[python] Fix and test setup.py flags to external shared objects

[jdblischak]

Issue and/or context:

• Closes [ci] dylib-copy logic appears to be broken on MacOS #2210
• Follow-up to [ci] Add MacOS job to test shared object copying #2220
• Address broken feedstock builds in tiledbsoma 1.8.0 pre-check TileDB-Inc/tiledbsoma-feedstock#91 and tiledbsoma 1.7.3 pre-check TileDB-Inc/tiledbsoma-feedstock#95

Changes:

• Restores the flags --tiledb and --libtiledbsoma that are passed to setup.py

Notes for Reviewer:

Here's what happened. PR #1937 fixed the shared object copying when an external libtiledb or libtiledbsoma is provided via the env vars TILEDB_PATH or TILEDBSOMA_PATH. Unfortunately, it broke the command-line flags --tiledb and --libtiledbsoma. It overwrites them with None. This is why the tiledbsoma-feedstock nightlies immediately started failing after it was merged, which is the only place where we still build tiledbsoma-py with setup.py.

We assumed there was a macOS-specific problem because 1) the linux-64 feedstock build continued to pass, and 2) #1937 only tested on linux and not macOS. However, after I added a macOS job in #2220, we knew it was not specific to macOS since macOS also properly copies the objects when using the env vars.

I figured it out by investigating the differences between linux and macOS in setup.py. It turns out that linux has a fall-back method to search for a system installation of libtiledbsoma. This is the only reason that the linux-64 feedstock build passed.

TileDB-SOMA/apis/python/setup.py

Lines 146 to 147 in 793818d

	if os.name == "posix" and sys.platform != "darwin":
	path = find_libtiledbsoma_full_path_on_linux(lib_name)

In order to catch this type of regression in this repository, I added an extra job to my python-so-copying workflow that confirms the flags --tiledb and --libtiledbsoma are functional. To emulate the feedstock build, I delete cmake from the runner. Thus if tiledbsoma-py is unable to use the command-line flags and tries to fall back to building libtiledbsoma from source with cmake, it will fail (which I confirmed on a branch in my fork by temporarily reverting my fixes to setup.py).

See also #2219 for tracking

[johnkerl]

See also #2210 for troubleshooting context.

[johnkerl]

🚢

[codecov]

Codecov Report

Merging #2221 (0a9b015) into main (793818d) will decrease coverage by 6.47%.
The diff coverage is n/a.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2221      +/-   ##
==========================================
- Coverage   78.49%   72.02%   -6.47%     
==========================================
  Files         137      103      -34     
  Lines       10683     6875    -3808     
  Branches      215      215              
==========================================
- Hits         8386     4952    -3434     
+ Misses       2198     1824     -374     
  Partials       99       99              

Flag	Coverage Δ
libtiledbsoma	67.44% <ø> (ø)
python	?
r	74.69% <ø> (ø)

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

Components	Coverage Δ
python_api	∅ <ø> (∅)
libtiledbsoma	48.85% <ø> (ø)