Re: Tepid: selective index updates for heap relations

Hello,

Attached is v74: 21 patches against aa7b9cf13. Changes since posted v73:

In v74 I've folded the five repair patches into their introducing commits.
Dependent pageinspect and recovery tests move into amcheck patch18.
Statistics remain patch17. Patches 1-19 are the production series;
patches 20/21 are separate [DO NOT MERGE] test-AM and benchmark patches.

01. Name the locator contract between table and index access methods
02. Rename TM_FailureData.traversed to retargeted
03. Report table AM tuple failures without heap metadata
04. Let the executor say which version of a row it names
05. Let an index AM store a variable-width locator
06. Carry row copies in after-trigger events
07. Support table AMs that update rows in place
08. Let a table AM flag bitmap groups that do not combine exactly
09. Improve test coverage for heap-only tuple (HOT) update behavior
10. Let table AMs compare candidate attributes before an update
11. Implement the heap AM's modified-attrs comparison callback
12. Identify modified indexed attributes in the executor on UPDATE
13. Prepare index maintenance for selective updates.
14. Add the selective-indexed on-disk format: inline attr bitmap and stubs
15. Add a per-block locator-split bit to the visibility map
16. Implement heap HOT with selective index updates.
17. Extend heap statistics to include selective-indexed updates
18. Teach amcheck to verify HOT-indexed chains and index coverage
19. Add a subscription option for HOT-indexed updates
20. [DO NOT MERGE] Add test table AMs for the locator contract
21. [DO NOT MERGE] Add a HOT-indexed benchmark harness

best.

-greg