Re: [PATCH] docs: automarkup.py: ignore exceptions when seeking for xrefs

[Mauro Carvalho Chehab]

Em Mon, 8 Jul 2019 09:23:36 -0600
Jonathan Corbet <corbet@xxxxxxx> escreveu:

> On Sat,  6 Jul 2019 13:28:42 -0300
> Mauro Carvalho Chehab <mchehab+samsung@xxxxxxxxxx> wrote:
> 
> > When using the automarkup extension with:
> > 	make pdfdocs
> > 
> > without passing an specific book, the code will raise an exception:
> > 
> > 	  File "/devel/v4l/docs/Documentation/sphinx/automarkup.py", line 86, in auto_markup
> > 	    node.parent.replace(node, markup_funcs(name, app, node))
> > 	  File "/devel/v4l/docs/Documentation/sphinx/automarkup.py", line 59, in markup_funcs
> > 	    'function', target, pxref, lit_text)
> > 	  File "/devel/v4l/docs/sphinx_2.0/lib/python3.7/site-packages/sphinx/domains/c.py", line 308, in resolve_xref
> > 	    contnode, target)
> > 	  File "/devel/v4l/docs/sphinx_2.0/lib/python3.7/site-packages/sphinx/util/nodes.py", line 450, in make_refnode
> > 	    '#' + targetid)
> > 	  File "/devel/v4l/docs/sphinx_2.0/lib/python3.7/site-packages/sphinx/builders/latex/__init__.py", line 159, in get_relative_uri
> > 	    return self.get_target_uri(to, typ)
> > 	  File "/devel/v4l/docs/sphinx_2.0/lib/python3.7/site-packages/sphinx/builders/latex/__init__.py", line 152, in get_target_uri
> > 	    raise NoUri
> > 	sphinx.environment.NoUri
> > 
> > This happens because not all references will belong to a single
> > PDF/LaTeX document.  
> 
> Interesting.  I'd like to understand better why the HTML builder doesn't do
> this...it seems like a bug in the latex builder somehow.

It took me a while to identify what part of the extension was causing the
build breakage with make pdfdocs, but this occurs upstream too, if you
try to build all documents.

The funny thing is that, if you try to build a single book, e. g.:

	make SPHINXDIRS=foo pdfdocs

this won't happen.

I didn't spend too much time trying to identify the exact breakage reason.
All I did were to add some prints inside latex/*.py while debugging it, in
order to get a rough idea about what was happening.

On several places, the code with calls "raise NoUri" is called, but the
caller silently ignores it (that happens, for example, when it parses
:ref:`genindex`).

What I suspect is that, when you do an html build - or when you build just
a single book with SPHINXDIRS=foo, all dependencies are either:

	- unsolved; or
	- solved within the same document/html URL

In other words, solved references will have a relative position within
a single documentation body. So, there won't be any need for a document
to refer to a symbol on some other document.

With PDF, a symbol defined under, let's say, "core-api" defines a
core_foo() function, and this function is referenced inside the "driver-api"
book. The automarkup.py will convert a driver-api core_foo() to:
:ref:`core_foo()`, instead of :doc:`core-api`, but core_foo label
doesn't exist within the "driver-api" book context.

As we're not using intersphinx extension, Sphinx doesn't have any
URL to convert the cross-reference in a way that, if one clicks at the
reference, it will open the referenced document at the proper page. 

So, it bails out.

That's said, I didn't try to use intersphinx in order to check if
LaTeX references to other documents would actually work.

> 
> > Better to just ignore those than breaking Sphinx build.
> > 
> > Fixes: d74b0d31ddde ("Docs: An initial automarkup extension for sphinx")
> > Signed-off-by: Mauro Carvalho Chehab <mchehab+samsung@xxxxxxxxxx>
> > ---
> >  Documentation/sphinx/automarkup.py | 9 +++++++--
> >  1 file changed, 7 insertions(+), 2 deletions(-)
> > 
> > diff --git a/Documentation/sphinx/automarkup.py b/Documentation/sphinx/automarkup.py
> > index b300cf129869..dba14374f269 100644
> > --- a/Documentation/sphinx/automarkup.py
> > +++ b/Documentation/sphinx/automarkup.py
> > @@ -55,8 +55,13 @@ def markup_funcs(docname, app, node):
> >                                            reftype = 'function',
> >                                            reftarget = target, modname = None,
> >                                            classname = None)
> > -            xref = cdom.resolve_xref(app.env, docname, app.builder,
> > -                                     'function', target, pxref, lit_text)
> > +
> > +            # When building pdf documents, this may raise a NoUri exception
> > +            try:
> > +                xref = cdom.resolve_xref(app.env, docname, app.builder,
> > +                                         'function', target, pxref, lit_text)
> > +            except:
> > +                xref = None  
> 
> So this absolutely needs to be "except sphinx.environment.NoUri".  I have
> seen catch-all "except" clauses paper over or otherwise hide too many
> problems over the years; I really try to avoid ever using them.

Makes sense to me. Feel free to change it when you apply it.

> 
> I want to look at this problem and understand it a bit better; then I'll
> probably end up applying this patch with the above tweak.

Btw, I got some other issues when changed Sphinx to include all books to the
pdf output on my documentation development tree:

	https://git.linuxtv.org/mchehab/experimental.git/commit/?h=convert_rst_renames_next_v2&id=cd72aaefc8b07ce7909be914e46dfb02bad5d86b

They're related to having nested tables, with got fixed by this patch:

	https://git.linuxtv.org/mchehab/experimental.git/commit/?h=convert_rst_renames_next_v2&id=fd0b22e391431f766e9be6f161fab9646c0dd9ca

Thanks,
Mauro