LaTeX: support for Greek and Cyrillic #5645

jfbu · 2018-11-16T20:16:25Z

do not escape Unicode Greek letters via LaTeX math mark-up: pass them
through un-modified to LaTeX document,
if "fontenc" receives extra option LGR, then pdflatex will support
Unicode Greek letters (not in math), and with extra option T2A it
will support (most) Unicode Cyrillic letters.
for pdflatex with LGR, this will use "textalpha" LaTeX package and
"substitutefont" package to set up some automatic font substitution
to work around the unavailability of Greek with "times"
package (which is default font package chosen by Sphinx for
pdflatex), same with T2A and "substitutefont" for Cyrillic.
for xelatex/lualatex, set up Computer Modern Unicode as default font,
as it supports Cyrillic and Greek scripts,
for platex, don't do anything special as the engine already has
its default font supporting Cyrillic and Greek (even in math mode!)

edit

The final version differs from the description above:

for xelatex/lualatex use FreeFont (because on Ubuntu xenial, CMU requires the big package texlive-fonts-extra, and FreeFont has its own separate fonts-freefont-otf)
also X2 encoding is handled for pdflatex, and the fonts are the base ones in LGR or X2 or T2A encoding, providedon Ubuntu xenial from cm-super package (and texlive-lang-greek, texlive-lang-cyrillic). These extra dependencies are optional only, to support the new feature of support of Greek and Cyrillic in European language document.

TODO: indexing with xindy has merge-rules for Greek, but for Cyrillic no, because they are loaded only for Cyrillic documents, so indexing in non-Cyrillic document of Cyrillic words will work but all such words will end-up in "non-alphabetical" group.

Closes: #5251
Fixes: #5248
Fixes: #5247
Closes: #1682

0. do not escape Unicode Greek letters via LaTeX math mark-up: pass them through un-modified to LaTeX document, 1. if "fontenc" receives extra option LGR, then pdflatex will support Unicode Greek letters (not in math), and with extra option T2A it will support (most) Unicode Cyrillic letters. 2. for pdflatex with LGR, this will use "textalpha" LaTeX package and "substitutefont" package to set up some automatic font substitution to work around the unavailability of Greek with "times" package (which is default font package chosen by Sphinx for pdflatex), same with T2A and "substitutefont" for Cyrillic. 3. for xelatex/lualatex, set up Computer Modern Unicode as default font, as it supports Cyrillic and Greek scripts, 4. for platex, don't do anything special as the engine already has its default font supporting Cyrillic and Greek (even in math mode!) Closes: sphinx-doc#5251 Fixes: sphinx-doc#5248 Fixes: sphinx-doc#5247

jfbu · 2018-11-16T23:15:08Z

CHANGES

+  the use of the ``LGR`` (Greek) and/or ``T2A`` (Cyrillic) font encoding. Even
+  then, the last four are font packages arising in the default value for
+  :confval:`latex_elements`.\ ``'fontpkg'``, and may be replaced by other font
+  packages providing ``LGR`` and/or ``T2A`` support.


These explanations may move later on to builders/index.rst from CHANGES, after 2.0 release

jfbu · 2018-11-16T23:17:17Z

CHANGES

+  will use the text font not the math font. If (and only if) the document
+  contains such Greek Unicode letters *and* the :confval:`latex_engine` is
+  ``'pdflatex'`` then the :confval:`latex_elements`.\ ``'fontenc'`` key
+  **must** be used to declare usage of the ``LGR`` font encoding.


Initially I added LGR and usage of packages textalpha and substitutefont to the default pdflatex setting so that no document would need any change for transition to Sphinx 2.0. Unfortunately the textalpha is on Ubuntu xenial available via texlive-lang-greek which wasn't previous dependency.

