[FBumann]

Nothing to revert — this is the rest of #511. 15 lines.

#511 landed the positional attach and pays for the ordinal with a Python dict built from the dimension table, plus a replace_strict per row per dim. Since #541 that lookup is redundant for most dimensions: encode_dimensions makes every string dimension an Enum over its labels in ordinal order, so the physical code already is the ordinal.

profiled/l build   1.240 s -> 0.901 s

Measured warm, median of 5, on main at 8efd207 against this branch.

The two PRs were written a day apart and neither could see the other — #511 predates #541 landing. This is the part of the idea it could not have: to_physical() where it had a dictionary, for the two of profiled's three dimensions that are strings. A non-string dimension (snapshot, here) still pays the dictionary, which is built from the small side — one entry per label, not per row.

Why there is nothing else to change

No behaviour change and no new gate. The ordinal is the same integer by either route, so #511's gate still says exactly what it said, and its three tests — the oracle, the shuffled input, both refusals — pin this unchanged. 942 passed, ruff clean, pyrefly 0 errors.

Context

Together with #511 this is the full effect the idea was worth: profiled/l build 1.325 s joined → 0.901 s attached, a 0.68x. #511 alone got about a fifth of that, because the dictionary it needed was most of what the attach saved.

Supersedes #566, which proposed the same thing as a wholesale replacement of #511 before #511 merged. That framing is obsolete now that the mechanism is on main; this is the same measured win as a 15-line delta instead.

🤖 Generated with Claude Code

[codspeed]

Merging this PR will regress 2 benchmarks

⚠️ 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

⚡ 2 improved benchmarks
❌ 2 regressed benchmarks
✅ 10 untouched benchmarks
⏩ 38 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Memory	test_emit[profiled-s-lpspec-lp]	24 MB	27 MB	-11.32%
❌	Memory	test_emit[sector-s-lpspec-lp]	4.5 MB	5 MB	-10.03%
⚡	Memory	test_emit[sector-s-lpspec-highs]	4.9 MB	4.1 MB	+20.14%
⚡	Memory	test_emit[transport-s-lpspec-lp]	20.7 MB	18.4 MB	+12.5%

Tip

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

---

_{Comparing perf/bound-ordinal-from-enum (191fe2e) with main (8efd207)²}

Footnotes

1. 38 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on main (15c9776) during the generation of this report, so 8efd207 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