fix: replace pkg_resources with importlib.metadata (#1816)

[jbbqqf]

Summary

Replace the runtime dependency on pkg_resources (removed in setuptools 81)
with the stdlib importlib.metadata, so import data_profiling no longer
crashes on fresh installs that pull in modern setuptools.

Fixes #1816 — Replace deprecated pkg_resources with importlib.metadata in profile_report.py.

Context

pkg_resources was removed from setuptools in v81 (changelog entry).
After Aug 2025, fresh pip install fg-data-profiling resolutions on a
clean environment routinely crash at import time with
ModuleNotFoundError: No module named 'pkg_resources', because
src/data_profiling/profile_report.py:11 does import pkg_resources
unconditionally and src/data_profiling/utils/versions.py keeps a
pkg_resources fallback.

That fallback was meant to support Python < 3.8 (where
importlib.metadata did not exist), but pyproject.toml already pins
requires-python = ">=3.10,<3.14" since #1778, so the fallback branch is
dead code under any supported runtime.

The original issue reporter validated the change locally with a 2195-test
green run on Python 3.13.

Changes

• src/data_profiling/profile_report.py: drop import pkg_resources,
  use importlib.metadata.version("Pillow") in ProfileReport.to_file.
  Wrap the lookup in a PackageNotFoundError guard (Pillow is an indirect
  dependency, so a trimmed environment without it should warn-and-continue,
  not crash). Parse numeric components defensively so pre-release versions
  like "11.0.0a1" don't blow up int().
• src/data_profiling/utils/versions.py: drop the pkg_resources fallback
  branch. importlib.metadata.version is unconditional now.
• tests/issues/test_issue1816.py (new): three regression guards.

The targeted comments inline explain why the branch was kept (Pillow may
be absent) and why the numeric-parts loop exists (pre-release segments) so
a reviewer reading the diff cold doesn't have to re-derive the reasoning.

Reproduce BEFORE/AFTER yourself (copy-paste)

# --- one-time setup ---
git clone https://github.com/Data-Centric-AI-Community/fg-data-profiling.git /tmp/repro && cd /tmp/repro
python3 -m venv /tmp/repro-venv
source /tmp/repro-venv/bin/activate
pip install --upgrade 'setuptools>=81' pip   # the env that triggers the bug

# --- BEFORE (origin/develop) ---
git checkout origin/develop
pip install -q -e . pytest
python3 -c "import data_profiling"
# Expected: ModuleNotFoundError: No module named 'pkg_resources'
git fetch https://github.com/jbbqqf/fg-data-profiling.git feat/1816-importlib-metadata-pillow
git checkout FETCH_HEAD -- tests/issues/test_issue1816.py
python3 -m pytest tests/issues/test_issue1816.py -v 2>&1 | tail -5
# Expected: collection error / import error before any test runs

# --- AFTER (this PR) ---
git fetch https://github.com/jbbqqf/fg-data-profiling.git feat/1816-importlib-metadata-pillow
git checkout FETCH_HEAD
pip install -q --force-reinstall --no-deps -e .
python3 -c "import data_profiling; print('import OK')"
# Expected: prints "import OK"
python3 -m pytest tests/issues/test_issue1816.py -v 2>&1 | tail -5
# Expected: 3 passed

The block above is a single bash run a reviewer can paste verbatim. The
only thing that changes between BEFORE and AFTER is the checked-out git
ref — that's the load-bearing piece of evidence.

What I ran locally

• pytest tests/issues/test_issue1816.py -v → 3/3 passed.
• pytest tests/unit/test_describe.py tests/unit/test_html_export.py tests/issues/test_issue1816.py → 33/33 passed.
• pytest tests/issues/ → 43 passed, 3 skipped, 1 unrelated failure
  (test_issue147 — needs pyarrow which isn't installed in my
  reproducible-build env; same failure on origin/develop, not introduced
  by this change).
• Python 3.13.12, setuptools 82.0.1 (>= 81, the version that triggers the
  bug). pkg_resources is genuinely absent in this env, which is why the
  regression test is meaningful.
• black --check clean on the three modified files.

GHA on the fork is not enabled, so the upstream pull-request.yml
workflow will be the first end-to-end signal across the supported Python
matrix; the pyproject pins >=3.10,<3.14, so the fix is in scope for
every supported version.

Edge cases tested

#	Scenario	Input	Expected	Verified by
1	setuptools >= 81 (no pkg_resources)	import data_profiling	succeeds; pkg_resources never imported	test_profile_report_module_does_not_import_pkg_resources
2	versions helper used elsewhere	import data_profiling.utils.versions	succeeds; no pkg_resources	test_versions_helper_does_not_import_pkg_resources
3	end-to-end report write	ProfileReport(df).to_file("...") on a 3-row df	writes HTML, sys.modules clean	test_to_file_runs_without_pkg_resources
4	Pillow absent	importlib.metadata.PackageNotFoundError	warning skipped, write continues	inline try/except PackageNotFoundError (covered by test 3 if Pillow optional)
5	Pillow pre-release version	"11.0.0a1"	numeric part loop short-circuits cleanly	inline numeric_parts loop (no separate test, exercised whenever Pillow isn't a final release)

Risk / blast radius

• Additive at the import layer — no public API change.
• The Pillow-version warning is preserved, just driven from
  importlib.metadata. pkg_resources and importlib.metadata agree on
  the version string for any package installed via PEP 517.
• One behavior change: if Pillow is somehow not installed, the previous
  code would have raised pkg_resources.DistributionNotFound (an
  uncaught crash from inside to_file); the new code warns nothing and
  continues. That is the safer default for a Pillow optional path. Worth
  a maintainer's nod, hence the inline comment.

Release note

fix: replace runtime pkg_resources lookups with importlib.metadata so
ProfileReport imports cleanly under setuptools >= 81. (#1816)

Upstream PR checklist (from .github/PULL_REQUEST_TEMPLATE/pull_request_template.md)

• make lint (black) — clean on the 3 modified files. (Pre-existing
  isort drift on profile_report.py that already exists on develop is
  not addressed here to keep the diff scoped.)
• make docs — n/a, no doc changes.
• make test — pytest tests/issues/test_issue1816.py -v → 3/3 passed
  on Python 3.13 with setuptools 82 (the env that triggers the bug).
• make examples — n/a, no example changes.

---

PR drafted with assistance from Claude Code. The change was reviewed
manually against the upstream codebase on develop and the setuptools
v81 release notes. The reproducer block above was used during development;
it is the same one a reviewer can paste verbatim.