So after some hesitation I decided on the contrary to not change anything to the default 'fontenc' but this means that if a document had a Greek Unicode in text, which previously was escaped to math mark-up, now, what will happens is that project author must explicitely use latex_elements 'fontenc' key to add usage of LGR encoding. And the TeX installation might need additional LaTeX packages (listed above).

jfbu · 2018-11-16T23:20:25Z

CHANGES

@@ -58,10 +97,18 @@ __ https://github.com/sphinx-contrib/sphinx-pretty-searchresults
 * #4018: htmlhelp: Add :confval:`htmlhelp_file_suffix` and
  :confval:`htmlhelp_link_suffix`
 * #5559: text: Support complex tables (colspan and rowspan)
+* LaTeX: support rendering (not in math, yet) of Greek and Cyrillic Unicode


it is possible via alphabeta package to support Greek Unicode literals in math directive with pdflatex, but the problem is that an upright Unicode alpha will be mapped to \alpha and rendered by italic math font, contrarilty to MathJax rendering, which will keep separate the upright and the italic. Anyway there is no issue yet about this on your tracker, and besides, only xelatex/lualatex are reasonable for supporting Unicode input in math, it is not worthwile to go to extreme lengths to try to support it with pdflatex.

I was surprised to discover that platex (Japanese) has all set-up and individual Greek and Cyrillic letters work fine both in text mode and in math mode...

jfbu · 2018-11-16T23:20:58Z

CHANGES

+* #5247: LaTeX: PDF does not build with default font config for Russian
+  language and ``'xelatex'`` or ``'lualatex'`` as :confval:`latex_engine`
+  (refs: #5251)
+* #5248: LaTeX: Greek letters in section titles disappear from PDF bookmarks


There still remains #5249, but see previous comment, user should choose xelatex and add usage of unicode-math package for this.

jfbu · 2018-11-16T23:22:52Z

doc/usage/configuration.rst

+                \substitutefont{LGR}{\ttdefault}{cmtt}
+                \substitutefont{T2A}{\rmdefault}{fcm}
+                \substitutefont{T2A}{\sfdefault}{fcs}
+                \substitutefont{T2A}{\ttdefault}{fct}


the T2A fonts are from the cm-lgc LaTeX font package

I think I should replace the last three lines with \substitutefont{X2}{\rmdefault}{cmr} etc.. because x2cmr.fd exists in texlive-lang-cyrillic, and covers more Cyrillic than T2A encoded font files.

jfbu · 2018-11-16T23:33:21Z

sphinx/writers/latex.py

+                        '  \\sphinxDUC{2502}{\\sphinxunichar{2502}}\n'
+                        '  \\sphinxDUC{2514}{\\sphinxunichar{2514}}\n'
+                        '  \\sphinxDUC{251C}{\\sphinxunichar{251C}}\n'
+                        '  \\sphinxDUC{2572}{\\textbackslash}\n'


this is only refactoring, for better readability of what the code does

jfbu · 2018-11-16T23:35:46Z

sphinx/writers/latex.py

-        'fontpkg':      '',
+        'fontpkg':     ('\\setmainfont{CMU Serif}\n'
+                        '\\setsansfont{CMU Sans Serif}\n'
+                        '\\setmonofont{CMU Typewriter Text}'),


These fonts are visually close to previous default with xelatex: previous default was LaTeX choice i.e. Latin Modern OpenType, which is not far from original Knuth Computer Modern. The CMU Serif is the Unicode version of Computer Modern, it exists in serif, sans serif, monospace and supports both Greek and Cyrillic.

Alternative is Libertinus, which is more Times-like (derived from Linux Libertine) but this is more recent font, currently still under development recently.

jfbu · 2018-11-16T23:36:49Z

sphinx/writers/latex.py

+            if ('T2A' in self.elements['fontenc'] and
+                not self.babel.uses_cyrillic()):
+                self.elements['substitutefont'] = '\\usepackage{substitutefont}'
+                self.elements['sphinxpkgoptions'] += ',cyrnocyr'


I prefered to move indeed to sphinx.sty the fancy latex, rather than have it in template

As said above I think now the fancy latex should go into some sphinxcyrillic.sty.

jfbu · 2018-11-16T23:37:47Z

sphinx/writers/latex.py

+                self.elements['substitutefont'] = '\\usepackage{substitutefont}'
+                self.elements['sphinxpkgoptions'] += ',cyrnocyr'
+            if 'LGR' in self.elements['fontenc']:
+                self.elements['substitutefont'] = '\\usepackage{substitutefont}'


