7

There seem to be significant differences in the horizontal alignment of the "dot-under" accent depending on whether pdfLaTeX, XeLaTeX, or LuaLaTeX is in use. (I use MacTeX2016, which (still) uses LuaTeX 0.95, with all packages updated thru today (2/22).)

With pdfLaTeX:

enter image description here

With XeLaTeX (and fontspec):

enter image description here

With LuaLaTeX (and fontspec):

enter image description here

The placement of the dot-under accents with pdfLaTeX and XeLaTeX is fairly similar, but it's rather different and, in some cases, rather haphazard-looking with LuaTeX; cf. the location of the dot-under accent below the letter i. The placement of the dot-under accent below the letter e isn't exactly too hot either.

Is there something a LuaLaTeX user can do to "fix" the placement of the dot-under accents?

Incidentally, if EB Garamond is used instead of Computer/Latin Modern, the placement of the dot-under accents is fairly similar (and, in particular, not objectionable) for pdfLaTeX, XeLaTeX, and LuaLaTeX.

\documentclass[border=0.5pt]{standalone}
\usepackage{ifluatex,ifxetex}
\ifluatex
    \usepackage{fontspec,luatex85}
\else
   \ifxetex
       \usepackage{fontspec}
   \else
       \usepackage[T1]{fontenc}
   \fi
\fi
\begin{document}
\d{a} \d{e} \d{i} \d{o} \d{u}
\end{document}
  • 1
    What happens is that XeTeX changes the combination \d{i} into the precomposed character (it's HarfBuzz doing it). LuaTeX doesn't do it and Latin Modern is quite weak in composing character placement. – egreg Feb 22 '17 at 22:24
  • @egreg - Thanks for this observation. Is it mainly the OpenType implementation of Latin Modern that's weak in placing composite characters? The reason I ask is that the accent placement happens to be ok under pdfLaTeX if either the lmodern or the cfr-lm font package is loaded explicitly. – Mico Feb 22 '17 at 22:29
  • not that it's that relevant but you don't need fontspec or luatex85 packages with xetex or luatex with recent releases. (and older releases would be using eu1 and eu2 encoding and xunicode not tu encoding, so the answers posted would not apply) – David Carlisle Feb 22 '17 at 23:13
  • @Mico With pdflatex a completely different method is used for the dot below accent. – egreg Feb 23 '17 at 9:04
10

What happens, at the moment, is that no \d composites are declared in the TU encoding, so \d{a} and so on are realized as combinations with the combining dot below (U+0323).

However, with XeLaTeX, the HarfBuzz library will seize the initiative and if the precomposed character exists in the font, it will use it. This doesn't happen with LuaTeX.

This is a grey area, I'd say. If the composite are declared, you get nothing if the font doesn't support the characters, if not, you have to rely on good positioning.

Indeed, if I try \showoutput with LuaLaTeX, I get for \d{i}

....\TU/lmr/m/n/10 i
....\TU/lmr/m/n/10 ̣

whereas XeLaTeX shows

....\TU/lmr/m/n/10 ị

Unfortunately, Latin Modern is quite weak in positioning certain combining characters, other fonts are better in this respect.

You can fix the issue, provisionally, by defining yourself the necessary composites:

\documentclass{article}
\usepackage{fontspec,ifluatex}

\makeatletter
% fix the hidden feature in tuenc.def
\def\add@unicode@accent#1#2{%
  \if\relax\detokenize{#2}\relax^^a0\else#2\fi
  \char#1\relax
}
\makeatother

\ifluatex
\usepackage{luatex85}
\DeclareTextComposite{\d}\UnicodeEncodingName{A}{"1EA0}
\DeclareTextComposite{\d}\UnicodeEncodingName{a}{"1EA1}
\DeclareTextComposite{\d}\UnicodeEncodingName{E}{"1EB8}
\DeclareTextComposite{\d}\UnicodeEncodingName{e}{"1EB9}
\DeclareTextComposite{\d}\UnicodeEncodingName{I}{"1ECA}
\DeclareTextComposite{\d}\UnicodeEncodingName{i}{"1ECB}
\DeclareTextComposite{\d}\UnicodeEncodingName{O}{"1ECC}
\DeclareTextComposite{\d}\UnicodeEncodingName{o}{"1ECD}
\DeclareTextComposite{\d}\UnicodeEncodingName{U}{"1EE4}
\DeclareTextComposite{\d}\UnicodeEncodingName{u}{"1EE5}
\fi    

\begin{document}
\d{a} \d{e} \d{i} \d{o} \d{u}
\end{document}

enter image description here

For the “hidden feature”, see How to place a dot below for example: e?

