Closed Jellby closed 2 years ago
jfbu
commented
2 years ago Thanks for report
Also singlehtml target reveals some problems with your example. second footnote mark hyperlinks to first target.
With latexpdf (and using PR #10169) the mark of the first footnote refers to footnote page 3. (at least the mark syncs with the hypertarget location).
So both targets have problems, but of opposite nature...
jfbu
commented
2 years ago This patch appears to fix it:
$ git diff
diff --git a/sphinx/writers/latex.py b/sphinx/writers/latex.py
index a2905e807..84986af71 100644
--- a/sphinx/writers/latex.py
+++ b/sphinx/writers/latex.py
@@ -653,6 +653,8 @@ class LaTeXTranslator(SphinxTranslator):
short = ''
if any(node.findall(nodes.image)):
short = ('[%s]' % self.escape(' '.join(clean_astext(node).split())))
+ if self.sectionlevel == self.top_sectionlevel:
+ self.body.append(r'\sphinxstepscope' + CR)
try:
self.body.append(r'\%s%s{' % (self.sectionnames[self.sectionlevel], short))@tk0miya As you see above, the fix can be obtained by inserting \sphinxstepscope suitably. Here I do it at top level sectioning, I am not sure if same as inserting for each document included in a toctree. Maybe the \sphinxstepscope should be inserted per input file (but currently I don't know how one does that, because currently I have only parcellar understanding of whole of our codebase, acting only a LaTeX maintenance technician...). What do you think?
I tested using this one.rst in place of OP's, and similar two.rst
One
----
========== ===
Test [#a]_
========== ===
1 One
========== ===
.. [#a] Footnote one.
sous-section
^^^^^^^^^^^^
========== ===
Test [#b]_
========== ===
1b Oneb
========== ===
.. [#b] Footnote oneb.
Jellby
commented
2 years ago I fixed it myself by monkey-patching visit_start_of_file to issue \sphinxstepscope.
Jellby
commented
2 years ago And for the page detection, see https://gitlab.com/Molcas/OpenMolcas/-/blob/db84fd20654312ff9d224697916b0b6ab526bafc/doc/source/conf.py#L454 (\thepage cannot be trusted).
jfbu
commented
2 years ago I fixed it myself by monkey-patching
visit_start_of_fileto issue\sphinxstepscope. @Jellby Thanks!
Let's wait for @tk0miya opinion if issuing \sphinxstepscope at visit_start_of_file is the better way indeed (I had overlooked it in latex.py...). The two approaches can be merged, it does not cause harm to issue \sphinxstepscope.
And for the page detection, see https://gitlab.com/Molcas/OpenMolcas/-/blob/db84fd20654312ff9d224697916b0b6ab526bafc/doc/source/conf.py#L454 (
\thepagecannot be trusted).
I will look into it. The LaTeX shipout routines assigns a correct value to \thepage however I guess interaction with inserts can result in problems when dealing with footnotes. Did you encounter a specific problematic instance?
Jellby
commented
2 years ago I will look into it. The LaTeX shipout routines assigns a correct value to
\thepagehowever I guess interaction with inserts can result in problems when dealing with footnotes. Did you encounter a specific problematic instance?
The sample above is a problematic case. For me, there are two footnote marks in page 4, one says 1 and the other says Page 3, 1, but they are both links to the same footnote in page 3. The first one is wrong because it "believes" it is appearing on page 3 (the value of \thepage). See https://texfaq.org/FAQ-wrongpn for why \thepage cannot be trusted.
jfbu
commented
2 years ago @Jellby thanks can you raise separate issue about \thepage, I will make PR. The extra structure to fix this could then serve for adding backreferences from footnotes in page footers to footnote marks, but currently only for those footnotes using the \sphinxfootnotemark mechanism.
Jellby
commented
2 years ago I can raise a new issue, but note that the problem with \thepage is already in the OP's 2nd point.
jfbu
commented
2 years ago Allright, let's keep one issue. But there is one aspect whose resolution I will prefer to delay to later which is about colliding footnote marks originating in named or explicit mark-up in sources spread across files, a problem which is made worse from adding latex_theme = "howto" to your example. Achieving correct hyperlinks appears to me worthy enough first goal.