Sphinx does not support Greek as main language I think, so it makes sense not to worry about LGR being last font encoding in 'fontenc' key.

jfbu · 2018-11-16T23:39:04Z

sphinx/writers/latex.py

+            if 'LGR' in self.elements['fontenc']:
+                self.elements['substitutefont'] = '\\usepackage{substitutefont}'
+            else:
+                self.elements['textgreek'] = ''


This way no dependency is added to existing Sphinx projects only needing T1 encoding (i.e. all (non-Cyrillic) projects which did not have Unicode Greek letters in their text, previously rendered via TeX math font)

codecov · 2018-11-16T23:42:19Z

Codecov Report

Merging #5645 into master will decrease coverage by <.01%.
The diff coverage is 33.33%.

@@            Coverage Diff            @@
##           master   #5645      +/-   ##
=========================================
- Coverage    83.2%   83.2%   -0.01%     
=========================================
  Files         294     294              
  Lines       39195   39219      +24     
  Branches     5864    5865       +1     
=========================================
+ Hits        32614   32632      +18     
- Misses       5220    5224       +4     
- Partials     1361    1363       +2

Impacted Files	Coverage Δ
tests/test_markup.py	`96.73% <ø> (ø)`	⬆️
sphinx/util/texescape.py	`100% <ø> (ø)`	⬆️
sphinx/writers/latex.py	`83.53% <33.33%> (-0.39%)`	⬇️
sphinx/config.py	`82.74% <0%> (-0.79%)`	⬇️
sphinx/testing/util.py	`93.51% <0%> (-0.12%)`	⬇️
sphinx/writers/text.py	`93.52% <0%> (-0.03%)`	⬇️
tests/test_util.py	`100% <0%> (ø)`	⬆️
setup.py	`0% <0%> (ø)`	⬆️
sphinx/writers/texinfo.py	`88.08% <0%> (ø)`	⬆️
... and 5 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6c94801...1edf2a4. Read the comment docs.

jfbu · 2018-11-17T00:27:35Z

@tk0miya sorry for flake8, but for some time now I can't execute the tests at my locale because I use Anaconda Python distribution and it has ceased supporting my system version of Mac OS X, and flake8 simply refuses to work.

As per circleci, the build succeeds with lualatex (showing CMU Serif is available) and fails with xelatex which can not find the font, I will try to modify it to call the font by filename but this is more cumbersome. ~~I wonder if testing is done on Mac OS ?~~

I had forgotten xetex can't find TeXLive fonts by font names, it needs filenames, because at my locale I have symlinks to work around that problem.

jfbu

While working on this PR, I did not pay attention to the dependencies on say Ubuntu Xenial. Indeed I mainly work on my own Mac OS X with a (quasi-full) TeXLive installation.

I feel I need to think a bit more about minimizing dependencies resulting from choice of fonts. For most Sphinx users wishing to use some Greek or Cyrillic exceptionally in some non-Cyrillic document, important is that the Unicode letter shows in PDF, not that the font is well-matched with the font for Latin text.

jfbu · 2018-11-17T13:46:41Z

CHANGES

+       - Greek letters (in text, not math)
+     * - cm-lgc
+       - texlive-fonts-extra
+       - Cyrillic letters (in non-Cyrillic documents)


They are alternative which may be better: texlive-fonts-extra appears to be rather big Ubuntu/Debian package, which installs many many TeX fonts. Thus, texlive-lang-cyrillic is much smaller and provides some Computer Modern fonts in "cyrillic" core LaTeX bundle which is not cm-lgc. They are available in X2 encoding which covers even more Cyrillic glyphs. I must check if in PDF they are rendered as scalable fonts. (I expect so)

It would be logical to tell people: for support of occasional Greek, you will need texlive-lang-greek and for support of occasional Cyrillic texlive-lang-cyrillic but not require them the big texlive-fonts-extra.

For xelatex/lualatex, this is different, I have no qualms about requiring full up-to-date installation if necessary. But for vast majority of projects using pdflatex it makes sense to try not to add big dependencies.

