks for the excellent help, @Kolaru!

[cormullion]

Work that needs to be done to support MathTeXEngine v0.5.0 correctly might include:

• integrating GeometryBasics.jl into Luxor *
• integrating FreeType(Abstraction).jl into Luxor
• adding call interfaces to Cairo.jl functions for showing glyphs, extents, fonts, etc.
• writing code to read TeXChar glyph indices from fonts and draw glyphs

( Luxor.jl is showing its age - it predates GeometryBasics.jl by some years 😃; perhaps it should be rewritten to use JuliaGeometry packages...)

I think this is a lot of work, so for now LaTex text setting is less than perfect.

[davibarreira]

Hey, @cormullion . Why does MathTeXEngine v0.5.0 requires such significant changes? I mean, more specifically, why GeometryBasics.jl?

[cormullion]

I was looking at Julius' work here - https://github.com/JuliaPlots/Makie.jl/pull/2139/files# - it seems to need stuff from FreeType and GeometryBasics. MTXE now references the font glyphs directly, by index, rather than by Unicode. I'm no expert here, though! :)

[davibarreira]

So, I took a look at the code to try to understand what was going on. Did the original fonts for the symbols change? In theory, the code I wrote for Luxor should work, as it get's the current font set by MathTeXEngine, and applies the fontface. Why doesn't it find the font anymore?

[cormullion]

The main thing I noticed (not being a Latex user) is that the italic fonts weren't found when the documentation job was run. I'm guessing that the use of FreeTypeAbstraction.jl finds/uses font names differently.

Version 0.4 of MTXE returned, for ^n:

(TeXChar 'n' [U+006E in cmmi10 - Regular], [10.481249809265137, 0.34492188692092896], 0.6)

Version 0.5 of MTXE returned, for ^n:

(TeXChar 'n' [index 79 in NewComputerModern - 10 Italic], [10.489200115203857, 0.3878999948501587], 0.6)

The Cairo Toy Api's text() function is happy to select a font with Luxor.fontface("cmmi10"), but I'm not sure if it's happy with "NewComputerModern - 10 Italic" - but it would probably work OK if it received Luxor.fontface("NewCM10-Italic") though.

Perhaps we could use Luxor.settext and Luxor.setfont for all glyphs (not just the few you added), and work out how to select the fonts via setfont given strings like "NewComputerModern - 10 Italic".

The other issue is the use of glyph indices rather than represented_char -> Unicode. Code for this is probably already written somewhere (in MTXE and/or Makie). I don't know how I would do it other than by adding dependencies for FreeType and scanning the fonts for the indices.

[Kolaru]

Did the original fonts for the symbols change?

Yes, I switch from the Computer Modern family (non unicode, a bunch of fonts for math symbols, weird glyph info in the font) to the New Computer Modern family (unicode, all symbols in a single font, but alternate glyph for some math symbols, requiring the glyph index business).

I don't think that GeometryBasics should be needed. The point type used for the position are from there, but that a mere detail I believe.

The other issue is the use of glyph indices rather than represented_char -> Unicode. Code for this is probably already written somewhere (in MTXE and/or Makie)

https://github.com/JuliaPlots/Makie.jl/pull/2139/files#diff-4c2354928757044a6b2d42b25e45db4aa14412f8a0367d3c28db496d62f6b64dR458-R463

This is it in the compat PR in Makie. I think the simpler way is probably to add a similar function to Cairo.jl and then use it there. I am not familiar with ccall so I can't promise anything, but I'll have a look next week-end if no one beats me to it.

[cormullion]

If anyone has some spare time, please check the current master - I've tried to update it to find the fonts. Haven't yet addressed the glyph indices though...

I think it works on MacOS, but can't test it on other platforms...

[davibarreira]

I was not able to figure it out :(