DEP: Deprecate private namespaces in scipy.optimize

[czgdp1807]

Reference issue

gh-14360

What does this implement/fix?

Additional information

cc: @rgommers

[rgommers]

The second commit needs a couple of changes to setup.py - there are still moduleTNC.so and minpack2.so left. Could you give those an underscore as well?

One other minor issue I spotted.

[rgommers]

Also, CI is still broken.

[czgdp1807]

there are still moduleTNC.so and minpack2.so left. Could you give those an underscore as well?

Ah! okay. No worries. I will do that.

[czgdp1807]

I am fixing mypy errors. Will push after verifying locally.

[rgommers]

The doc build is still broken, trying to use optimize.optimize

[czgdp1807]

I get the following warning in docs,

docstring of _odrpack.odr:16: WARNING: py:obj reference target not found: ODR
docstring of _odrpack.odr:28: WARNING: py:obj reference target not found: ODR
/Users/czgdp1807/scipy_project/scipy/build/testenv/lib/python3.9/site-packages/scipy/optimize/_root.py:docstring of scipy.optimize._root.root:86: WARNING: py:obj reference target not found: nonlin
/Users/czgdp1807/scipy_project/scipy/build/testenv/lib/python3.9/site-packages/scipy/odr/__init__.py:docstring of scipy.odr:29:<autosummary>:1: WARNING: py:obj reference target not found: scipy.odr.ODR
/Users/czgdp1807/scipy_project/scipy/build/testenv/lib/python3.9/site-packages/scipy/optimize/__init__.py:docstring of scipy.optimize:396: WARNING: py:mod reference target not found: scipy.optimize.nonlin
release/0.11.0-notes.rst:85: WARNING: py:obj reference target not found: nonlin
release/1.6.0-notes.rst:47: WARNING: py:obj reference target not found: scipy.odr.ODR

And the following message,

build finished with problems, 14 warnings.
make: *** [html-scipyorg] Error 1
Traceback (most recent call last):
  File "/Users/czgdp1807/scipy_project/scipy/runtests.py", line 566, in <module>
    main(argv=sys.argv[1:])
  File "/Users/czgdp1807/scipy_project/scipy/runtests.py", line 215, in main
    subprocess.run(cmd, check=True)
  File "/Users/czgdp1807/mambaforge/envs/scipy-dev/lib/python3.9/subprocess.py", line 528, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['make', '-Cdoc', 'PYTHON="/Users/czgdp1807/mambaforge/envs/scipy-dev/bin/python"', 'html-scipyorg', 'latex', 'SPHINXOPTS="-j2"']' returned non-zero exit status 2.

Are they okay?

[czgdp1807]

I think other test failures are unrelated?

[rgommers]

I'll have a look, the docs warnings are slightly odd - unclear to me why the ODR ones are showing up now for example.

[czgdp1807]

I think in release notes we use module names before this deprecation. Hence, now nonlin doesn't exist because it's converted to _nonlin. Should we update release notes as well?

[rgommers]

I think in release notes we use module names before this deprecation. Hence, now nonlin doesn't exist because it's converted to _nonlin. Should we update release notes as well?

yes we should, just to avoid doc build warnings

[rgommers]

Hmm, I fixed most things, but nonlin is a little annoying. While there is nothing in optimize.nonlin that is not also present in optimize, it had its own submodule docs for some reason: http://scipy.github.io/devdocs/reference/optimize.nonlin.html#module-scipy.optimize.nonlin.

I'd prefer to get rid of that, moving to _nonlin and merging the docs content with the main optimize docs. But I want to flag that and make sure we're okay with that. The alternative is to add optimize.nonlin to the list of public API modules in http://scipy.github.io/devdocs/reference/index.html#api-definition and revert some of the changes in this PR (and keep the duplicate exposing of functions in the two namespaces). Anyone have any objections to getting rid of the nonlin namespace @scipy/scipy-core-team?

[larsoner]

Anyone have any objections to getting rid of the nonlin namespace @scipy/scipy-core-team?

None here, sounds like a good plan to me

[rgommers]

Only thumbs up for not making nonlin public. Also, I realized after writing this that all functions in it are legacy, root(..., method='xxx') is recommended instead.

CI is green except for the expected linting complaints, so in it goes. Thanks @czgdp1807, all.