bug: when using `import x.y.z as a`, Griffe thinks `a` is an alias for `x`, not `x.y.z`

[the-13th-letter]

Description of the bug

When importing modules within a package and using an alias, Griffe incorrectly determines the target of that alias to be the top-level package, not the leaf module.

To Reproduce

$ cd /tmp/XYZ
$ export PYTHONPATH=/tmp/XYZ
$ echo "import urllib.parse as parse" > bug_demo1.py
$ griffe dump -LDEBUG bug_demo1
INFO       Loading package bug_demo1
DEBUG      Found bug_demo1: loading
DEBUG      Loading path /tmp/XYZ/bug_demo1.py
INFO       Finished loading packages
{
  "bug_demo1": {
    "kind": "module",
    "name": "bug_demo1",
    "labels": [],
    "members": [
      {
        "kind": "alias",
        "name": "parse",
        "target_path": "urllib",
        "lineno": 1,
        "endlineno": 1
      }
    ],
    "filepath": "/tmp/XYZ/bug_demo1.py"
  }
}

Note here that the alias parse is resolving to urllib. It should resolve to urllib.parse instead.

$ echo "import xml.etree.ElementTree as ET" > bug_demo2.py
$ griffe dump -LDEBUG bug_demo2
INFO       Loading package bug_demo2
DEBUG      Found bug_demo2: loading
DEBUG      Loading path /tmp/XYZ/bug_demo2.py
INFO       Finished loading packages
{
  "bug_demo2": {
    "kind": "module",
    "name": "bug_demo2",
    "labels": [],
    "members": [
      {
        "kind": "alias",
        "name": "ET",
        "target_path": "xml",
        "lineno": 1,
        "endlineno": 1
      }
    ],
    "filepath": "/tmp/XYZ/bug_demo2.py"
  }
}

Here too the alias ET is resolving to xml instead of to xml.etree.ElementTree.

Full traceback

(Included above.)

Expected behavior

The aliases should resolve correctly.

Environment information

$ griffe --debug-info
- __System__: Linux-######
- __Python__: cpython 3.11.###
- __Environment variables__:
  - `PYTHONPATH`: `/tmp/XYZ`
- __Installed packages__:
  - `griffe` v0.42.1

(Griffe was installed from PyPI.)

Additional context

I originally encountered this issue while documenting (with mkdocs, mkdocstrings and mkdocs-autorefs) a module xxx.yyy.B that was importing xxx.yyy.A as A. mkdocstrings failed to resolve/expand references to A.SomeClass as xxx.yyy.A.SomeClass, and instead linked to the (non-existing) xxx.SomeClass instead.

[pawamoy]

Thanks for bug report @the-13th-letter :slightly_smiling_face: I was able to replicate. Will push a fix asap :smile:

[pawamoy]

Relevant commit: https://github.com/mkdocstrings/griffe/commit/e9867f78044a2a33b575e274224d3a4c16b62439

[pawamoy]

Should be fixed in v0.42.2 :slightly_smiling_face:

[the-13th-letter]

That was really fast. Thank you kindly.