Re: [PATCH] scripts: kernel-doc: add warning for comment not following kernel-doc syntax

[Aditya Srivastava <yashsri421@gmail.com>]

On 1/4/21 1:02 am, Jonathan Corbet wrote:
> Aditya Srivastava <yashsri421@gmail.com> writes:
>
>> On 29/3/21 7:26 pm, Jonathan Corbet wrote:
>>> Aditya Srivastava <yashsri421@gmail.com> writes:
>>>
>>>> Currently, kernel-doc start parsing the comment as a kernel-doc comment if
>>>> it starts with '/**', but does not take into account if the content inside
>>>> the comment too, adheres with the expected format.
>>>> This results in unexpected and unclear warnings for the user.
>>>>
>>>> E.g., running scripts/kernel-doc -none mm/memcontrol.c emits:
>>>> "mm/memcontrol.c:961: warning: expecting prototype for do not fallback to current(). Prototype was for get_mem_cgroup_from_current() instead"
>>>>
>>>> Here kernel-doc parses the corresponding comment as a kernel-doc comment
>>>> and expects prototype for it in the next lines, and as a result causing
>>>> this warning.
>>>>
>>>> Provide a clearer warning message to the users regarding the same, if the
>>>> content inside the comment does not follow the kernel-doc expected format.
>>>>
>>>> Signed-off-by: Aditya Srivastava <yashsri421@gmail.com>
>>>> ---
>>>>  scripts/kernel-doc | 17 +++++++++++++----
>>>>  1 file changed, 13 insertions(+), 4 deletions(-)
>>>
>>> This is definitely a capability we want, but I really don't think that
>>> we can turn it on by default - for now.  Experience shows that if you
>>> create a blizzard of warnings, nobody sees any of them.  How many
>>> warnings does this add to a full docs build?
>>>
>>
>> Hi Jonathan, here's the diff I have created for the warnings before
>> and after the changes:
>> https://github.com/AdityaSrivast/kernel-tasks/blob/master/random/kernel-doc/kernel_doc_comment_syntax.txt
>>
>> Around ~1320 new warnings of this type are added to the kernel tree,
>> and around ~1580 warnings are removed.
>
> So I finally got around to looking at this again...  How did you
> generate that file?
>

I ran scripts/kernel-doc -none on all the files in the kernel tree
before and after appying the changes, and then generated their diff to
find the warnings removed and added.

> I tried applying the patch and doing a normal full htmldocs build and
> got all of four warnings:
>
>   ./include/linux/seqlock.h:829: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
>    * DEFINE_SEQLOCK(sl) - Define a statically allocated seqlock_t
>   ./fs/jbd2/journal.c:1391: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
>    *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
>   ./fs/jbd2/journal.c:1422: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
>    *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
>   ./include/linux/dcache.h:309: warning: This comment starts with '/**', but isn't a kernel-doc comment. Refer Documentation/doc-guide/kernel-doc.rst
>    *      dget, dget_dlock -      get a reference to a dentry
>

I think there should be more warnings. For e.g., running kernel-doc
-none ./drivers/usb/mtu3/mtu3.h gives these warnings:

./drivers/usb/mtu3/mtu3.h:75: warning: This comment starts with '/**',
but isn't a kernel-doc comment. Refer
Documentation/doc-guide/kernel-doc.rst
./drivers/usb/mtu3/mtu3.h:86: warning: This comment starts with '/**',
but isn't a kernel-doc comment. Refer
Documentation/doc-guide/kernel-doc.rst
./drivers/usb/mtu3/mtu3.h:143: warning: This comment starts with
'/**', but isn't a kernel-doc comment. Refer
Documentation/doc-guide/kernel-doc.rst

> Two observations:
>
>  - This is not an awful lot of warnings - not the blizzard I had
>    feared.  At this level, I think we can just merge the patch and
>    then, hopefully, fix those cases.
>
>  - All of the warned-about places are *attempts* to write real kerneldoc
>    comments, they just got the syntax wrong in one way or another.  It's
>    probably not worth the effort to try to detect this case - the
>    warning is enough to draw attention to the comment in question.
>

I agree. Above are some of the cases which are not getting detected by
this patch.
This may be so as I am only allowing the function syntax as mentioned
in the rst file, i.e., "^\s*\*\s*([\w\s]+?)(\(\))?\s*([-:].*)?$" or
("* foo(\(\))? - description")

I probably need to check for pointers as well and other similar case(s).
Maybe I should design a separate check for functions than assigning
$decl_type = 'function' in the first check.

What do you think?

Thanks
Aditya