Waiting on epics-base/setuptools_dso#31

When building this and running tests with local copies of setuptools_dso, epicscorelibs and pvxslibs, I noticed that there is a circular import error when attempting to run the nose2 tests.

❯ python -m nose2 p4p                                21s  p4p  14:12:16
E
======================================================================
ERROR: p4p (nose2.loader.LoadTestsFailure)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/scratch/controls/dev/p4p/src/p4p/__init__.py", line 14, in <module>
    from .wrapper import Value, Type
  File "/scratch/controls/dev/p4p/src/p4p/wrapper.py", line 5, in <module>
    from . import _p4p
ImportError: cannot import name '_p4p' from partially initialized module 'p4p' (most likely due to a circular import) (/scratch/controls/dev/p4p/src/p4p/__init__.py)

----------------------------------------------------------------------
Ran 1 test in 0.000s

FAILED (errors=1)

The seems to occur no matter the Python version (at least >=3.8) .

ImportError: cannot import name '_p4p' from partially initialized module 'p4p' (most likely due to a circular import) (/scratch/controls/dev/p4p/src/p4p/init.py)

I recall errors like this when accidentally running nose from within a partially built git clone. eg. if $PWD/p4p exists. This is why some of the GHA jobs will cd dist before running nose.

https://github.com/mdavidsaver/p4p/blob/70b030de2f30cf690bb317429f59bd092e65420e/.github/workflows/build.yml#L357-L359

Looking again at this PR. Is it necessary to replace all of the quotes?

I am not opposed to code "beautification", but would much prefer these changes be separated. At least in separate commits, preferably separate PRs.

I recall errors like this when accidentally running nose from within a partially built git clone.

Following the Python Build commands listed in p4p/.github/workflows/build.yml seems to have fixed this issue.
The ones:

          python setup.py sdist --formats=gztar

and

          python -m pip wheel -v -w dist --no-index --no-deps \
           --no-build-isolation --no-use-pep517 \
           --build-option --plat-name=${{ matrix.piparch }} \
           $PRE \
           dist/*.tar.*

          cd dist
          python -m pip install p4p*.whl
          python -m ci_core_dumper exec python -m nose2 -v p4p

or anything similar are not mentioned in the docs for "Building" at all. It might be worth adding these so that anyone else trying to build from source doesn't get confused if these errors show when trying to run the unittests. Unless of course I missed an obvious step in the guide...

Looking again at this PR. Is it necessary to replace all of the quotes?

Oh whoops, that must have been my vscode auto-formatter and I didn't realise... I can change them back

I recall errors like this when accidentally running nose from within a partially built git clone.

Forgive me for belaboring this point. Did switching to a different directory avoid this failure?

... not mentioned in the docs ...

These docs do not cover building wheels from a manually created git checkout. Only the pip --no-binary ... recipe to build from downloaded source. This was a deliberate decision on my part back in ~2018. I thought the process was too complex for reasonable "user" documentation. For that matter, I was having trouble getting pip to do what I wanted. This situation has evolved since then. Maybe there is now a sensible recipe to document? Does pip have a command for generating source tars?

Does pip have a command for generating source tars?

That comes from the build system not pip. It looks like you are using setuptools so I expect python setup.py sdist to do what you want.

@mdavidsaver are you happy with these changes?