The X2 encoded Computer Modern seem to require not only texlive-lang-cyrillic but also cm-super-minimal (because PDF needs subsetting in particular font file sfrm1000.pfb, if I look at log of pdflatex at my locale). Investigating if cm-super-minimal is enough or cm-super is needed.

Investigating if cm-super-minimal is enough or cm-super is needed.

Yes, cm-super-minimal is enough, but \usepackage[10pt]{type1ec} must be used before \usepackage[...]{fontenc}. From the README.debian of the cm-super-minimal package,

If you DON'T have cm-super (the full package) installed you have to use
\usepackage[10pt]{type1ec}
which uses only those cm-super fonts available in cm-super-minimal and
scale those fonts for other design sizes.

and I confirmed by looing at contents of type1ec.sty file to see what it does.

jfbu · 2018-11-17T13:47:36Z

doc/conf.py

@@ -56,10 +56,17 @@
                    'Georg Brandl', 'manual', 1)]
 latex_logo = '_static/sphinx.png'
 latex_elements = {
+    'fontenc': r'\usepackage[LGR,T2A,T1]{fontenc}',


I will probably switch to X2 in place of T2A to cover even more of Cyrillic Unicode block.

jfbu · 2018-11-17T13:51:13Z

doc/usage/builders/index.rst

@@ -164,8 +164,11 @@ The builder's "name" must be given to the **-b** command-line option of
   * ``texlive-latex-recommended``
   * ``texlive-fonts-recommended``
   * ``texlive-latex-extra``
+   * ``texlive-fonts-extra``, ``texlive-lang-greek`` (if needed to


It would be much better to say here texlive-fonts-cyrillic in place of texlive-fonts-extra

jfbu · 2018-11-17T13:51:33Z

doc/usage/configuration.rst

+                \substitutefont{LGR}{\ttdefault}{cmtt}
+                \substitutefont{T2A}{\rmdefault}{fcm}
+                \substitutefont{T2A}{\sfdefault}{fcs}
+                \substitutefont{T2A}{\ttdefault}{fct}


I think I should replace the last three lines with \substitutefont{X2}{\rmdefault}{cmr} etc.. because x2cmr.fd exists in texlive-lang-cyrillic, and covers more Cyrillic than T2A encoded font files.

jfbu · 2018-11-17T13:52:28Z

doc/usage/configuration.rst

+             both Cyrillic and Greek scripts (contrarily to the
+             default font configured by LaTeX for ``xelatex/lualatex``
+             if ``'fontpkg'`` is left to empty string, as was the case
+             prior to 2.0).


TODO: \setmainfont{CMU Serif} syntax works only with lualatex, for xelatex one needs in fact filename, as I did in latex.py. Update doc.

jfbu · 2018-11-17T13:55:36Z

sphinx/texinputs/sphinx.sty

+  }%
+  \DeclareTextSymbolDefault{\CYRpalochka}{T2A}%
+\fi
+


I will probably replace it by more extensive code covering X2 which is superset of T2A for Cyrillic. And tex.sx answer did already the work, in a more agreeable syntax using Unicode letters. I think I should externalize this to a sphinxcyrillic.sty file probably, to keep sphinx.sty simpler. Besides with a separate sphinxcyrillic.sty file, its loading can be done from template rather than pass an option to sphinx.

jfbu · 2018-11-17T13:57:12Z

sphinx/texinputs/sphinx.xdy

+(merge-rule "\IeC {\textChi }" "Χ" :string)
+(merge-rule "\IeC {\textPsi }" "Ψ" :string)
+(merge-rule "\IeC {\textOmega }" "Ω" :string)
+(merge-rule "\IeC {\textohm }" "Ω" :string)


all these merge rules do nothing bad with xelatex/lualatex where Unicode letters remain identical to themselves when written to file; they serve only with pdflatex because the .idx file will have those macros rather than the Unicode letters. So we map the macros back to the Unicode letters.

jfbu · 2018-11-17T13:58:15Z

sphinx/writers/latex.py

+% T2A was declared as font encoding
+  \substitutefont{T2A}{\rmdefault}{fcm}
+  \substitutefont{T2A}{\sfdefault}{fcs}
+  \substitutefont{T2A}{\ttdefault}{fct}


