fix: make sorted __all__ and literals black-compatible (#2280)

[lord-haffi]

Resolves: #2280, #1815

Problem

--sort-reexports (and the # isort: list/dict/tuple/… comments) format the sorted result with stdlib pprint. Under --profile=black the result is single-quoted, packed in pprint's wrap style and lacks a trailing comma — output black immediately reformats, so isort + black never reach a fixpoint.

Fix

Format the sorted literal to be black-compatible: double quotes, a single line when it fits line_length, otherwise one item per line (vertical-hanging-indent) with a trailing comma when include_trailing_comma is set, keeping the literal's own bracket type. It's a fixed black/PEP 8 layout — it deliberately does not reproduce every multi_line_output mode for literals (VHI is what black uses).

isort's output for a long __all__ under --profile=black, before → after this fix:

-__all__ = ['Address', 'AliasAddress', 'BankAccountConnection', 'Certificate',
- 'ConcessionFee']
+__all__ = [
+    "Address",
+    "AliasAddress",
+    "BankAccountConnection",
+    "Certificate",
+    "ConcessionFee",
+]

Why drop pprint

pprint exposes only width; quote style, wrap mode and trailing comma are hardcoded to non-black output, so it could never satisfy --profile=black. It was the sole cause of the bug, so it's removed entirely — dict sorting now uses the same formatter and gets the same fix.

Also fixes a --check crash on multi-line __all__

Emitting multi-line __all__ exposed a latent bug in the --sort-reexports rollback: it wrote the __all__ opening line then did output_stream.seek(output_stream.tell() - reexport_rollback) to overwrite it with the sorted literal. In check mode the output is a no-op stream whose tell() is always 0, so the seek went to a negative position and raised ValueError: Negative seek position. The seek target is now clamped to 0; real seekable output is unaffected.

Note: sorted string literals are now double-quoted by default (isort has no quote setting); existing literal-sort test expectations are updated to match.

[lord-haffi]

@Helveg Please take a look. For me it would work. I deliberately did not try to merge it with the formatting logic for the imports. In my head it made sense to keep that seperated from this logic here even though it is very similar at some parts. What are your thoughts on this?

[lord-haffi]

I just realized, it probably addresses the same problem as #2573.

[lord-haffi]

Claude says:

Overlap with #2573

Heads-up that #2573 (thanks @HarperZ9) targets the same area and also closes #1815, so these two should be reconciled. This PR also resolves #1815 and is a superset of #2573 — I ran #2573 locally against the relevant cases to compare fairly:

Case	#2573	this PR (#2576)
# isort: tuple under --profile=black (#1815)	wraps to VHI but keeps single quotes (via pprint.pformat), so black --check still reformats it	double-quoted, black --check clean
Long __all__ under --profile=black	single-quoted VHI → not black-stable	black-stable (verified on a real 196-entry __all__)
isort --check on a multi-line __all__ preceded by imports	still raises ValueError: Negative seek position	fixed
# isort: dict	unchanged (still pprint)	same config-aware formatting

Root differences:

• Wrap sorted literals with black profile settings #2573 adds VHI wrapping only for the over-length + use_parentheses + VERTICAL_HANGING_INDENT path and otherwise keeps pprint, which emits single quotes — so the output isn't black-stable (black renormalizes the quotes).
• This PR removes pprint from literal sorting entirely and applies black's quote rule + single-line-or-VHI to every literal type, and additionally clamps the sort_reexports rollback seek in core.py (isort --check on a multi-line __all__ currently crashes — see the check-mode note above).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.39%. Comparing base (fd8bd07) to head (339aaba).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2576      +/-   ##
==========================================
+ Coverage   99.32%   99.39%   +0.06%     
==========================================
  Files          41       41              
  Lines        3111     3137      +26     
  Branches      674      680       +6     
==========================================
+ Hits         3090     3118      +28     
  Misses         11       11              
+ Partials       10        8       -2     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Helveg]

Hi, I'm not a maintainer of this repo so I can't merge it, but FWIW I approve these changes :D

[DanielNoord]

Thanks