[Prateek99-s]

Summary

Fixes #5270.
Closes #5275
Closes #5305

Formatting lines containing long chains of consecutive bracket trailers (such as subscript chains a[0][1]...[160]) suffered from quadratic runtime due to combinatorial omit trials in right_hand_split_with_omits.

Root Cause

Unlike attribute chains which have dot delimiters and are split cleanly in a single pass by delimiter_split, consecutive
subscript chains ][ have no intermediate delimiters and fall back to right_hand_split_with_omits.

generate_trailers_to_omit yields bracket pairs from the tail up to line_length. In long lines with many small trailers
(like [0]...), Black was performing up to 20 full right_hand_split calls per recursion level, only for
is_line_short_enough(lines[0]) to fail on every single omit candidate because each candidate split point was still hundreds of
characters from the start of the line.

Fix

1. Precompute cumulative prefix lengths for all leaves in line.leaves once at the start of right_hand_split_with_omits.
2. For each candidate omit, inspect the target opening_bracket the split would break on. If its prefix length from the start of the line alone exceeds mode.line_length, any head line produced by that split is mathematically guaranteed to exceed line_length. We safely skip executing right_hand_split for that omit.
3. Candidate omits that break on an opening bracket within mode.line_length (e.g. foo(long_args...)[key]) are retained and evaluated normally.
4. In the final fallback, reuse first_lines instead of redundantly re-running right_hand_split.
5. Output formatting is 100% invariant across all cases (diff-shades shows 0 changes).

Benchmark

• $n = 100$ subscripts: ~2.8s -> ~0.35s (~8x speedup, 2,574 RHS calls -> 156 RHS calls)
• $n = 160$ subscripts: ~9.5s -> ~0.35s (~27x speedup)
• $n = 300$ subscripts: >30s -> ~0.95s (~30x speedup)

Test Plan

• Added test_subscript_chain_avoids_quadratic_rhs_splits in tests/test_black.py to verify that right_hand_split invocations remain linearly bounded ($O(n)$) rather than quadratic ($O(n^2)$).
• Full unit test suite passes.
• All pre-commit checks pass.

[github-actions]

diff-shades results comparing this PR (0fcc9da) to main (a3cb9e1):

--preview style: no changes

--stable style: no changes

---

What is this? | Workflow run | diff-shades documentation

[cobaltt7]

This PR also has a significantly malformed description, please fix it

[Prateek99-s]

Updated the test to wrap right_hand_split and assert that split calls are bounded linearly (mock_rhs.call_count < 2 * n). Without pruning unreachable omits, n = 100 made 2,574 calls (O(n²)), whereas with prefix pruning it stays bounded at 156 calls (O(n)). This avoids fragile wall-clock timings on CI while guaranteeing quadratic backtracking cannot regress.

[cobaltt7]

Thanks!