refactor: pkg_resources -> importlib.metadata

[cwegener]

pkg_resources is in the process of being deprecated in favor of
importlib.metadata and importlib.resources

importlib.metadata has been available in the standard library since 3.8

And as of 3.10 the importlib.metadata API is no longer provisional

In order to support Python version older than 3.10, the
importlib_metadata backport is used instead of the actual
importlib.metadata API.

[john-bodley]

Thanks @cwegener for the PR. Note you'll also have to re-compile the Python dependencies which ensures we have a frozen set of Python packages (and sub-packages) for testing purposes. Please refer to the CONTRIBUTING.md file with regards to how to do this.

[cwegener]

Thanks @cwegener for the PR. Note you'll also have to re-compile the Python dependencies which ensures we have a frozen set of Python packages (and sub-packages) for testing purposes. Please refer to the CONTRIBUTING.md file with regards to how to do this.

Hi John. I tried, but pip-compile-multi seems to just blissfully ignore the install_dependencies from setup.py

Do you have any hints on how to make pip-compile-multi read the install_dependencies collection from setup.py?

[john-bodley]

@cwegener I gather you mean install_requirements rather than install_dependencies? Note that pip-compile-multi should adhere to the requirements. Would you mind sharing the command(s) you ran?

[cwegener]

@cwegener I gather you mean install_requirements rather than install_dependencies? Note that pip-compile-multi should adhere to the requirements. Would you mind sharing the command(s) you ran?

Sorry, yes. Typo. install_requires is what I'm referring to.
All good. It turns out that the result of pip-compile-multi -P importlib-metadata is simply a noop, because there is no change to the version of importlib-metadata that is already present.
And pip-compile-multi just doesn't need to know that importlib-medata is now a top-level require instead of a transitive one.

[cwegener]

I did just force push the 'noop' result from pip-compile-multi, which is basically just adding the comment into the requirements/base.txt for good measure.

[john-bodley]

Yes!

[cwegener]

On a different note, once I'm done with this pkg_resources migration (I'll publish the other part which is the migration to importlib.resources shortly) and setuptools should then no longer be a dependency,I hope to convince the superset contributors to migrate from setuptools+pip-compile-multi to hatch+pdm ... 😄

[codecov]

Codecov Report

Merging #24514 (5361e38) into master (ba3bdc0) will increase coverage by 0.01%.
The diff coverage is 94.73%.

❗ Current head 5361e38 differs from pull request most recent head c2c4485. Consider uploading reports for the commit c2c4485 to get more accurate results

@@            Coverage Diff             @@
##           master   #24514      +/-   ##
==========================================
+ Coverage   69.06%   69.07%   +0.01%     
==========================================
  Files        1901     1903       +2     
  Lines       74019    74027       +8     
  Branches     8116     8118       +2     
==========================================
+ Hits        51121    51136      +15     
+ Misses      20787    20780       -7     
  Partials     2111     2111              

Flag	Coverage Δ
hive	53.89% <100.00%> (-0.01%)	⬇️
mysql	79.41% <100.00%> (-0.01%)	⬇️
postgres	79.50% <100.00%> (-0.01%)	⬇️
presto	53.79% <100.00%> (-0.01%)	⬇️
python	83.48% <100.00%> (-0.01%)	⬇️
sqlite	78.00% <100.00%> (-0.01%)	⬇️
unit	54.65% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
.../plugins/plugin-chart-echarts/src/BoxPlot/index.ts	50.00% <ø> (ø)
...d/plugins/plugin-chart-echarts/src/Funnel/index.ts	50.00% <ø> (ø)
...nd/plugins/plugin-chart-echarts/src/Gauge/index.ts	50.00% <ø> (ø)
...nd/plugins/plugin-chart-echarts/src/Graph/index.ts	50.00% <ø> (ø)
.../plugin-chart-echarts/src/MixedTimeseries/index.ts	25.00% <ø> (ø)
...tend/plugins/plugin-chart-echarts/src/Pie/index.ts	50.00% <ø> (ø)
...nd/plugins/plugin-chart-echarts/src/Radar/index.ts	50.00% <ø> (ø)
...plugins/plugin-chart-echarts/src/Sunburst/index.ts	50.00% <ø> (ø)
.../plugin-chart-echarts/src/Timeseries/Area/index.ts	33.33% <ø> (ø)
...-chart-echarts/src/Timeseries/Regular/Bar/index.ts	33.33% <ø> (ø)
... and 23 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[cwegener]

@john-bodley Sorry about the missing import sort. I just pushed the fix for that. Can you run the GH actions again.

[john-bodley]

@cwegener with regards to,

I hope to convince the superset contributors to migrate from setuptools+pip-compile-multi to hatch+pdm ...

it might make sense to first create an issue (or possibly a SIP dependency on the complexity) regarding your intentions as it's probably worth discussing the pros/cons especially given the slew of Python package managers floating around.

[john-bodley]

@cwegener after rereading your PR description you mention that,

importlib.metadata has been available in the standard library since 3.8

and given we only support Python 3.9+ I was wondering why we need to use the backport.

[cwegener]

and given we only support Python 3.9+ I was wondering why we need to use the backport.

The backport is needed because the API of importlib.metadata is actually different in 3.9. The API has only been stable in 3.10.

Specifically, the entry_points() signature in 3.9 was still different, it didn't have the groups parameter in the 3.9 standard library. So the stable API is only available in the backport. (Found that out when I developed this change in 3.10 accidentally and then ran unit tests in 3.9)

[cwegener]

Specifically, the entry_points() signature in 3.9 was still different, it didn't have the groups parameter in the 3.9 standard library. So the stable API is only available in the backport. (Found that out when I developed this change in 3.10 accidentally and then ran unit tests in 3.9)

The best I can tell, the 'select' feature wasn't there in the entry_points()

I did just see though that there also is a more specific backport of JUST the select feature for entry_points() 😁

https://github.com/jaraco/backports.entry_points_selectable