Horton
Menu

Re: Index AM API cleanup

Peter Eisentraut <peter@eisentraut.org>

View thread
From: Peter Eisentraut <peter@eisentraut.org>
To: Mark Dilger <mark.dilger@enterprisedb.com>
Cc: PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2025-03-20T11:59:29Z
Lists: pgsql-hackers

Attachments

Here is a new version of the remaining main patch set.  I've made a lot 
of changes to reduce the size and scope.

- In version v21, you had included a bunch of expected files in the 
"treeb" module, which wasn't necessary, since the test driver overwrote 
those anyway.  I removed those, which makes the patch much smaller.

- The patch set made various attempts to be able to use a non-btree 
operator family as the referenced operator family for ordering 
operators, but this was not fully developed, since you had just 
hardcoded BTREE_AM_OID in most places.  I have cut all of this out, 
since this just blows up the patch surface but doesn't add any 
functionality at this time.  I believe the patch for ALTER OPERATOR 
FAMILY / ADD falls into this category as well, so I have moved this to 
the end of the patch series, as "WIP".

(I think this is ok because the purpose of this patch set is to allow 
new index types that behave like btree in many ways, whereas what the 
above feature would allow is to have a type that doesn't require a btree 
operator family to communicate its semantics.  These things are morally 
related but technically distinct.)

- For similar reasons, I removed the changes you made in typcache.c. 
With the new infrastructure we have in place, we might someday 
reconsider how the type cache works and make it more independent of 
hardcoded btree and hash behavior, but this is not necessary for this 
patch set.

- Move the changes in rangetypes.c to a separate "WIP" patch.  This 
patch is incomplete, as described in the patch, and also not needed for 
this patch series' goal.

- There were also a couple of other minor "excessive" generalizations 
that I skipped.  For example, we don't need to make the function 
btcostestimate() independent of btree strategies.

- Conversely, I moved the changes in network.c to a separate patch at 
the front of the patch series, and fixed them up so that they actually 
work.  The effect of this is easily visible in the "inet" test in the 
"xtree" test module.

- I separated out the changes dealing with row compare support into a 
separate patch.  I think this is not fully ready.  You'll see the FIXME 
in src/backend/executor/nodeIndexscan.c.  Also, this still had a test 
for BTREE_AM_OID in it, so it didn't really work anyway.  I think 
leaving this out for now is not a huge loss.  Index support for row 
compare is a specialized feature.

- I squashed together the remaining feature patches into one patch.  I 
observed that you had gradually adjusted the interfaces to use 
CompareType instead of strategy numbers, but then left it so that they 
would accept either one (for example, PathKey).  But then I went through 
and removed the strategy number support for the most part, since nothing 
was using it anymore.  (For example, PathKey was using strategy numbers 
for fake purposes anyway, so we don't need them there anymore, and it 
all trickles from there in a way.)  The resulting patch is much smaller 
and much much simpler now, since it is mostly just a straight conversion 
from strategy to compare type and not much else.

Here is the current lineup:

v22-0001-TEST-Add-loadable-modules-as-tests-of-the-amapi.patch
v22-0002-TEST-Fixup-for-test-driver.patch

This is the test setup, as before.  I added a small fix to reduce the 
diffs further.

v22-0003-TEST-Add-treeb-deep-copy-of-btree.patch.gz
v22-0004-TEST-treeb-fixups.patch

As before, with a compilation fix to keep up with some other changes.

v22-0005-Generalize-index-support-in-network-support-func.patch

As explained above, I think this patch stands on its own and is ready to 
go.  Check please.

v22-0006-Convert-from-StrategyNumber-to-CompareType.patch

This is all that remains now.  I think with a bit more polishing around 
the edges, some comment updates, etc., this is close to ready.

v22-0007-TEST-Rotate-treeb-strategy-numbers.patch

This still works. ;-)

v22-0008-WIP-Support-row-compare-index-scans-with-non-btr.patch
v22-0009-WIP-Generalize-index-support-in-range-type-suppo.patch
v22-0010-WIP-Allow-more-index-AMs-in-ALTER-OP-FAMILY.ADD.patch

These are postponed, as described above.

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Relax ordering-related hardcoded btree requirements in planning

  2. Support non-btree indexes in get_actual_variable_range()

  3. Convert PathKey to use CompareType

  4. Generalize index support in network support function

  5. Update a code comment

  6. Allow non-btree unique indexes for matviews

  7. Allow non-btree unique indexes for partition keys

  8. Add some opfamily support functions to lsyscache.c

  9. Simplify and generalize PrepareSortSupportFromIndexRel()

  10. Generalize hash and ordering support in amapi

  11. Drop opcintype from index AM strategy translation API

  12. Allow non-btree speculative insertion indexes

  13. Support non-btree indexes for foreign keys

  14. Integrate GistTranslateCompareType() into IndexAmTranslateCompareType()

  15. Convert strategies to and from compare types

  16. Move CompareType to separate header file

  17. Add get_opfamily_name() function

  18. Rename GistTranslateStratnum() to GistTranslateCompareType()

  19. Change gist stratnum function to use CompareType

  20. Rename RowCompareType to CompareType

  21. Improve slightly misleading internal error message

  22. Track scan reversals in MergeJoin

  23. Track sort direction in SortGroupClause

  24. Add stratnum GiST support function

  25. Add amgettreeheight index AM API routine