[OffgridwithJD]

Part of #1341: create_rollup's remaining refusals and rebase_rollup's, from the medium and low items. Every arm is in both twins, next to the existing refusal arms. Each is red when the check it names is deleted from pgcolumnar--1.0-alpha5.sql, and main survives each deletion.

arm	mutation	main	this branch
a ts_column that is not timestamptz is refused	the timestamptz check removed	sh 72/72, py passed	"got [ok] want [22023]", and "no refused rollup left a table behind: got [1]"
a group_by column the source does not have is refused	the group_by column check removed	green	"got [42601] want [22023]": the CREATE TABLE breaks instead
a sum over a column that cannot be folded is refused	the cannot-be-folded check removed	green	"got [42601] want [22023]"
rebase_rollup of a table that is not a rollup is refused (42704)	its IF NOT FOUND removed	green	"got [42601] want [42704]"

The timestamp case is the one that matters most. A timestamp (no zone) source column would be bucketed in whatever zone the refresh happens to run in, so the refusal is the documented behaviour. With the check gone, the create succeeds, which is why the existing "no table left behind" arm also gains this mutation in the ledger.

Evidence

• Clean: rollup 76/76 and test_rollup.py passed on each of PG 15–19.
• Ledger: four new dated rows; 2336 → 2340 rows, census unchanged at 1803, restated at 8af86e8.
• Gate: rc=0.
• Database-free set: 429 passed.

Overlap with #1357: both touch rollup.sh and test_rollup.py in different places, and both move the budget, so whichever lands second restates.

🤖 Generated with Claude Code

[jdatcmd]

Approving the content at 8ed2b65. On PG17 I reproduced the timestamptz claim with create_rollup's check disabled:

• main: rollup passes 72/72.
• this head: "a ts_column that is not timestamptz is refused: got [ok] want [22023]" and "and no refused rollup left a table behind: got [1] want [0]". That second arm gaining this mutation is right, as you say.

Before merge: CI hadn't reported on this head when I looked, and main has moved to 1933577 (2343 / 1796). With your four dated rows I expect 2347 / 1796. It also overlaps #1357 in rollup.sh and test_rollup.py. I suggest #1357 lands first, then this one rebases onto it.

[OffgridwithJD]

@jdatcmd, rebased onto 8c61dff (#1357 merged). The new head is 7af80c9, replacing 8ed2b65. Ledger: 2353 / 1798 (2349 + 4 dated).

• Conflict: only check_ledger_budget.txt. The tsv equals main plus my delta, with no duplicate keys, and the changed test lines match 8ed2b65 exactly.
• Merged test files: rollup.sh and test_rollup.py auto-merged with test: the rollup catalogs' write side and the definer helpers refuse another role, in both twins (#1341) #1357's arms, so I ran the merged files on all five majors: rollup 82/82 and test_rollup.py passed on each of PG 15–19. The gate over those logs gives rc=0 with the census agreeing.
• Database-free set: 429 passed.

I'll ping on green CI.

[OffgridwithJD]

@jdatcmd, CI is green on 7af80c9 (15/15). With my run of the merged rollup files on 15–19 (82/82, posted above), it's ready for your merge. #1361 is next, and I'll rebase it as soon as this lands.

[jdatcmd]

Re-approving 7af80c9.

• Ledger: a recount gives 2353 / 1798, which agrees with the budget. No duplicate keys, no dated premise: row.
• Patch: the rollup changes are identical to 8ed2b65, the head I reviewed and reproduced.
• Database-free set: 429 passed.

Merging.