On 2017-07-18, Kornel Benko wrote:
...
> 2.) Can we replace outputting \textcompwordmark with \vphantom{} in
> exported tex file?
> I could not see any difference in the pdflatex output. Also
> lualatex and xelatex displayed correctly.
However, lib/unicodesymbols says:
Do only add commands that give correct output, no hacks that look "similar".
but \textcompwordmark is defined as the LaTeX equivalent to
200C ZERO WIDTH NON-JOINER (ZWNJ) in the LaTeX base:
t1enc.dfu:261:\DeclareUnicodeCharacter{200C}{\textcompwordmark}
utf8.def:247:\DeclareUnicodeCharacter{200C}{\textcompwordmark}
The macro itself is defined in the LaTeX base as
latex.ltx:2036:\DeclareTextCommandDefault{\textcompwordmark}{\leavevmode\kern\z@}
For T1-encoded fonts, it is mapped to character 23:
latex.ltx:8275:\lccode 23 =23 % textcompwordmark in T1
t1enc.def:112:\DeclareTextSymbol{\textcompwordmark}{T1}{23}
Since the recent additon of the standard Unicode text font encoding "TU",
there is a conversion to ZWNJ if Unicode fonts are used with fontspec:
tuenc.def:169:\DeclareTextSymbol{\textcompwordmark}
\UnicodeEncodingName{"200C}
The problem is, that this character is missing in the default (LatinModern
Unicode) font. It works fine with, e.g., DejaVu.
> Moreover, using
> \textcompwordmark with xelatex displays wrong.
I cannot reproduce. In my tests, the output is correct: \textcompwordmark
suppresses the ff ligature without other side-effects in the PDF.
However, there is a difference if you copy-paste from the PDF output:
With evince, I get
* for pdflatex and T1-encoded lmodern fonts:
test with textcompwordmark: ff
test with vphantom: ff
\textcompwordmark translates to the unprintable character u0017 because
this is the position of the ZWNJ character in T1 encoding
(This is a LaTeX or Evince bug.)
* for xetex or luatax with fontspec and LatinModern
test with textcompwordmark: ff
test with vphantom: ff
Output and copy-paste OK despite the "missing character" warning.
The invisible ZWNJ is simply dropped but prevents the ligature.
* for xetex with DejaVu:
test with textcompwordmark: f f
test with vphantom: ff
The ZWNJ is converted to an ordinary space when pasting into LyX
* for luatex with DejaVu:
test with textcompwordmark: ff
test with vphantom: ff
The ZWNJ is kept (and correctly suppresses the ligature in GUI and output)
when pasting into LyX
Options:
a) Do not turn the "missing character" warning into an error for
characters where this is no data loss, e.g. 200C (ZWNJ)
+ helps also, if there are literal ZWNJ characters in the source,
- side-effects for copy-paste with pdflatex and xetex
b) Re-set \textcompwordmark to the Default in the preamble if TU is
used:
\DeclareTextCommand{\textcompwordmark}{TU}{\leavevmode\kern\z@}
As immediate action, maybe also
c) Invert the test(s) with a comment declaring the reason.
d) Use DejaVu or FreeSerif in the affected tests
(preferably added to the documents instead of the script-based
replacement).
Günter