I will change that presumably to X2 with cmr, so that only texlive-lang-cyrillic is needed not the much bigger texlive-fonts-extra.

jfbu · 2018-11-17T13:59:23Z

sphinx/writers/latex.py

+                        '\\setmonofont{cmuntt.otf}[\n'
+                        '  BoldFont = cmuntb.otf,\n'
+                        '  ItalicFont = cmunit.otf,\n'
+                        '  BoldItalicFont = cmuntx.otf]'),


Well what a pain that is. On my MacOS, the problem did not arise because I used symlinks from font repertory ~/Library/Fonts to the TeXLive location, so XeTeX knows how to find fonts per font name, not filename, so in initial commit I had done for xelatex as for lualatex.

Unfortunately this seems to require texlive-fonts-extra. But one may expect from Sphinx xetex/lualatex users complete up-to-date TeX installations, so perhaps this is acceptable. Libertinus font is more recent yet.

jfbu · 2018-11-17T14:03:07Z

sphinx/writers/latex.py

+            if ('T2A' in self.elements['fontenc'] and
+                not self.babel.uses_cyrillic()):
+                self.elements['substitutefont'] = '\\usepackage{substitutefont}'
+                self.elements['sphinxpkgoptions'] += ',cyrnocyr'


As said above I think now the fancy latex should go into some sphinxcyrillic.sty.

jfbu · 2018-11-21T21:31:13Z

I am not sure the testing includes building out document to PDF: if it does, the cm-super package is required now.

In my last commit I have

tried to remove the clutter and be less verbose in explanations,
decided for FreeFont as default choice of fonts for the Unicode engines,
added support to pdflatex for X2-encoded Cyrillic which covers a lot of Cyrillic glyphs
reduced extra requirements on Ubuntu xenial to a minimum: texlive-fonts-extra never required, only individual packages such as fonts-freefont-otf or cm-super(-minimal). And these requirements are optional, only to support previously non existing support of Greek/Cyrillic in English or European languages documents.

jfbu · 2018-11-21T21:47:59Z

Testing completed. I will push one last commit to avoid using same option prefix for sphinx.sty and (new) sphinxcyrillic.sty and then I will merge.

Forgot to add it to previous commit :(

jfbu · 2018-11-21T22:12:40Z

@tk0miya merged! thanks for reviewing, I since externalized to new sphinxcyrillic.sty, fixed up font choice for xelatex/lualatex, clarified the Ubuntu package dependencies and improved the documentation.

jfbu · 2018-11-22T08:44:35Z

@tk0miya It seems our LaTeX testing does not include building our PDF documentation sphinx.pdf (I agree this takes time, less now thanks to your work on speeding up the gathering of references phase). MEMO: to build sphinx.pdf this PR adds the following requirements texlive-lang-greek, texlive-lang-cyrillic, and cm-super (because Greek and Cyrillic is used for demonstration in document). ~~fonts-latex-extra~~ texlive-fonts-extra is not required.

tk0miya · 2018-11-22T14:12:56Z

At present, we only do testing to build testroot document to PDF, not our document.

On the other hand, PDF is always built in readthedocs.org on every commit. I know it's not testing, but might be enough.

jfbu · 2018-11-22T15:51:05Z

ah yes, and RTD build succeeded. As per the requirements for our continuous testing I see they are changed e.g. at circleci via a docker file sphinxdoc/docker-ci? I was worried I could not modify it. What is needed for LaTeX PDF builds if one tests the new features of this PR is texlive-lang-greek, texlive-lang-cyrillic, cm-super. For xelatex/lualatex, fonts-freefont-otf. The (big) Ubuntu package texlive-fonts-extra is not needed.

