As we want to call Python code directly at the Sphinx extension,
convert get_feat.pl to Python.
The code was made to be (almost) bug-compatible with the Perl
version, with two exceptions:
1. Currently, Perl script outputs a wrong table if arch is set
to a non-existing value;
2. the ReST table output when --feat is used without --arch
has an invalid format, as the number of characters for the
table delimiters are wrong.
Those two bugs were fixed while testing the conversion.
Additionally, another caveat was solved:
the output when --feat is used without arch and the feature
doesn't exist doesn't contain an empty table anymore.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <03c26cee1ec567804735a33047e625ef5ab7bfa8.1763492868.git.mchehab+huawei@kernel.org>
Now that we have tools/lib/python for our Python modules, turn them into
proper packages with a single namespace so that everything can just use
tools/lib/python in sys.path. No functional change.
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <20251110220430.726665-3-corbet@lwn.net>
"scripts/lib" was always a bit of an awkward place for Python modules. We
already have tools/lib; create a tools/lib/python, move the libraries
there, and update the users accordingly.
While at it, move the contents of tools/docs/lib. Rather than make another
directory, just put these documentation-oriented modules under "kdoc".
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <20251110220430.726665-2-corbet@lwn.net>
Move this tool out of scripts/ to join the other documentation tools; fix
up a couple of erroneous references in the process.
It's worth noting that this script will fail badly unless one has a
PYTHONPATH referencing scripts/lib/abi.
Reviewed-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
The scripts for managing the features docs are found in three different
directories; unite them all under tools/docs and update references as
needed.
Reviewed-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Quoth Mauro:
This series should probably be called:
"Move the trick-or-treat build hacks accumulated over time
into a single place and document them."
as this reflects its main goal. As such:
- it places the jobserver logic on a library;
- it removes sphinx/parallel-wrapper.sh;
- the code now properly implements a jobserver-aware logic
to do the parallelism when called via GNU make, failing back to
"-j" when there's no jobserver;
- converts check-variable-fonts.sh to Python and uses it via
function call;
- drops an extra script to generate man pages, adding a makefile
target for it;
- ensures that return code is 0 when PDF successfully builds;
- about half of the script is comments and documentation.
I tried to do my best to document all tricks that are inside the
script. This way, the docs build steps is now documented.
It should be noticed that it is out of the scope of this series
to change the implementation. Surely the process can be improved,
but first let's consolidate and document everything on a single
place.
Such script was written in a way that it can be called either
directly or via a Makefile. Running outside Makefile is
interesting specially when debug is needed. The command line
interface replaces the need of having lots of env vars before
calling sphinx-build:
$ ./tools/docs/sphinx-build-wrapper --help
usage: sphinx-build-wrapper [-h]
[--sphinxdirs SPHINXDIRS [SPHINXDIRS ...]] [--conf CONF]
[--builddir BUILDDIR] [--theme THEME] [--css CSS] [--paper {,a4,letter}] [-v]
[-j JOBS] [-i] [-V [VENV]]
{cleandocs,linkcheckdocs,htmldocs,epubdocs,texinfodocs,infodocs,mandocs,latexdocs,pdfdocs,xmldocs}
Kernel documentation builder
positional arguments:
{cleandocs,linkcheckdocs,htmldocs,epubdocs,texinfodocs,infodocs,mandocs,latexdocs,pdfdocs,xmldocs}
Documentation target to build
options:
-h, --help show this help message and exit
--sphinxdirs SPHINXDIRS [SPHINXDIRS ...]
Specific directories to build
--conf CONF Sphinx configuration file
--builddir BUILDDIR Sphinx configuration file
--theme THEME Sphinx theme to use
--css CSS Custom CSS file for HTML/EPUB
--paper {,a4,letter} Paper size for LaTeX/PDF output
-v, --verbose place build in verbose mode
-j, --jobs JOBS Sets number of jobs to use with sphinx-build
-i, --interactive Change latex default to run in interactive mode
-V, --venv [VENV] If used, run Sphinx from a venv dir (default dir: sphinx_latest)
the only mandatory argument is the target, which is identical with
"make" targets.
The call inside Makefile doesn't use the last four arguments. They're
there to help identifying problems at the build:
-v makes the output verbose;
-j helps to test parallelism;
-i runs latexmk in interactive mode, allowing to debug PDF
build issues;
-V is useful when testing it with different venvs.
When used with GNU make (or some other make which implements jobserver),
a call like:
make -j <targets> htmldocs
will make the wrapper to automatically use POSIX jobserver to claim
the number of available job slots, calling sphinx-build with a
"-j" parameter reflecting it. ON such case, the default can be
overriden via SPHINXDIRS argument.
