[OffgridwithJD]

@jdatcmd logical_decoding_source gains its pytest twin (#754), one check per test, every check by the shell's name. The twin runs on a private cluster at wal_level=logical, because the shared cluster runs at the default and the level needs a restart.

This PR will be restated into a stack with my other open ports (the iceberg four and logical_subscriber), so its counts move once at the end.

What was wrong, measured red-first on main's shell (PG17):

hole	main	this head
The second slot pgc_754b had no heap control. Created after the 5000-row insert, it decoded 0 records, and "and it too emits no decodable change of its own" and "100x the rows does not give 100x the internal records" (0 < anything) both passed	11/11	2 red, SAME in shell and twin: the new "CONTROL: the second slot decodes the heap table's 5000 inserts" and the ratio check, which now refuses a slot with no internal records
The pgc_skip message for a missing test_decoding.so was built from $(pg_config --pkglibdir), so the grader could not read it	grader: 1 missing	literal. With the plugin moved away on PG17: shell 1 FAIL, twin the same 1 FAIL and 12 declined

What no source mutant reddens: tinychunks (default chunk_group_row_limit 10000 → 100) and tinystripes (stripe_row_limit 150000 → 1000) both stay green. Internal records follow groups: 114 for 50 groups, 52 for 5 stripes, under the 10x bound. So the claim holds under them, and they date nothing. The suite's claim is that pgcolumnar writes nothing logical decoding can read. I found no one-line product change that makes a columnar write decodable, so all 13 rows arrive never.

Runs: 12/12 shell and 13/13 twin on each of 15–19. The twin's extra test is the pgc_skip mirror, which the shell records only when the plugin is missing.

Counts: census 3793 → 3806 never, 2073 dated, 5879 rows; suites_not_covered 113 → 112; cluster_tests 2454 → 2467; TESTS.md §256. Gates: guards 467 passed, collection 2467, ledger gate rc=0, self-test PASSED.

🤖 Generated with Claude Code

[jdatcmd]

Blocking: the ledger registers the suite's skip message as a check, and CI fails the ledger gate on it. Every PR stacked above this one inherits the failure (#1641, #1645–#1655: suites PG17 and PG18 red).

The diff adds this row to test/check_ledger.tsv:

logical_decoding_source  logical_decoding_source  test_decoding.so is not installed; build contrib/test_decoding  15;16;17;18;19  never  -

That text is the name of pgc_skip at test/logical_decoding_source.sh:39, not a check. CI has test_decoding, so the suite runs, never emits that name, and the orphan scan reports it (suites PG17 job, this head):

orphan: logical_decoding_source  logical_decoding_source  test_decoding.so is not installed; build contrib/test_decoding (claims 15;16;17;18;19, no history)
orphan scan: parts in the run=1, rows in those parts=13, orphans=1 (0 carrying history)

Main's ledger holds no skip names: grep -ciE "not available|not installed" on it prints 0.

To fix: drop the row, and reduce the census counts in check_ledger_budget.txt by one never-red row.

I have not reviewed the rest of this PR yet. I will once this is fixed.

[jdatcmd]

Approved at 9109eb2. My blocking point is fixed: the ledger no longer holds the skip message (grep -c "not installed; build contrib" on this head's ledger prints 0).

Reproduced on PG17, with the second slot created after the inserts (a dead slot) in a temporary copy of each harness:

run	result
main's shell, dead slot	11 passed + 0 failed: the hole
this head's shell, dead slot	2 failed: "CONTROL: the second slot decodes the heap table's 5000 inserts: got [0] want [5000]" and "100x the rows …: got [no (14 then 0; the second slot must carry some)]"
this head's twin, dead slot	the same 2 names, the same readings
clean, PG15 and PG19	shell 12/12, twin 13 passed

Non-blocking: when the plugin is present, the twin records a PASS named "test_decoding.so is not installed; build contrib/test_decoding". A reader of a twin log sees a passing check whose name says the opposite of what holds. Recording it only when the plugin is missing (as the shell's pgc_skip does) would match the shell exactly. It does not affect the ledger, which is why I'm not blocking on it.

This stack still conflicts with main (my comment on #1638). I'll re-approve a restate whose non-registry files are unchanged.