BTW i need to do s/fonts-latex-extra/texlive-fonts-extra/g everywhere... (I got mixed up with Ubuntu package names, and I am not on Ubuntu at my locale...)

edit: actually I mixed up only in my comments, not in this PR

tk0miya · 2018-11-22T16:48:27Z

Yes, Circle CI uses sphinxdoc/docker-ci image (https://hub.docker.com/r/sphinxdoc/docker-ci/).
And it comes from https://github.com/sphinx-doc/docker-ci .
Circle CI has always fetch our image on each testing. So it would be nice if we can make it slim :-)

jfbu · 2018-11-22T17:39:34Z

@tk0miya afaict, texlive-fonts-extra (in https://github.com/sphinx-doc/docker-ci, which is a big dependency) could be replaced by
fonts-freefont-otf (to support xelatex and lualatex builds on master branch), and fonts-lmodern (to support them on 1.8 branch). (I think no platex builds are done)

edit I see lmodern (which includes fonts-lmodern) is already listed as dependency (the dependencies are in alphabetical order but for some reason the name of this one does not start with texlive- so I missed it). After 2.0 release it will become superfluous I think and fonts-freefont-otf will be enough.

If we add some Greek or Cyrillic letters to the testroot (still with language English), then it would be needed to add texlive-lang-greek, texlive-lang-cyrillic and cm-super for the PDFLaTeX build of testroot to succeed. But this is a bit redundant because RTD build of our own documentation tests this now.

tk0miya · 2018-11-24T10:38:34Z

We should test for Greek and Cyrillic at least once. And I think we've already done it on rtd. So it's okay to keep as is.

jfbu added type:bug builder:latex labels Nov 16, 2018

jfbu added this to the 2.0.0 milestone Nov 16, 2018

jfbu commented Nov 16, 2018

View reviewed changes

Fix flake8

f56431b

jfbu force-pushed the latex_greek_cyrillic_letters branch from 53bf824 to 2b88140 Compare November 17, 2018 00:42

Load CMU by filename, not font name, for XeLaTeX

6283324

jfbu force-pushed the latex_greek_cyrillic_letters branch from 2b88140 to 6283324 Compare November 17, 2018 00:46

jfbu commented Nov 17, 2018

View reviewed changes

tk0miya approved these changes Nov 20, 2018

View reviewed changes

LaTeX: font choices for Greek and Cyrillic support, and (Xe|Lua)LaTeX

1edf2a4

new file: sphinx/texinputs/sphinxcyrillic.sty

5974c40

Forgot to add it to previous commit :(

jfbu merged commit 8412bdf into sphinx-doc:master Nov 21, 2018

jfbu deleted the latex_greek_cyrillic_letters branch November 22, 2018 08:47

jfbu added a commit that referenced this pull request Dec 13, 2018

Update CHANGES for PR #5645

6a40dd9

jfbu mentioned this pull request Dec 13, 2018

LaTeX: use \small fontsize in code-blocks with xelatex and lualatex #5768

Merged

tk0miya mentioned this pull request Dec 17, 2018

Tests fail if fonts are missing #5823

Closed

jfbu mentioned this pull request Dec 18, 2018

Latex writer should not translate Greek unicode, but use textgreek package #1682

Closed

github-actions bot locked as resolved and limited conversation to collaborators Aug 20, 2021

LaTeX: support for Greek and Cyrillic #5645

LaTeX: support for Greek and Cyrillic #5645

Conversation

jfbu commented Nov 16, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Nov 16, 2018 • edited Loading

Codecov Report

jfbu commented Nov 17, 2018 • edited Loading

jfbu left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jfbu commented Nov 21, 2018

jfbu commented Nov 21, 2018

jfbu commented Nov 21, 2018

jfbu commented Nov 22, 2018 • edited Loading

tk0miya commented Nov 22, 2018

jfbu commented Nov 22, 2018 • edited Loading

tk0miya commented Nov 22, 2018

jfbu commented Nov 22, 2018 • edited Loading

tk0miya commented Nov 24, 2018

jfbu commented Nov 16, 2018 •

edited

Loading

codecov bot commented Nov 16, 2018 •

edited

Loading

jfbu commented Nov 17, 2018 •

edited

Loading

jfbu commented Nov 22, 2018 •

edited

Loading

jfbu commented Nov 22, 2018 •

edited

Loading

jfbu commented Nov 22, 2018 •

edited

Loading