Visiable changes when compared with the old behavior:
When V=0, the only visible difference is that:
- pdfdocs target now returns 0 on success, 1 on failures.
This addresses an issue over the current process where we
it always return success even on failures;
- it will now print the name of PDF files that failed to build,
if any.
In verbose mode, sphinx-build-wrapper and sphinx-build command lines
are now displayed.
Improve the suggestions algorithm by using get_close_matches() if
no suggestions with the same name are found. As we're now building
a dict, when the name is identical, but on a different domain,
the search is O(1), making it a lot faster.
The get_close_matches is also fast, as there is just one loop,
instead of 3.
This can be useful to detect typos on references, with could
be the base of a futuere extension that will handle ref unmatches
for the entire build, allowing someone to find typos and fix them.
As difflib and get_close_matches are there since the early
Python 3.x days, we don't need to handle any extra dependencies
to use it.
We're keeping the default values for the search, e.g. n=3, cutoff=0.6.
With that, we now have things like:
$ make SPHINXDIRS="userspace-api/media" htmldocs
...
include/uapi/linux/videodev2.h:199: WARNING: Invalid xref: c:type:`v4l2_memory`. Possible alternatives:
c:type:`v4l2_meta_format` (from v4l/dev-meta)
c:type:`v4l2_rect` (from v4l/dev-overlay)
c:type:`v4l2_area` (from v4l/ext-ctrls-image-source) [ref.missing]
...
include/uapi/linux/videodev2.h:1985: WARNING: Invalid xref: c:type:`V4L.v4l2_queryctrl`. Possible alternatives:
std🏷️`v4l2-queryctrl` (from v4l/vidioc-queryctrl)
std🏷️`v4l2-query-ext-ctrl` (from v4l/vidioc-queryctrl)
At the first example, it was not a typo, but a symbol that doesn't
seem to be properly documented. The second example points to
v4l2-queryctrl, which is a close match for the symbol.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <7365feb74cbdd6b982c87baf5863360ab98cf727.1759329363.git.mchehab+huawei@kernel.org>
Specially when using c::namespace, it is not hard to break
a reference by forgetting to add a domain. Also, different
cases and using "-"/"_" the wrong way are typical cases that
people often gets wrong.
We might use a more complex logic here to also check for typos,
but let's keep it plain, simple.
This is enough to get thos exeptions from media controller:
.../include/uapi/linux/media.h:26: WARNING: Invalid xref: c:type:`media_device_info`. Possible alternatives:
c:type:`MC.media_device_info` (from mediactl/media-ioc-device-info)
.../include/uapi/linux/media.h:149: WARNING: Invalid xref: c:type:`media_entity_desc`. Possible alternatives:
c:type:`MC.media_entity_desc` (from mediactl/media-ioc-enum-entities)
.../include/uapi/linux/media.h:228: WARNING: Invalid xref: c:type:`media_link_desc`. Possible alternatives:
c:type:`MC.media_link_desc` (from mediactl/media-ioc-enum-links)
.../include/uapi/linux/media.h:235: WARNING: Invalid xref: c:type:`media_links_enum`. Possible alternatives:
c:type:`MC.media_links_enum` (from mediactl/media-ioc-enum-links)
.../include/uapi/linux/media.h:212: WARNING: Invalid xref: c:type:`media_pad_desc`. Possible alternatives:
c:type:`MC.media_pad_desc` (from mediactl/media-ioc-enum-links)
.../include/uapi/linux/media.h:298: WARNING: Invalid xref: c:type:`media_v2_entity`. Possible alternatives:
c:type:`MC.media_v2_entity` (from mediactl/media-ioc-g-topology)
.../include/uapi/linux/media.h:312: WARNING: Invalid xref: c:type:`media_v2_interface`. Possible alternatives:
c:type:`MC.media_v2_interface` (from mediactl/media-ioc-g-topology)
.../include/uapi/linux/media.h:307: WARNING: Invalid xref: c:type:`media_v2_intf_devnode`. Possible alternatives:
c:type:`MC.media_v2_intf_devnode` (from mediactl/media-ioc-g-topology)
.../include/uapi/linux/media.h:341: WARNING: Invalid xref: c:type:`media_v2_link`. Possible alternatives:
c:type:`MC.media_v2_link` (from mediactl/media-ioc-g-topology)
.../include/uapi/linux/media.h:333: WARNING: Invalid xref: c:type:`media_v2_pad`. Possible alternatives:
c:type:`MC.media_v2_pad` (from mediactl/media-ioc-g-topology)
.../include/uapi/linux/media.h:349: WARNING: Invalid xref: c:type:`media_v2_topology`. Possible alternatives:
c:type:`MC.media_v2_topology` (from mediactl/media-ioc-g-topology)
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <4c75d277e950e619ea00ba2dea336853a4aac976.1759329363.git.mchehab+huawei@kernel.org>
The code here was meant to handle 3 functions:
1. allow having a separate conf.py file, per subdir;
2. generate a list of latex documents.
