Re: [PATCH 00/34] docs: avoid using ReST :doc:`foo` tag

[Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx>]

Em Mon, 7 Jun 2021 21:34:58 -0300
Nícolas F. R. A. Prado <n@xxxxxxxxxxxxx> escreveu:

> Hi Mauro,
> 
> On Mon, Jun 07, 2021 at 09:34:22AM +0200, Mauro Carvalho Chehab wrote:
> > Em Sun, 6 Jun 2021 19:52:25 -0300
> > Nícolas F. R. A. Prado <n@xxxxxxxxxxxxx> escreveu:
> >   
> > > On Sat, Jun 05, 2021 at 09:08:36PM +0200, Mauro Carvalho Chehab wrote:  
> > > > Em Sat, 5 Jun 2021 12:11:09 -0300
> > > > Nícolas F. R. A. Prado <n@xxxxxxxxxxxxx> escreveu:
> > > >     
> > > > > Hi Mauro,
> > > > > 
> > > > > On Sat, Jun 05, 2021 at 03:17:59PM +0200, Mauro Carvalho Chehab wrote:    
> > > > > > As discussed at:
> > > > > > 	https://lore.kernel.org/linux-doc/871r9k6rmy.fsf@xxxxxxxxxxxx/
> > > > > > 
> > > > > > It is better to avoid using :doc:`foo` to refer to Documentation/foo.rst, as the
> > > > > > automarkup.py extension should handle it automatically, on most cases.
> > > > > > 
> > > > > > There are a couple of exceptions to this rule:
> > > > > > 
> > > > > > 1. when :doc:  tag is used to point to a kernel-doc DOC: markup;
> > > > > > 2. when it is used with a named tag, e. g. :doc:`some name <foo>`;
> > > > > > 
> > > > > > It should also be noticed that automarkup.py has currently an issue:
> > > > > > if one use a markup like:
> > > > > > 
> > > > > > 	Documentation/dev-tools/kunit/api/test.rst
> > > > > > 	  - documents all of the standard testing API excluding mocking
> > > > > > 	    or mocking related features.
> > > > > > 
> > > > > > or, even:
> > > > > > 
> > > > > > 	Documentation/dev-tools/kunit/api/test.rst
> > > > > > 	    documents all of the standard testing API excluding mocking
> > > > > > 	    or mocking related features.
> > > > > > 	
> > > > > > The automarkup.py will simply ignore it. Not sure why. This patch series
> > > > > > avoid the above patterns (which is present only on 4 files), but it would be
> > > > > > nice to have a followup patch fixing the issue at automarkup.py.      
> > > > > 
> > > > > What I think is happening here is that we're using rST's syntax for definition
> > > > > lists [1]. automarkup.py ignores literal nodes, and perhaps a definition is
> > > > > considered a literal by Sphinx. Adding a blank line after the Documentation/...
> > > > > or removing the additional indentation makes it work, like you did in your
> > > > > 2nd and 3rd patch, since then it's not a definition anymore, although then the
> > > > > visual output is different as well.    
> > > > 
> > > > A literal has a different output. I think that this is not the case, but I 
> > > > didn't check the python code from docutils/Sphinx.    
> > > 
> > > Okay, I went in deeper to understand the issue and indeed it wasn't what I
> > > thought. The reason definitions are ignored by automarkup.py is because the main
> > > loop iterates only over nodes that are of type paragraph:
> > > 
> > >     for para in doctree.traverse(nodes.paragraph):
> > >         for node in para.traverse(nodes.Text):
> > >             if not isinstance(node.parent, nodes.literal):
> > >                 node.parent.replace(node, markup_refs(name, app, node))
> > > 
> > > And inspecting the HTML output from your example, the definition name is inside
> > > a <dt> tag, and it doesn't have a <p> inside. So in summary, automarkup.py will
> > > only work on elements which are inside a <p> in the output.  
> > 
> > 
> > Yeah, that's what I was suspecting, based on the comments.
> > 
> > Maybe something similar to the above could be done also for some
> > non-paragraph data. By looking at:
> > 
> > 	https://docutils.sourceforge.io/docs/ref/doctree.html
> > 
> > It says that the body elements are:
> > 
> > 	admonition, attention, block_quote, bullet_list, caution, citation, 
> > 	comment, compound, container, danger, definition_list, doctest_block, 
> > 	enumerated_list, error, field_list, figure, footnote, hint, image, 
> > 	important, line_block, literal_block, note, option_list, paragraph, 
> > 	pending, raw, rubric, substitution_definition, system_message, 
> > 	table, target, tip, warning  
> 
> Ok, I went through each one by searching the term on [1] and inspecting the
> element to see if it contained a <p> or not. The vast majority did. These are
> the ones I didn't find there or didn't make sense:
> 
> 	comment
> 	container
> 	image
> 	pending
> 	raw
> 	substitution_definition
> 	system_message
> 	target
> 
> We can safely ignore them. And these are the ones that matter and don't have
> paragraphs:
> 
> 	1. literal_block
> 	2. doctest_block
> 	3. definition_list
> 	4. field_list
> 	5. option_list
> 	6. line_block
> 
> 1 and 2 are literals, so we don't care about them.
> 
> 3 is the one you noticed the issue with. It's worth mentioning that the
> definition term doesn't have a paragraph, but its definition does (as can be
> checked by inspecting [2]).
> 
> 4 is basically the same as 3, the rst syntax is different but the output is the
> same. That said, I believe we only use those to set options at the top of the
> file, like in translations, and I can't see automarkup being useful in there.
> 
> 5 is similar to 3 and 4, but the term is formatted using <kbd>, so it's like a
> literal and therefore not relevant.
> 
> 6 is useful just to preserve indentation, and I'm pretty sure we don't use it in
> the docs.
> 
> So in the end, I think the only contenders to be added to automarkup are
> definition lists, and even then I still think we should just substitute those
> definition lists with alternatives like you did in your patches. Personally I
> don't see much gain in using definitions instead of a simple paragraph. But if
> you really think it's an improvement in some way, it could probably be added to
> automarkup in the way you described.

Thank you for checking this!

Kernel docs use a lot definition lists. At the initial versions, it was
equivalent to:

	**Something to be written with emphasis**

	  Some description

Sphinx later changed the look-and-feel for the term, on html output, but
the thing is that:

	Something to be written with emphasis
	   Some description

looks a lot better when read as a text file.

Also, on some cases, the first notation doesn't work. The definition-list
was the only way I know that would allow to apply an emphasis to a literal
block.

We can avoid using Documentation/foo on description lists: the current 4 
cases where doc:`foo` are already addressed in this series, and the output
is acceptable.

Yet, I have a couple of concerns:

1. It might have some unknown places where a description list is used
   for Documentation/foo;
2. It is not trivial to identify if someone add Documentation/foo in
   the future;
3. I suspect that there are several places where functions and structs
   appear at the definition lists.

(1) can probably be checked with a multi-line grep. So, not a big
    problem;

(2) is something that would require someone to verify from time to
    time;

but (3) are harder to check and seems to be a valid use-case.

Due to (3), I think we should let automarkup to parse non-literal
terms on description lists. At very least it should emit a warning when
it won't be doing auto-conversions for known patterns at definition
lists (if doing that would generate false-positives).

Thanks,
Mauro