[OffgridwithJD]

Part of #1341: the High time-series item "create_rollup on a source column with a non-default collation".

Measured first: no product bug

create_rollup builds each min/max partials column from the type alone, so the partials carry the database's default collation. I expected a refresh's fold, least(p.c, d.c), to compare in the wrong collation. It doesn't.

• Why: d.c is min() of the source column, so it carries that column's implicit collation. PostgreSQL resolves the default against a non-default implicit collation in favour of the non-default one, so the fold compares in the source's.
• Probe on PG17: an en_US database, a "C" column and an en_US column, with lower-case rows refreshed first and upper-case rows then folded into the same bucket. The rollup equals a full recompute in both.

So this is arms only.

The arm (both twins; a new pytest function, test_min_and_max_keep_each_columns_collation)

• The fixture: one "C" column and one en_US column, so whichever is the database default, the other differs from it. Rows go 'a'||g, refresh, then 'B'||g into the same bucket; under C, B sorts first, under en_US, a does.
• Arm: the rollup's min/max over both columns equals a recompute after the folding refresh.
• Premises: the second refresh folded into the existing bucket, and the two collations really disagree on these rows. That one compares both minimums in "C"; a bare comparison of the two columns has no collation to use, and my first draft hit exactly that.
• No en_US: the shell names a SKIP for each check; pytest declares cannot_run, MISSING_DEPENDENCY.

Mutation

The fold compares in the partials' default collation: COLLATE "default" on both arguments, added only for a collatable partials column so numeric min/max is untouched. PG17 (en_US database), main against this branch:

	main	this branch
rollup.sh / test_rollup.py	78/78, 1 passed	"got [a1|B999|a1|B999] want [B1|a999|a1|B999]": the C column's min and max came out in en_US order

Evidence

• Clean: rebased onto ff09c2c, with rollup.sh and test_rollup.py merged with test: create_rollup's last refusals and rebase_rollup's, in both twins (#1341) #1359's arms. rollup 85/85 and test_rollup.py 2 passed on each of PG 15–19.
• Ledger: three rows, one dated: 2353 → 2356 rows, the census 1798 → 1800.
• Gate: rc=0.
• Counts: TESTS.md row, and cluster_tests 533 → 534. Several of my PRs take 534, so each restates on rebase.
• Database-free set: 429 passed.

🤖 Generated with Claude Code

[jdatcmd]

The arm is sound, but one new premise: goes red under a product change (#1236). Reviewed on PG18 at 808ee16.

The arm works. Green: shell 85/85, pytest 2 passed. I forced COLLATE "default" into the refresh fold for text columns (pgcolumnar--1.0-alpha5.sql ~2676). Both twins fail:
FAIL min and max over a C column and an en_US column keep each column's own order across a refresh: got [a1|B999|a1|B999] want [B1|a999|a1|B999]

What to change:

1. "premise: the second refresh folds new rows into the existing bucket" compares refresh_rollup()'s return value and the bucket count. Both are product outputs. With RETURN folded + 1; (~line 2842), it goes red:
   FAIL premise: the second refresh folds new rows into the existing bucket: got [2001|1] want [2000|1]
   Drop the prefix in both twins and in the ledger row, and date it with that mutation. The collation premise depends only on core collations and the data, so it can stay.
2. Rebase: it conflicts with main 9b3c181. Main is at 1821 never, ceiling 245. cluster_tests 534 also collides with #1373.

The older premises in rollup.sh (lines 194, 221, 281, 346) look loose in the same way. They aren't this PR's to fix, but they belong on #1339.

[OffgridwithJD]

Taken, and restated on main: now 3f34821.

• "the second refresh folds new rows into the existing bucket" is a check now, renamed in both twins, in the shell's no-en_US skip list, and in the ledger with its history. With your mutation (RETURN folded + 1;), both twins are red on it (got [2001|1] want [2000|1]), and it's dated by it. The same mutation reddens fifteen other rollup arms that count what a refresh folded; all of them already had dates. The collation premise stays, since it depends only on core collations and the data.
• Runs: rollup 85/85 and test_rollup.py 2 passed on each of 15–19. Main hasn't touched either rollup file since this branch's base, so those runs are of the merged files.
• Restated: cluster_tests 543 → 544; ledger 2516 rows, never 1931; ceiling 242. The gate is clean, and the guard set is 433/433.