3. set "subproject" tag if SPHINXDIRS points to a subdir.
We don't have (1) anymore, and (3) is now properly handled
entirely inside conf.py.
So, only (3) is still needed, and this is a single-line change
at conf.py.
So, drop it, moving the remaining code to conf.py.
While here, drop a duplicated $(RUSTDOC) command-line argument.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <ec998f9f268a401ca6aa36e3221d39c97efeccaa.1758361087.git.mchehab+huawei@kernel.org>
The ErrorString() and SafeString() docutils functions were helpers meant to
ease the handling of encodings during the Python 3 transition. There is no
real need for them after Python 3.6, and docutils 0.22 removes them,
breaking the docs build
Handle this by just injecting our own one-liner version of ErrorString(),
and removing the sole SafeString() call entirely.
Reported-by: Zhixu Liu <zhixu.liu@gmail.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Message-ID: <87ldmnv2pi.fsf@trenco.lwn.net>
While the original code came from the Sphinx Include class,
such class is monolithic: it has only one function that does
everything, and 3 variables that are used:
- required_arguments
- optional_arguments
- option_spec
So, basically those are the only members that remain from
the original class, but hey! Those are the same vars that every
other Sphinx directive extension has to define!
In summary, keeping inheritance here doesn't make much sense.
Worse than that, kernel-include doesn't support the current set
of options that the original Include class has, but it also
has its own set of options.
So, let's fill in the argument vars with what it does
support, dropping the rest.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/a9f2eebf11c6b0c3a2e3bf42e71392cdfd2835d1.1755872208.git.mchehab+huawei@kernel.org
It is best to point to the original line of code that generated
an error than to point to the beginning of a directive.
Add support for it. It should be noticed that this won't work
for literal or code blocks, as Sphinx will ignore it, pointing
to the beginning of the directive. Yet, when the output is known
to be in ReST format, like on TOC, this makes the error a lot
more easier to be handled.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/a0953af8b71e64aaf2e0ba4593ad39e19587d50a.1755872208.git.mchehab+huawei@kernel.org
kernel_include extension was originally designed to be used by the
media comprehensive uAPI documentation, where, instead of simpler
kernel-doc markups, the uAPI documentation is enriched with a larger
text, with images, complex tables, graphs, etc.
There, we wanted to include the much simpler yet documented .h
file.
This extension is needed to include files from other parts of the
Kernel tree outside Documentation, because the original Sphinx
include tag doesn't allow going outside of the directory passed
via sphinx-build command line.
Yet, the cross-references themselves to the full documentation
were using a perl script to create cross-references against the
comprehensive documentation.
As the perl script is now converted to Phython and there is a
Python class producing an include-compatible output with cross
references, add two optional arguments to kernel_include.py:
1. :generate-cross-refs:
If present, instead of reading the file, it calls ParseDataStructs()
class, which converts C data structures into cross-references to
be linked to ReST files containing a more comprehensive documentation;
Don't use it together with :start-line: and/or :end-line:, as
filtering input file line range is currently not supported.
2. :exception-file:
Used together with :generate-cross-refs:. Points to a file containing
rules to ignore C data structs or to use a different reference name,
optionally using a different reference type.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/efc39c8e54a2056ae2fdb94d5006fcb19e227198.1755872208.git.mchehab+huawei@kernel.org
When printing --help, we'd like the name of the files
from __doc__ to match the displayed positional arguments at
both usage and argument description lines.
Use a custom formatter class to convert ``foo`` into ANSI SGR
code to bold the argument, if is TTY, and adjust the help
text to match the argument names.
Here on Plasma, that makes it display it colored, wich is
really cool. Yet, I opted for SGR, as the best is to follow
the terminal color schema for bold.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/2c1e61d1fb1b2a2838b443beee89c1528831997f.1755872208.git.mchehab+huawei@kernel.org
When the Kernel started to use Sphinx, we had to come up with
a solution to parse media headers. On that time, we didn't have
much experience with Sphinx extensions. So, we came up with our
own script-based solution that were basically implementing a
set of rules we used to have at the Makefile.
Convert it to Python, keeping it bug-compatible with the
original script.
