sql: disallow DESC option in the last column of inverted indexes

[mgartner]

The last column of inverted indexes cannot have the DESC direction.
Prior to this commit it was allowed, but caused internal errors. To
avoid propagating the notion that inverted indexes have a semantic
direction, we stop printing the ASC option for inverted index columns
in the output of SHOW CREATE TABLE.

Fixes #84388

Release note (sql change): The last column of an INVERTED INDEX can no
longer have the DESC option. If DESC was used in prior versions, it
could cause internal errors.

[cockroach-teamcity]

This change is 

[rytaft]

Reviewed 16 of 16 files at r1, all commit messages.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @chengxiong-ruan, @mgartner, and @michae2)

---

-- commits line 16 at r1:
I don't think this is true. There are some cases where we scan a range. For example, look at the second span here:

> EXPLAIN (opt) SELECT a FROM json_tab WHERE b @> '{"a": {}}' ORDER BY a
;
                                 info
----------------------------------------------------------------------
  sort
   └── project
        └── inverted-filter
             ├── inverted expression: /5
             │    ├── tight: true, unique: false
             │    └── union spans
             │         ├── ["7a\x00\x019", "7a\x00\x019"]
             │         └── ["7a\x00\x02\x00\xff", "7a\x00\x03")
             └── scan json_tab@foo_inv
                  └── inverted constraint: /5/1
                       └── spans
                            ├── ["7a\x00\x019", "7a\x00\x019"]
                            └── ["7a\x00\x02\x00\xff", "7a\x00\x03")

Why can't we just disallow DESC for now?

[mgartner]

-- commits line 16 at r1:

Previously, rytaft (Rebecca Taft) wrote…

I don't think this is true. There are some cases where we scan a range. For example, look at the second span here:

> EXPLAIN (opt) SELECT a FROM json_tab WHERE b @> '{"a": {}}' ORDER BY a
;
                                 info
----------------------------------------------------------------------
  sort
   └── project
        └── inverted-filter
             ├── inverted expression: /5
             │    ├── tight: true, unique: false
             │    └── union spans
             │         ├── ["7a\x00\x019", "7a\x00\x019"]
             │         └── ["7a\x00\x02\x00\xff", "7a\x00\x03")
             └── scan json_tab@foo_inv
                  └── inverted constraint: /5/1
                       └── spans
                            ├── ["7a\x00\x019", "7a\x00\x019"]
                            └── ["7a\x00\x02\x00\xff", "7a\x00\x03")

Why can't we just disallow DESC for now?

Good point. I still think it's odd to give the impression that inverted index columns have a semantic ordering. The indexed column contains only one dimension of the value, and an inverted index could never support sort-free ordering on the inverted index column.

Postgres disallows ASC on GIN indexes, which is one of the reasons I removed it:

marcus=# create table t (a int[]);
CREATE TABLE

marcus=# create index on t using gin (a asc);
ERROR:  0A000: access method "gin" does not support ASC/DESC options
LOCATION:  ComputeIndexAttrs, indexcmds.c:2033

But allowing it would be more backward compatible...

@cockroachdb/sql-experience any thoughts?

[rafiss]

assorted thoughts:

• Are we able to use any form of product telemetry to determine how common using gin (a asc) expressions are?
• What if we showed a NOTICE saying that ASC is ignored (if it is specified), rather than an error?
• When it comes to backward compat concerns, let's involve the Product team. cc @vy-ton

[mgartner]

I've updated this PR to only disallow DESC and stop printing ASC in the output of SHOW CREATE TABLE for inverted index columns. This should not be considered a breaking change because creating an inverted index column with DESC would cause internal errors prior to this commit.

[rytaft]

Reviewed 8 of 8 files at r2, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @chengxiong-ruan and @michae2)

[mgartner]

I've confirmed that the bug was reproducible without DESC, so this regression test remains useful.

[mgartner]

TFTR!

bors r+

[craig]

Build failed (retrying...):

• GitHub CI (Cockroach)

[craig]

Build succeeded:

• GitHub CI (Cockroach)