chore: add type annotations to cloudinit.user_data

[Mohith1612]

Proposed Commit Message

chore: add type annotations to cloudinit.user_data

_do_include only computes include_once_fn when include_once_on is
set, so nest the cached-file lookup inside that branch and fall
through to the fetch when nothing was read. mypy can then see
include_once_fn as a str at the isfile and load_text_file calls.

_explode_archive assigns msg a MIMEText in one branch and a MIMEBase
in the other. MIMEText subclasses MIMEBase, so declare msg as
MIMEBase.

Drop cloudinit.user_data from the mypy override list in
pyproject.toml.

Refs GH-5445

Additional Context

Refs GH-5445. This module is marked as a priority in the issue checklist.

Removing it from the override list surfaces three errors:

user_data.py:233: Argument 1 to "isfile" has incompatible type
                  "Any | None"                          [arg-type]
user_data.py:234: Argument 1 to "load_text_file" has incompatible type
                  "Any | None"                          [arg-type]
user_data.py:303: Incompatible types in assignment (expression has type
                  "MIMEBase", variable has type "MIMEText") [assignment]

The first two. include_once_fn is only assigned inside
if include_once_on:, but the lookup below sat outside that branch, so mypy
still saw it as Optional at the isfile and load_text_file calls. Nesting
the lookup inside the branch it depends on resolves that without any extra
guard.

The fall-through changes from else to if content is None, which is
equivalent: the only assignment in the nested branch is
util.load_text_file(), which is annotated -> str and so never returns
None. The fetch therefore runs in exactly the same cases as before, and a
cached file that happens to be empty still suppresses the fetch, as it did
previously.

I left the later if include_once_on and resp.ok() guard alone. It does not
produce an error and changing it would be noise.

The third. In _explode_archive, msg gets a MIMEText in the text
branch and a MIMEBase in the other, so mypy pins it to MIMEText from the
first assignment. MIMEText is a subclass of MIMEBase and both are already
imported, so a bare msg: MIMEBase declaration is enough. No assignment
changes.

Test Steps

No runtime behaviour change, so no new tests. tests/unittests/test_data.py
already covers the include and include-once paths through _do_include, and
the archive handling through _explode_archive.

$ python -m pytest tests/unittests/test_data.py -q
34 passed

$ tox -e py3
5743 passed, 5 skipped, 13 xfailed, 10 warnings in 173.08s

$ tox -e check_format
ruff: All checks passed!
pylint: Your code has been rated at 10.00/10
black: 595 files would be left unchanged.
isort: Skipped 8 files
mypy: Success: no issues found in 591 source files
congratulations :)

Merge type

• Squash merge using "Proposed Commit Message"
• Rebase and merge unique commits. Requires commit messages per-commit each referencing the pull request number (#<PR_NUM>)

[holmanb]

Thanks - one comment.

[holmanb]

It would read more cleanly if these were just nested together. I think this is safe since isfile returns a string, so the None check is redundant anyway.

[Mohith1612]

Understood, will update the code!

[Mohith1612]

Nested as suggested, thanks. The redundant None check is gone:

if include_once_on:
    include_once_fn = self._get_include_once_filename(include_url)
    if os.path.isfile(include_once_fn):
        content = util.load_text_file(include_once_fn)
if content is None:
    try:

One note on the fall-through, since it moved from else to if content is None. That is equivalent because the only assignment in the nested branch is util.load_text_file(), which is annotated -> str and never returns None, so the fetch still runs in exactly the same cases. A cached file that happens to be empty still suppresses the fetch, as before.

Small correction on the reasoning, though it does not change the outcome: os.path.isfile returns a bool, not a string. The thing that made the None check redundant is _get_include_once_filename, which returns an os.path.join result and so is always a str once we are inside the include_once_on branch.

I also rewrote the commit message to describe the current state rather than what the code used to do, per @blackboxsw's note on #7021.

tox -e py3, tox -e check_format and tests/unittests/test_data.py are all green on the updated branch.

[holmanb]

This has conflicts, please fix.

[Mohith1612]

Rebased onto main, conflict resolved.

The conflict was in the GH-5445 override list in pyproject.toml. Main had dropped cloudinit.sources.helpers.vultr in #7051 while this branch drops cloudinit.user_data, and the two removals landed adjacent to each other. Both entries are gone in the resolved version, and nothing else in that list moved.

The code change is unchanged from the version you reviewed. tox -e py3, tox -e check_format and tests/unittests/test_data.py are all green on the rebased branch.