Also, I pointed out your example also has sub-optimal outcome with the singlehtml builder. So perhaps let's work out first a latex patch addressing at least some of the implied issues and then examine what is left later on and possibly open up new issues then.
jfbu
commented
2 years ago The whole \sphinxstepexplicit from #8832 has some core flaw anyhow. A fix to this issue will have to be incorporated as one aspect of the renewed thinking needed to fix #10188. This will probably entail changes both in latex.py and in sphinxpackagefootnote.sty latex code.
tk0miya
commented
2 years ago I fixed it myself by monkey-patching visit_start_of_file to issue \sphinxstepscope.
Looks good. The start_of_file nodes are visited before each item in toctree. So it's a good timing to increase the scope counter.
jfbu
commented
2 years ago I fixed it myself by monkey-patching visit_start_of_file to issue \sphinxstepscope.
Looks good. The
start_of_filenodes are visited before each item in toctree. So it's a good timing to increase the scope counter.
@tk0miya Ok, I will incorporate this to fix of #10188. I am expecting to drop entirely the business with sphinxexplicit. And this will go with dropping the recently added referred attribute, which will not be needed.
But there will be a change in our Sphinx LaTeX mark-up for \sphinxfootnotemark, i.e. sphinxpackagefootnote.sty will be modified to diverge from current behaviour inherited from LaTeX original \footnotemark. Ours will get as additional argument the refid. This is needed to be able to robustly construct working hyperlinks. And in so doing I will add structure to the sphinxpackagefootnote.sty footnotetext (and footnote) environments which will allow to robustly know if footnote mark and footnote are on same or diverging pages.
I started looking into this and I observe that footnotetext node have lostbackrefs attribute. Can you give me a pointer where I should modify the footnote transform to let footnotetext modes inherit backrefs attribute to the various footnotemark's referring it ?
I don't need theses backrefs immediately I think, but in future it will be nice to implement them, as what I need to add for fixing current issues opens up possibility with little effort in future to support back references like currently done for html output.
The sphinxscope, with modification at each input file, is a simple numerical alternative to work with docname attributes which require processing to provide safe latex labels ; besides I see that footnotemark and footnotetext nodes have lost docname. WIth sphinxscope we can do without it.
tk0miya
commented
2 years ago I just posted #10190 to fix this. Please reject it if you'll post another work.
I started looking into this and I observe that footnotetext node have lostbackrefs attribute. Can you give me a pointer where I should modify the footnote transform to let footnotetext modes inherit backrefs attribute to the various footnotemark's referring it ?
You can do it by modifying FootnoteCollector in sphinx/builders/latex/transforms.py. It generates footnottext nodes from the original footnote_definition nodes (see unrestrict() and depart_table()).
The sphinxscope, with modification at each input file, is a simple numerical alternative to work with docname attributes which require processing to provide safe latex labels ; besides I see that footnotemark and footnotetext nodes have lost docname. WIth sphinxscope we can do without it.
I think putting \sphinxscope per each input file is the first step. It would be better to change the location of the scope to each file, each part, or each arbitrary sectioning unit (via some new configuration). I'm not sure the future work is related to the docname. But it's better to consider the plan, if possible.
Describe the bug
These are a couple of issues with LaTeX footnotes, which may be related to #9529 and possibly partially fixed by #10169.
How to Reproduce
Default new project, with
latex_elements = {'fontpkg': r'\usepackage[notextcomp]{kpfonts}'}and these two files:one.rst
two.rst
Expected behavior
[#a], they're in different files).Page 3, 1.Your project
See "How to Reproduce" above
Screenshots
No response
OS
Linux Ubuntu 20.04
Python version
3.6.9
Sphinx version
4.4.0
Sphinx extensions
No response
Extra tools
No response
Additional context
No response