  • Many thanks for this detailed analysis and the proposed work-around! – Mico Feb 23 '17 at 8:20
8
+100

As egreg notes this is tricky, if we leave it as the default \d accent then you at least get some kind of dot under as long as the font has U+0303 however if we declare the composites, you get a better positioned dot if the font has the composite but a missing glyph in luatex if it doesn't, which is not really an improvement over a mis-aligned dot. xetex (harfbuzz) will find the right glyph in any case if the font has the pre-composed character.

I need to do an update anyway for the error reported

How to place a dot below for example: e?

but I'm wondering whether to add some \d declarations. I suspect not at least for the update which needs to be today or tomorrow.

Unicode has 42 pre-composed characters with a base followed by U+0303, but not all fonts have them, so it isn't clearly an advantage to declare the composites in the tuenc default.

\documentclass{article}
\showoutput
\makeatletter
\input{tuenc.def}
\begin{document}

1 \d{i}

2 i^^^^0323

3 ^^^^1ecb

\end{document}

in luatex and xetex:

enter image description here


To see what happens if all the composites with latin letters are declared

\documentclass{article}
\showoutput
\makeatletter
\input{tuenc.def}
\begin{document}



x: \d{A} \char"1EA0

x: \d{B} \char"1E04

x: \d{D} \char"1E0C

x: \d{E} \char"1EB8

x: \d{H} \char"1E24

x: \d{I} \char"1ECA

x: \d{K} \char"1E32

x: \d{L} \char"1E36

x: \d{M} \char"1E42

x: \d{N} \char"1E46

x: \d{O} \char"1ECC

x: \d{R} \char"1E5A

x: \d{S} \char"1E62

x: \d{T} \char"1E6C

x: \d{U} \char"1EE4

x: \d{V} \char"1E7E

x: \d{W} \char"1E88

x: \d{Y} \char"1EF4

x: \d{Z} \char"1E92



x: \d{a} \char"1EA1

x: \d{b} \char"1E05

x: \d{d} \char"1E0D

x: \d{e} \char"1EB9

x: \d{h} \char"1E25

x: \d{i} \char"1ECB

x: \d{k} \char"1E33

x: \d{l} \char"1E37

x: \d{m} \char"1E43

x: \d{n} \char"1E47

x: \d{o} \char"1ECD

x: \d{r} \char"1E5B

x: \d{s} \char"1E63

x: \d{t} \char"1E6D

x: \d{u} \char"1EE5

x: \d{v} \char"1E7F

x: \d{w} \char"1E89

x: \d{y} \char"1EF5

x: \d{z} \char"1E93


\end{document}

If the composites are declared then use of the \d command with the specified base would be equivalent to \char with the composite character, the following shows the result with luatex and xetex.

enter image description here

Note that luatex drops some characters, reporting

Missing character: There is no Ḅ (U+1E04) in font [lmroman10-regular]:+tlig;!
Missing character: There is no Ḳ (U+1E32) in font [lmroman10-regular]:+tlig;!
Missing character: There is no Ṿ (U+1E7E) in font [lmroman10-regular]:+tlig;!
Missing character: There is no Ẉ (U+1E88) in font [lmroman10-regular]:+tlig;!
Missing character: There is no ḅ (U+1E05) in font [lmroman10-regular]:+tlig;!
Missing character: There is no ḳ (U+1E33) in font [lmroman10-regular]:+tlig;!
Missing character: There is no ṿ (U+1E7F) in font [lmroman10-regular]:+tlig;!
Missing character: There is no ẉ (U+1E89) in font [lmroman10-regular]:+tlig;!

So those combinations work with \d in Latin Modern only if the composite isn't declared.

Note it would be possible to declare \d (or all accents) to do a check to see if the composite character is present and if not use the combining combination, but currently that would mean doing the test on every use, as latex stores these accent commands per-encoding not per-font. fontspec provides a mechanism to declare variants of TU encoding which could be used to address this, but it is not really convenient to declare a custom encoding for every font.

Another possibility of course would be to use harfbuzz from luatex, which is not an impossible aim.

  • Wow, many thanks for these detailed explanations. (I really wish I could have accepted both answers!) I will definitely give the idea of combining harfbuzz with luatex a try, probably over the weekend. I'll report back if I succeed. :-) About the OpenType version of the Latin Modern fonts not providing "native" dot-under accent glyphs for b, k, v, w, B, K, V, and W: that does seem like an oversight, especially as other opentype fonts (e.g., EB Garamond) don't have this limitation. – Mico Feb 23 '17 at 8:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.