CLN: use absolute imports throughout the codebase

[idanmiara]

Related to #1711
Replace relative imports (i.e. from . import xyz) with absolute imports (from shapely import xyz) and run isort as recommended by PEP8 (https://peps.python.org/pep-0008/#imports): Absolute imports are recommended, as they are usually more readable and tend to be better behaved.
This helps much with debugging.

[idanmiara]

Hi,
This PR ready for review.
I've rebased and cleaned commit history.
Thanks!

[coveralls]

Pull Request Test Coverage Report for Build 4161382935

• 71 of 72 (98.61%) changed or added relevant lines in 17 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.3%) to 87.085%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
shapely/geometry/conftest.py	0	1	0.0%

Totals
Change from base Build 4161378179:	0.3%
Covered Lines:	2387
Relevant Lines:	2741

---

💛 - Coveralls

[idanmiara]

@jorisvandenbossche please review

[idanmiara]

@jorisvandenbossche can his be included in 2.0.1 ?

[jorisvandenbossche]

Personally, I have a (slight, but not strong) preference for keeping the relative imports, because I like using those (there are also several other projects that do that as well, like numpy and scikit-learn, although those projects might fit more in the "complex package layouts" bucket compared to shapely)

There is certainly some inconsistency right now internally, because the usage of this has come from merging the pygeos code, while what remained from the original shapely code uses absolute imports.
So we should certainly choose one way or the other.

cc @caspervdw since you started using this originally in pygeos

[idanmiara]

@jorisvandenbossche Thanks for your comments.

Sorry, I'm not sure I follow what's the advantage of using relative imports (You noted that some other popular projects use them, but why is it an advantage? many other popular projects use absolute imports exclusively).

The disadvantages that I know of, for using relative imports are:

1. It's a bad practice (https://peps.python.org/pep-0008/#imports).
2. You can't copy and paste some code from a module and test it without modifying the imports.
3. BUG/Regression: Running doctests fail because of relative imports #1711

[idanmiara]

Hi @jorisvandenbossche any further thoughts about this PR and with respect to #1711 ?

[idanmiara]

@brendan-ward @mwtoews @caspervdw any thoughts about this?

[brendan-ward]

Have not reviewed the PR (will try to do so in the coming days), but I agree with the direction; I stumbled with the relative imports in pygeos several times while trying to test code there, and the precedent for absolute imports was already set in shapely.

[brendan-ward]

Thanks for working on this @idanmiara ! Only spotted a couple of issues (not sure why isort is putting unittest into wrong position), but otherwise this looks good to me.

It brings a nice consistency to the imports, since the precedent was already established in shapely < 2.0

[idanmiara]

Thanks for working on this @idanmiara ! Only spotted a couple of issues (not sure why isort is putting unittest into wrong position)

Please see #1681 which will fix that issue.

[idanmiara]

@brendan-ward please approve this PR as we are addressing your requested change in #1743 and #1681

[idanmiara]

@jorisvandenbossche @brendan-ward
I think we are getting close with #1681 so I'm waiting for #1681 to be merged so I can rebase this PR - as there would probably be more conflicts once it's merged.

[jorisvandenbossche]

Sorry, I'm not sure I follow what's the advantage of using relative imports (You noted that some other popular projects use them, but why is it an advantage? many other popular projects use absolute imports exclusively).

I was just giving those example to show that shapely is not some kind of oddity in using relative imports. It's not bad practice, it's just a subjective preference on development workflows (PEP8 recommends absolute imports, but also says that explicit relative ones are an "acceptable alternative").

But consistency within a code base is of course important.

I'm waiting for #1681 to be merged so I can rebase this PR - as there would probably be more conflicts once it's merged.

Given that #1681 also includes updating the unittest import, I removed the changes to the /tests subdirectory in this PR, then there might not be conflicts to resolve.

[jorisvandenbossche]

Thanks @idanmiara

[idanmiara]

@jorisvandenbossche thanks for your review!

[mwtoews]

I'm late to say thanks, looks good @idanmiara ! I generally prefer absolute imports within modules and appreciate consistency. A modern tool of interest is https://github.com/MarcoGorelli/absolufy-imports