isort only if imports at the top of the module were edited

[akaihola]

Fixes #72:

The problem with applying isort modifications "surgically" is in the fact that isort may (and often does) change the order of lines.

Could we consider all changes by isort as one contiguous diff chunk, and apply it if any line in edited_linenums falls inside that chunk? Is there an edge case where isort only modifies part of the import lines, and a user edit happens to an import outside that part? I can't see a problem there. The important thing is that either all or none of the isort modifications are applied.

The changes are pretty involved. Here's a summary for reviewers:

src/darker/__main__.py:

• some variable renames for clarity:
  • path_in_repo -> relative_path_in_rev2
  • src -> absolute_path_in_rev2
  • relative_path -> relative_path_in_rev2
• _isort_and_blacken_single_file() and _isort_and_blacken_single_file() now accept
  • a pre-created EditedLinenumsDiffer object instead of creating its own
  • a list of files to skip when reformatting using Black, instead of just a boolean for the single file

src/darker/git.py:

• rename get_rev1_path() to get_path_in_repo()

src/darker/import_sorting.py:

• apply_isort() now
  • accepts a pre-created EditedLinenumsDiffer object instead of creating its own
  • calls into helper functions refactored out of it:
    • isort arguments are created in _build_isort_args()
    • isort is invoked in _call_isort_code()
  • generates a diff for the result of isort
• _diff_overlaps_with_edits() does the meat of the new functionality: it checks whether the range of user edits touches the range of reformatted imports

src/darker/diff.py:

• diff_chunks() now combines diff_and_get_opcodes() and opcodes_to_chunks() behind one call

[akaihola]

@yajo, I noticed you gave a thumbs up for this idea in #72. I could include this change in release 1.5.0 – would you be able to do a code review of my pull request here?

[akaihola]

I hopefully managed to clarify the changes a bit in the description of this PR to make reviewing easier. @roniemartinez would you happen to have a bit of time to do a sanity check on this patch?

[yajo]

I've got no idea about the code sorry 😕

However I've tested this and it doesn't seem to do any isort at all. I'm not sure if I'm testing well, though. AFAICS only black runs:

Kooha-04-08-2022-07-31-04.mp4

[akaihola]

@yajo, thanks for testing this! You do need the -i/--isort option to sort imports as well. It's not enabled by default.

Is there something we could do to improve documentation so this becomes more obvious?

[akaihola]

@roniemartinez, if the review feels too messy, I could also separate out e.g. the variable renames, function refactorings and EditedLinenumsDiffer changes into their own PRs which don't change functionality. Would this help?

[roniemartinez]

LGTM

[akaihola]

Thanks @roniemartinez! We're getting close to a 1.5.0 release!