[OffgridwithJD]

@jdatcmd Stacked on #1517. Review only the top commit, 9bc605c.

What changes

A new check. native_roundtrip.sh's header says the read crosses several row groups, and nothing checked that. A new check, "the native table holds five row groups", counts them in pgcolumnar.row_group: stripe_row_limit 1000 over 5000 rows gives five. It isn't a premise:, because a writer that ignores the limit would redden it.

The twin is test/pytest/test_native_roundtrip.py. It covers all 9 checks by the shell's name, one check per test. The reads and the answers are parametrized by the shell's names (name,sql), so one red can't stop the next. The loop-reader list no longer needs the file.

Evidence: the shell against the twin as shipped (raising expect)

mutant (columnar_reader.c)	shell	twin	what goes red
none	9/9	9 passed	—
a null by-value column reads back as a present 0	2	2, the same	the float null count, all columns
a null by-reference column reads back as a present empty value (yours)	4	4, the same	all columns, the text null count, filtered scan, projection
the last row group is never loaded	6	6, the same	everything except the row count, the row-group count and the filtered scan
zone-map pruning inverted	1	1, the same	the filtered scan

All of these are PG18 runs, and yours gives the same 4 on PG17. The head run is 9/9 on each of PG15–19.

Bookkeeping

• Ledger: native_roundtrip seeded with 9 rows: 4424 → 4433, never 3069 → 3071, dated 1355 → 1362, suites_not_covered 188 → 187.
  • The 2 never are the row count, which the metadata path answers (TESTS.md now says so), and the row-group count.
• Gate: rc 0 against test: native_gap gains its pytest twin, and its indexed lookup reads the index (#1341) #1517's branch. The orphan scan finds 0.
• Collection: cluster_tests 763 → 772, nine tests.
• Docs: TESTS.md gets §178, and the stem leaves unported_suites.txt.
• Guards: 440/440.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD, requesting changes at 5be44b28 for one thing: the twin still hides part of the red set, which the body says the split prevents.

My mutant: a by-reference null reads back as a present empty value

In columnar_reader.c's per-value decode, the null branch returns an empty text with nulls[c] = false when !att->attbyval. That's a different class from your by-value mutant: only v text is affected. On PG17:

	red
main's native_roundtrip.sh	4: all columns, null handling (v), filtered scan, projection
this PR's shell	the same 4 (9 checks, 5 pass)
this PR's twin	2: all columns, null handling (v). 2 pass, 2 fail

test_a_native_table_reads_back_what_the_heap_holds runs "all columns", "filtered scan" and "projection" in one loop. The first red raises, so the other two never run. The body says the two-test split is there "so one red doesn't hide the rest of a red set", and under this mutant it hides half of it. Your three mutants happened not to show it: each one's red names fell one per test, or into the row-count test.

Suggested fix: parametrize the two loop tables, so each name is its own test. cluster_tests then goes up by the number of names rather than by 2, and the loop-reader entry can go, because the names become parametrize ids. Or anything else that runs each check whatever the one before it did.

Non-blocking

• premise: on the row-group count. A writer that ignores stripe_row_limit reddens it, so by #1236 it's a property rather than a premise. Same as the plan premises in #1513 and #1517.
• "native row count" is answered from metadata, as your body notes, so it can't see a lost row group. Worth a word in the TESTS.md row.

[OffgridwithJD]

@jdatcmd agreed on all of it, and thank you for both mutants. I took (a), one check per test, in all three PRs:

PR	head	tests	how
#1517	03c31de	9	the module fixture runs every step in the shell's order and records each reading; each test asserts one
#1518	9bc605c	9	the reads and answers are parametrized by the shell's names (name,sql); the loop-reader entry is gone
#1519	e53ba5d	7	one function per check

The red-set tables now come from the twin as shipped, not from my local soft runner. You were right that the soft runner showed only that a name could go red. My runner now runs the real twin file and reads the AssertionError: <name>: got lines. Every mutant matches the shell, yours included:

• the by-reference null: 4 and 4 on PG18 and PG17;
• the inverted bloom setting: 2 and 2.

Also done:

• test: native_roundtrip gains its pytest twin, and asserts the read crosses its row groups (#1341) #1518: the row-group check drops premise:. It reads "the native table holds five row groups", and its ledger row is reseeded from new five-major logs.
• test: native_roundtrip gains its pytest twin, and asserts the read crosses its row groups (#1341) #1518: the TESTS.md text says the row count comes from metadata.
• The stack is rebased onto bbda341. The counts are unchanged, apart from cluster_tests (754 → 763 → 772 → 779).
• At each tip: gate rc 0, guards 440, and the collection count matches.

On (b), a deferred-failure mode for expect: I agree it's the owner's call. One-check-per-test fixes these three without touching the vacuity layer. I'll open an issue for it if the pattern keeps recurring in new twins.

🤖 Generated with Claude Code

[jdatcmd]

Approving at 9bc605ce. I took (a) in this PR and reran my mutant on the twin as shipped, comparing exact red-name sets from E AssertionError: <name>: got lines against the shell's FAIL lines. PG17:

mutant	shell red	twin red	sets
a by-reference null reads back present and empty (`columnar_reader.c`)	4	4	identical: all columns, null handling (v), filtered scan, projection

Counts. One cluster_tests line, 772, and collection at this head is 772.

At the stack's tip (e53ba5d6), test_harness_deps, test_compare_to_bash and test_mutation_ledger pass 105.

I'll merge in stack order as CI goes green: #1517, then #1518, then #1519.

[jdatcmd]

@OffgridwithJD merging #1518 now at 9bc605ce: approved on this head, 15/15 checks, CLEAN.