fix(rag)!: preserve string boundaries when indexing a string-under-axis Ragged

[d-laub]

Closes #71.

Integer indexing on a string-under-axis Ragged concatenated a whole group into one blob, dropping the per-string boundaries already sitting in str_offsets. s.lengths[0] == 2 but len(s[0]) == 3.

Two sites, not one

• Ragged.__getitem__ integer branch — the one reported.
• _getitem_record_rows integer branch — worse, and the one GenVarLoader actually hits. RaggedVariants is a record Ragged whose alt/ref are opaque-string fields sharing offsets with numeric start/ilen, so a peeled row returned a concatenated S1 char array (not even bytes) sitting next to a per-variant numeric array in the same dict:

rec[0] -> {'alt':   array([b'A', b'G', b'G'], dtype='|S1'),   # 3 chars
           'start': array([1, 2], dtype=int32)}                # 2 variants

_getitem_record_rows_r2 delegates to the first site and inherits the fix; it's covered by a test.

Fix

Both now peel to the standalone opaque-string layout (offsets == [], str_offsets set, shape == (k,)), which Spec C already defines as the zero-real-level special case of string-under-axis — so this is what "peel one real level" already means in the layout algebra, not a new return type. Shared helper _peel_string_row, mirroring the narrowing _slice_contig_string already does for slices. Data stays zero-copy; only the small (k+1,) offsets slice is copied, rebased to zero as is_contiguous requires.

Result:

lengths     : [2 1]
numeric[0]  : [10 20]        len 2
string [0]  : [b'A', b'GG']  len 2
zip aligned : [(10, b'A'), (20, b'GG')]

This makes zip(rv.start[0][h], rv.alt[0][h]) correct in GenVarLoader (mcvickerlab/GenVarLoader#330).

Alternatives considered and rejected in the spec: an object array of bytes (same ergonomics but allocates k Python objects per row access, against the repo's hot-path rule), to_chars()[i] (forces consumers to re-assemble strings), and raising NotImplementedError (indexing is far more central than the to_numpy()/to_packed() that set that precedent).

Breaking

The integer index returns a Ragged of strings rather than one concatenated bytes. b"".join(s[i]) recovers the old value. major_version_zero is set, so this bumps 0.21.2 → 0.22.0.

test_string_under_axis_integer_index pinned the old behavior but used exactly one string per group, so it could never distinguish the two interpretations — that's why the bug survived. It has been rewritten.

Testing

16 new/rewritten tests: the issue repro, numeric/string length parity, multi-dim (batch, ploidy, ~variants), record rows with a multi-byte (indel) entry so 1-byte coincidence can't mask a regression, zero-copy, empty group, to_chars() agreement, negative/OOB indexing, and a regression guard that the standalone string case still returns bytes.

25 of 26 test files pass. tests/test_coords.py fails to collect with OSError: Could not find/load shared object file from llvmlite — pre-existing environment breakage, verified identical on unmodified main.

Spec: docs/superpowers/specs/2026-07-27-ragged-string-getitem-design.md
Plan: docs/superpowers/plans/2026-07-27-ragged-string-getitem.md

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will regress 1 benchmark

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 1 regressed benchmark
✅ 10 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	test_bench_ragged_cres	978.1 µs	1,465.7 µs	-33.27%
⚡	test_bench_baseline_ragged_short_alleles	4.2 ms	2.7 ms	+57.94%
⚡	test_bench_baseline_ragged_cres	8.2 ms	6.2 ms	+33.01%
⚡	test_bench_dense_batch	3.4 ms	2.9 ms	+17.24%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing worktree-issue-71-string-under-axis-getitem (e970469) with main (c7f4a5a)}