C, distinguish between ordinary identifiers and tag names

[jakobandersen]

Describe the bug
It should be possible to have a type with the same name as a struct/union/enum, as in the language.

To Reproduce
Steps to reproduce the behavior: as example, consider the following documentation (where every instance of struct could be replaced with union or enum instead):

.. c:struct:: foo
.. c:type:: struct foo foo

   No duplicate declaration warning.

- :c:struct::`foo`, link to the struct.
- :c:type::``foo``, link to the typedef.
- :c:any:``foo``, probably link to the type?
- Perhaps new feature: :c:any:``struct foo``, link to the struct.
- :c:expr:``foo*``, hyperlink ``foo`` to the typedef.
- c:expr:``struct foo*``, hyperlink ``foo`` (or all of ``struct foo``?) to the struct.

Expected behavior
See above.

Additional context

• See also Duplicate definition with C "typedef struct foo foo" breathe-doc/breathe#543.
• An implementation of this could perhaps be split into something backwards compatible, and then the breaking part for the next major version.

[Shamino0]

Hi. I reported this over at the Breathe site. Attached is a test-case that shows the problem. I've got four cases:

1. A C struct where the typedef and struct have the same name (processed via Doxygen and Breathe)
2. A C struct where the typedef and the struct have different names (processed via Doxygen and Breathe)
3. Sphinx markup declaring a C type and struct with the same name
4. Sphinx markup declaring a C type and struct with different names

Cases 2 and 4 work correctly - I can link to the types or the structs.

Cases 1 and 3 are broken - trying to reference the struct results in a link to the typedef.

I don't know what the cause is, but I did notice that the URLs for the links don't include the full domain. They are of the form .../html/foo.html/#c.foo_t. The top-level domain (:c:) is present, but the next-level domain (:type: or :struct:) is not, making it impossible to distinguish between the two.

testcase.zip

[jakobandersen]

(Just to avoid nomenclature confusion: "domain" in Sphinx means something specific, so for example "c", "cpp", and "python" are domains, without further subdivision.)

The cause is relatively simple: the C domain keeps track of declarations without partitioning names into tag and non-tag names. The links include an ID generated by the C domain (if you try other domains you will see very different patterns for the IDs), and you are absolutely correct that the ID needs to include tagging.

[jayaddison]

Building the Linux 6.13 Kernel documentation with Sphinx v8.1.3 emits a warning of the form WARNING: Duplicate C declaration, also defined at ... for identifier(s) named usb_string -- used variously in the codebase to refer either to a struct/datatype, or separately to a refer to a function. Would fixing this bug remove that duplication warning (by treating the identifiers as distinct)?

[cmarqu]

@jayaddison Looking at #12162 (comment) and the comment above, this PR would indeed remove the erroneous warning.