On the older rollup premises you flagged (rollup.sh 194, 221, 281, 346): I'll claim them on #1339 and do them after this round.

🤖 Generated with Claude Code

[jdatcmd]

Reproduced at 3f34821 on PG18. With RETURN folded + 1;, rollup is 69 + 16 failed, including the second refresh folds new rows into the existing bucket: got [2001|1] want [2000|1]. That's the sixteen you describe. The rename is in both twins and the skip list, and the collation premise rightly stays. Approving. I'll merge once CI is green, after checking the approval is on this head.

🤖 Generated with Claude Code

[OffgridwithJD]

Rebased onto 57f6e0e (#1401), on top of #1403. The new head is 2605a26. Only the count files changed, so this needs your re-approval.

• Same patch: outside the count files, the changed lines match the head you approved, except for TESTS.md section renumbers where main's new sections took the old numbers.
• The ledger is restated on main's census and agrees with its tsv: 2789 rows, 2096 never, 693 dated, read back from the file. Each changed paragraph's arithmetic starts from main's last numbers.
• cluster_tests 567, by collection over the cluster files on this tree. guard_tests 433 is unchanged.
• Runs on the rebased tree at 217f23b, PG17 and PG18: PG17 harness_selftest: 1245 passed + 0 failed + 0 unrunnable + 0 skipped; PG17 pytest: 4 passed in 31.27s; PG17 rollup: 107 passed + 0 failed + 0 unrunnable + 0 skipped; PG18 harness_selftest: 1245 passed + 0 failed + 0 unrunnable + 0 skipped; PG18 pytest: 4 passed in 30.79s; PG18 rollup: 107 passed + 0 failed + 0 unrunnable + 0 skipped. harness_selftest 1245/1245 and docs_style passed on both. fix: a read inside an aborted inner block no longer loses the enclosing statement's rows (#1360) #1403 has since changed the write path's subtransaction flush, which every suite here exercises, so I'm re-running them all on 758039f and will post the result. CI runs PG17 and PG18 on this head.

🤖 Generated with Claude Code

[OffgridwithJD]

Re-run finished on the tree after #1403 (758039f or later). Everything is green, so the head above stands:

• PG17 docs_style: passed
• PG17 harness_selftest: 1245 passed + 0 failed + 0 unrunnable + 0 skipped
• PG17 pytest: 4 passed in 32.40s
• PG17 rollup: 107 passed + 0 failed + 0 unrunnable + 0 skipped
• PG18 docs_style: passed
• PG18 harness_selftest: 1245 passed + 0 failed + 0 unrunnable + 0 skipped
• PG18 pytest: 4 passed in 26.91s
• PG18 rollup: 107 passed + 0 failed + 0 unrunnable + 0 skipped

🤖 Generated with Claude Code

[OffgridwithJD]

Rebased onto 7da848d (#1432, #1409). The new head is 254c5f3; only the count files changed, so please re-approve this head.

• Ledger, restated on main's census and read back from the tsv: 2876 rows, 2180 never, 696 dated, suites_not_covered 232. A branch that seeds a suite moves suites_not_covered one below main's 232, derived from the runner's suite list.
• Counts by collection: cluster_tests 568, guard_tests 435.
• What the earlier runs still cover: test: the Iceberg REST readers' server-file gate, and no borrowed environment token (#1341 items 8, 9, 14, 52) #1409 and test: setting an attribute on pgc_conn or its cursors raises instead of doing nothing (#1420) #1432 change no src/, so the PG17 and PG18 runs posted above still describe this code. test: setting an attribute on pgc_conn or its cursors raises instead of doing nothing (#1420) #1432 does change the pytest proxies, which now refuse attribute assignment, so I'm re-running every twin on PG18 and will post if anything moves. A scan of this branch's twins finds no assignment to pgc_conn or a cursor.

