Italic text and conjugate alignment

[AshtonSBradley]

A very common axis label use case is some function, e.g. f(x), or f(t). An italic argument has a noticeable offset - anything of the form ax.xlabel=L"g(t)", etc in both GLMakie, and CairoMakie, so I assume its a font issue.

Visually, it appears that standard latex formatting is to horizontally centre the vertical midpoint of the character.

Here is an example of a CairoMakie figure inserted into a latex manuscript. The caption shows clearly the normal latex alignment, while the ylabels in Makie have an offset

Is is possible to centre the character vertical midpoint to get a balanced result like latex?

MakieOrg/Makie.jl#4777

[AshtonSBradley]

here is another example of unusual math alignment. The complex conjugate is quite high, compared to normal latex in the caption. Is this related to the first problem somehow?

[Kolaru]

Thanks for reporting!

Using the still experimental prototype/prototype.jl, we can look at the bounding boxes and it gives some clues (yellow is extra space before a glyph, green extra space after, according to the font)

Those are two separate issues:

• The spacing of (t) is related to the kerning of the letter: the font is expecting an italic character after an italic t, so the space is reduced, which is ugly when followed by a roman character (see Cursive / normal text boundaries #9). Probably the solution is to just ignore the kerning at the transition.
• The \alpha^* is different: * is an operator, and thus extra space is added around it by default (like for +). This means we should identify whether the symbol is starting or alone in its group. Related: Unary Minus looks too wide, math superscript too high #61

Unfortunately, I don't know when I'll have time to look into this in more details :( (by backlog here is growing)

[AshtonSBradley]

Thanks for this insight. It must have surfaced more recently - although the current build of beautifulmake does show the problem (x not as bad as t for this), I found a test figure I made with Makie in May 2022 that doesn't:

[manuelbb-upb]

By default, for L"f(t)" the glyph for f and t will be from the NewComputerModern-Italic face, and ( and ) will be rendered with NewComputerModernMath-Regular.

I think a few years back, instead of NewComputerModern the default typeface family was ComputerModern, where the glyph extents might have been defined differently?

A possible solution is to force the same typeface for everything (via font_mapping) and explicitly use the correct unicode glyphs:

using GLMakie
using LaTeXStrings
import MathTeXEngine as MTE

ffam = MTE.get_texfont_family()
ffam.font_mapping[:delimiter]=:math
ffam.font_mapping[:char]=:math
ffam.font_mapping[:punctuation]=:math

Z = rand(2, 10)
fig = Figure(; size=(200,200), fontsize=20)
ax = Axis(fig[1,1]; ylabel = L"f(t),𝑓(𝑡),g(x),𝑔(𝑥)")
scatter!(ax, Z)

The y-label looks much nicer now:

This is of interest to me because of the unicode substitution mechanism I am thinking about in #143.
A unicode normalization mechanism (e.g., substituting latin letters for the italic symbols from the unicode math block) would allow us to just use :math font for everything with an :inline_math state and simplify the font settings a bit.

[AshtonSBradley]

awesome, that looks really good. How does it cope with superscripts like * ?

[AshtonSBradley]

does this mean it would be limited to supported unicode? I think this might be a bit narrow for some things like super and subscripts, there are a bunch of limitations

[AshtonSBradley]

Is it right that the gs are swapping?

[manuelbb-upb]

Unfortunately, it does not resolve the height issue with superscripts.
As @Kolaru has noted above, that is a separate issue.

I kind of hijacked this thread because I noticed that the spacing is more consistent if the glyphs are taken from the same typeface (NewComputerModernMath-Regular in this case).
Even though the typeface is called "...-Regular", it has italic variants of some characters (latin+greek) in the Unicode Math Block.
That is also why "g" and "𝑔" look so different: it is how the font designers decided to distinguish the letter variants for the NewComputerModernMath-Regular typeface.
I don't know how it would render with a LaTeX compiler if that same typeface was used.
Edit: Indeed, it seems correct that the "mathematical italic small g" has no belly in NCM, whilst the upright letter g, as well as the italic g have a belly, cf the font sample or the symbol tables.
The difference is not so stark with other typeface families, see below examples with TeXGyrePagella.

My motivation to get involved in this thread is just to show my findings to @Kolaru ;)
Originally, I wanted to get consistent styling for the plots in my thesis, where I am using the unicode-math package with ISO styling conventions and the TeX Gyre Pagella typeface family.
At the moment, I have a small package UnicodeMath to test things out, but it is not part of MathTeXEngine (and I kind of made a mess of my pull request(s)).
You can still try it out as shown in the script below, which has generated these plots:

using GLMakie
using LaTeXStrings
import MathTeXEngine as MTE

#=
using Pkg
Pkg.add(;url="https://github.com/manuelbb-upb/UnicodeMath.jl", rev="192524f9ea59debf398bd772f1108d4f3224bb9c")
=#
import UnicodeMath: _sym, symup, symit

ffam = MTE.set_texfont_family!(MTE.default_font_families["TeXGyrePagella"])
ffam.font_mapping[:delimiter] = :math
ffam.font_mapping[:digit] = :math
ffam.font_mapping[:function] = :math
ffam.font_mapping[:punctuation] = :math
ffam.font_mapping[:symbol] = :math
ffam.font_mapping[:char] = :math

Z = rand(2, 10)
function test_fig()
    global Z
    fig = Figure(; size=(320,240), fontsize=20)
    ax = Axis(fig[1,1]; 
        ylabel = L"f(t),𝑓(𝑡),g(x),𝑔(𝑥)", 
        xlabel = L"x^*" # will use upright x
    )
    scatter!(ax, Z)
    return fig    
end

function test_fig2()
    global Z
    fig = Figure(; size=(320,240), fontsize=20)
    ax = Axis(
        fig[1,1]; 
        ylabel = latexstring("$(symup("f(t)")),$(symit("f(t)"))"),
        xlabel = latexstring("$(_sym("x^*"))")  # will use `:tex` style => italic x
    )
    scatter!(ax, Z)
    return fig    
end

fig1 = test_fig()
fig2 = test_fig2()

save(joinpath(@__DIR__, "fig.png"), fig1)
save(joinpath(@__DIR__, "fig2.png"), fig2)

In test_fig2, for the ylabel, I explicitly request upright glyphs with symup and italic glyphs with symit.
For the xlabel I used the _sym command which automatically uses a global styling configuration like unicode-math does in LaTeX.

[manuelbb-upb]

Until there is progress with the unicode stuff, you could checkout this branch:

Pkg.add(; url="https://github.com/manuelbb-upb/MathTeXEngine.jl", rev="italic_correction")

I have fiddled with some heuristics that could help with spacing using the current typesetting method (and thus maybe help with #9)

using GLMakie
using LaTeXStrings
import MathTeXEngine as MTE

Z = rand(2, 10)

function test_fig(it_corr=true)
    global Z
    MTE.ITALIC_CORRECTION[] = it_corr
    fig = Figure(; size=(320,240), fontsize=20)
    ax = Axis(
        fig[1,1]; 
        xlabel=L"\eta(t), \mathrm{y}(x), g(f(x)), \mathrm{g}t"
    )
    scatter!(ax, Z)
    return fig    
end

fig1 = test_fig(true)
save(joinpath(@__DIR__, "fig_it_corr.png"), fig1)
fig2 = test_fig(false)
save(joinpath(@__DIR__, "fig_not_corr.png"), fig2)