[jdatcmd]

@OffgridwithJD This ports test/parquet_nested_import.sh to pytest for #1341 (7 checks). The two pyarrow arms decline by their own names without pyarrow, as the shell's two skips do. The set hashes are taken in SQL, as pgc_set_hash does, with a failed query reading as a unique query_error().

One shell fix, in both twins. "import count matches" compared two counts with check, under which two empty readings agree. It is now check_num. The round-trips already gave each failed query its own hash, so they could not agree on two failures.

Measured on PG16, unless a row says otherwise:

run	main's shell	this shell	twin
clean	7/7	7/7 on each of PG15–19	7/7 on each of PG15–19
both counts read from missing tables	passes	not a measurement, so nothing was compared	—
readempty (the reader takes an empty Parquet list for a NULL array)	—	3 red: both round-trips and the pyarrow oracle; the counts stay green, as they should	the same 3

Two twin test names carry "parquet" where they would otherwise collide with test_arrow_nested_import.py's; the guards caught that.

Gates over d5c03a34:

• ledger gate rc=0; census 3607 never;
• suites_not_covered: 137 → 136;
• cluster_tests: 2045 → 2052, by collection;
• guards: 464.

🤖 Generated with Claude Code

[OffgridwithJD]

@jdatcmd Approving 66089227. The check_num fix is real, and the twin matches the shell under every mutant. Measured on PG17 (~/pg/17-m), per mutant make clean, then the shell, then the twin as shipped:

run	shell	twin
head, PG17 and PG18	7/7	7/7; SAME 0 reds
readempty (yours, rebuilt: an empty Parquet list read back as a NULL array)	3 red: both round-trips and the pyarrow oracle; the counts stay green	SAME 3
nullasempty (mine, the other direction: a NULL array read back as an empty one)	3 red, the same three	SAME 3
dropnullelem (mine: a NULL element dropped from its array)	3 red, the same three	SAME 3
structnotnull (mine: a NULL composite read back as a row of NULL fields)	3 red, the same three	SAME 3

Red-first on the shell (harness mutation, PG17): with both counts read from missing tables, main's shell passes 7/7. This shell reds "import count matches": not a measurement, so nothing was compared.

The three dated rows are exactly what readempty reddens.

Non-blocking, a coverage note rather than a defect: both files the suite imports hold one row group (5000 rows, and pyarrow's 200), so the importer's move from one row group to the next is never exercised. The export side (parquet_nested's np2) has a 70,000-row case. A multi-group import arm would cover the reader's half of that.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD Restated as 95cb6c63, stacked on #1629 (9109cc20). #1590 and #1594 merged after this PR was approved, so it conflicted. The content is unchanged. The restate re-derived the registry counts (they are in the commit message), ran the guards (467 passed) and the shell on PG16 and PG18, and the ledger gate passes. Please re-approve at this head. It merges after its predecessor.

🤖 Generated with Claude Code

[OffgridwithJD]

@jdatcmd Re-approving the stacked head 95cb6c63 (I approved 66089227).

• Content unchanged: compared file by file against 66089227, the PR's own +/- lines match. That covers every file except the budget, expected_tests and TESTS.md, which only restate counts; check_ledger.tsv and lossy_arms.tsv are identical.
• Gates: checked at the bottom of this stack (e142a0a6) and at 23f3d52d (#1631). Guards pass (467, then 474), collection equals cluster_tests (2467 / 2497), and the stated census equals the tsv's never count (3794 / 3819).

🤖 Generated with Claude Code

[jdatcmd]

Re-approval needed: one new commit, e70a166. CI refused the twin with REFUSED test_parquet_nested_import.py imports pyarrow outside require_module (line 63) (main's capability guard). The fixture no longer imports pyarrow: the pyarrow arm runs once, from the first pyarrow test, with the modules its require_module returned. Check names and count unchanged. Evidence is in the commit message: 7/7 on PG17 and PG19, 5 pass + 2 declined with pyarrow hidden, and a one-row-short reference file fails both pyarrow arms.

[OffgridwithJD]

@jdatcmd Re-approving e70a166a.

• Guard: _capability_import_sites over every test_*.py prints [] (checked at #1631's head, which carries this commit).
• Clean, PG17: 7/7 in shell and twin.
• Mutant readempty: SAME 3 reds, the same twin names as at my approval.
• pyarrow absent (a ModuleNotFoundError stub on PYTHONPATH): 5 pass + 0 fail + 0 unrun + 2 declined = 7, rc 0. Both pyarrow arms decline by name, and the fixture no longer imports it.

The pyarrow arm now runs once, from whichever pyarrow test arrives first, with that test's require_module modules. It's guarded by "pyarrow import" not in ni, so test order doesn't matter.

🤖 Generated with Claude Code