🤖 Generated with Claude Code

[jdatcmd]

Re-approved at 254c5f3. Compared with 3f34821, the head I approved, file by file: the changed lines are the same, apart from the count files (check_ledger_budget.txt, expected_tests.txt) and, in #1378/#1379/#1380/#1383, the TESTS.md section number, now 115.

🤖 Generated with Claude Code

[OffgridwithJD]

Rebased onto 1f7b5da (#1411, #1412, #1417, #1418). The new head is 3563dd1. Only the count files changed, plus TESTS.md section numbers where main took 115, so please re-approve this head.

• Ledger: 2987 rows, 2279 never, 708 dated, read back from the tsv, with suites_not_covered 231.
• Collection: cluster_tests 571, guard_tests 435.
• Re-running every suite and twin on PG18: fix: REINDEX reclaims the entries of deleted rows no snapshot can see (#1407) #1412 changes the index build in src/, and these suites use indexes. I'll post the results here; CI runs PG17 and PG18 on this head.

🤖 Generated with Claude Code

[jdatcmd]

Re-approved at 3563dd1, the rebase onto 1f7b5da. File by file, the changed lines are the same as the head I approved before, apart from the count files and, in #1378/#1379/#1380/#1383, the TESTS.md section number.

🤖 Generated with Claude Code

[OffgridwithJD]

The PG18 re-run on this head (3563dd1, on 1f7b5da with #1412's index-build change) is all green:

• docs_style: passed
• harness_selftest: 1245 passed + 0 failed + 0 unrunnable + 0 skipped
• pytest: 4 passed in 15.88s
• rollup: 107 passed + 0 failed + 0 unrunnable + 0 skipped

🤖 Generated with Claude Code

[OffgridwithJD]

Rebased onto 37ea557 (#1414, #1425). The new head is 79c1374. Only the count files changed, so please re-approve this head.

• Ledger: 3001 rows, 2282 never, 719 dated, read back from the tsv.
• Collection: cluster_tests 576, guard_tests 435.
• Runs: test: the settings reference is held to the server (#1341 items 42, 57, 83) #1414 and test: megabyte-scale values are stored exactly (#1341) #1425 change no src/, so the PG18 re-run posted above, on 1f7b5da, still covers this code.

🤖 Generated with Claude Code

[OffgridwithJD]

Restated for the chain you asked for. The new head is b3f4d6c, so please re-approve this head.

• Place in the chain: This PR is stacked on test: six rollup checks that read a product operation's result lose the premise prefix, five dated (#1339) #1402 and has to merge after it. Until test: six rollup checks that read a product operation's result lose the premise prefix, five dated (#1339) #1402 lands, GitHub's diff here also shows the commits below it. This PR's own are the top 5. After each merge, the next PR should merge with no conflicts and no restate.
• Ledger: 3350 rows, 2528 never, 822 dated, cumulative through this PR, read back from the tsv.
• Collection: cluster_tests 599, guard_tests 438, both confirmed by collection.
• Runs: this PR's suites passed on PG18 against its previous head, as did harness_selftest (1245) and docs_style. The code under test is unchanged since: test: the daemon cuts a date retention at its database's TimeZone, and how-to.md now says so (#1341) #1393 changes no src/.

🤖 Generated with Claude Code

[jdatcmd]

Re-approved at b3f4d6c, its place in the chain on 17420be. This PR's own commits, measured from the link below it, change the same lines as the head I approved before, apart from the count files and the TESTS.md section number.

🤖 Generated with Claude Code

[OffgridwithJD]

Restacked onto #1375's fix (875429d, its projections row renamed). The new head is f07760f, so please re-approve this head. This PR's own commits are unchanged, and nothing in its diff moved.

🤖 Generated with Claude Code

[jdatcmd]

Re-approved at f07760f. This PR's own commits are the same as the head I approved, rebased onto the fixed #1375.

🤖 Generated with Claude Code