While here, try to better document it.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/ae5cfa8dff37e280cc9493fc95a51cd0cc0ba127.1755872208.git.mchehab+huawei@kernel.org
As reported by Donald, this code:
rst_parser = RSTParser()
rst_parser.parse('\n'.join(result), document)
breaks line parsing. As an alternative, I tested a variant of it:
rst_parser.parse(result, document)
but still line number was not preserved. As Donald noted,
standard Parser classes don't have a direct mechanism to preserve
line numbers from ViewList().
So, instead, let's use a mechanism similar to what we do already at
kerneldoc.py: call the statemachine mechanism directly there.
I double-checked when states and statemachine were introduced:
both were back in 2002. I also tested doc build with docutils 0.16
and 0.21.2. It worked with both, so it seems to be stable enough
for our needs.
Reported-by: Donald Hunter <donald.hunter@gmail.com>
Closes: https://lore.kernel.org/linux-doc/m24ivk78ng.fsf@gmail.com/T/#u
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Instead of printing line numbers from the temp converted ReST
file, get them from the original source.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Add a simple sphinx.Parser to handle yaml files and add the
the code to handle Netlink specs. All other yaml files are
ignored.
The code was written in a way that parsing yaml for different
subsystems and even for different parts of Netlink are easy.
All it takes to have a different parser is to add an
import line similar to:
from doc_generator import YnlDocGenerator
adding the corresponding parser somewhere at the extension:
netlink_parser = YnlDocGenerator()
And then add a logic inside parse() to handle different
doc outputs, depending on the file location, similar to:
if "/netlink/specs/" in fname:
msg = self.netlink_parser.parse_yaml_file(fname)
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Reviewed-by: Donald Hunter <donald.hunter@gmail.com>
The logic there which adds a dependency note to Sphinx cache
is not taking into account that the build dir may not be
the source dir. This causes a performance regression:
$ time make O=/tmp/foo SPHINXDIRS=admin-guide htmldocs
[OUTDATED]
Added: set()
Changed: {'abi-obsolete', 'abi-removed', 'abi-stable-files', 'abi-obsolete-files', 'abi-stable', 'abi', 'abi-removed-files', 'abi-testing-files', 'abi-testing', 'gpio/index', 'gpio/obsolete'}
Removed: set()
All docs count: 385
Found docs count: 385
real 0m11,324s
user 0m15,783s
sys 0m1,164s
To get the root cause of the problem (ABI files reported as changed),
I used this changeset:
diff --git a/Documentation/conf.py b/Documentation/conf.py
index e8766e689c1b..ab486623bd8b 100644
--- a/Documentation/conf.py
+++ b/Documentation/conf.py
@@ -571,3 +571,16 @@ def setup(app):
"""Patterns need to be updated at init time on older Sphinx versions"""
app.connect('config-inited', update_patterns)
+ app.connect('env-get-outdated', on_outdated)
+
+def on_outdated(app, env, added, changed, removed):
+ """Track cache outdated due to added/changed/removed files"""
+ print("\n[OUTDATED]")
+ print(f"Added: {added}")
+ print(f"Changed: {changed}")
+ print(f"Removed: {removed}")
+ print(f"All docs count: {len(env.all_docs)}")
+ print(f"Found docs count: {len(env.found_docs)}")
+
+ # Just return what we have
+ return added | changed | removed
Reported-by: Akira Yokosawa <akiyks@gmail.com>
Closes: https://lore.kernel.org/linux-doc/c174f7c5-ec21-4eae-b1c3-f643cca90d9d@gmail.com/
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Tested-by: Akira Yokosawa <akiyks@gmail.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/e25673d87357457bc54ee863e97ff8f75956580d.1752752211.git.mchehab+huawei@kernel.org
Docutils emits a deprecation warning when the set_class() element method is
used; that warning disappears into the ether, but it also causes a crash
with docutils 0.19.
Avoid the deprecated function and just append directly to the "classes"
attribute like the documentation says instead.
Reported-by: Akira Yokosawa <akiyks@gmail.com>
Tested-by: Akira Yokosawa <akiyks@gmail.com>
Closes: https://lore.kernel.org/de7bae91-3200-481f-9db2-c0dc382c91dd@gmail.com/
Fixes: d6d1df92c2 ("docs: automarkup: Mark up undocumented entities too")
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
The automarkup code generates markup and a cross-reference link for
functions, structs, etc. for which it finds kerneldoc documentation.
Undocumented entities are left untouched; that creates an inconsistent
reading experience and has caused some writers to go to extra measures to
cause the markup to happen.
Mark up detected C entities regardless of whether they are documented.
Reviewed-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
