feat: add sqllogictest

[kemingy]

• fix test: Migrate sql test in pgvecto.rs to this repo #25

[kemingy]

It's ready to review.

Since this lib depends on pgvector, I didn't include many basic logic tests here. Let me know if you want to add more @VoVAllen

cc @usamoi @cutecutecat

[VoVAllen]

https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/issue_427.slt
https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/null.slt
https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/partition.slt
https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/pushdown_plan.slt
https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/pushdown_range.slt
https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/reindex.slt

[VoVAllen]

Some of the function (partition index/partial index/reindex concurrently) needs test verification. I don't know whether we support it for now. Fine to fail the test, but need verification

[kemingy]

https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/issue_427.slt https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/null.slt https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/partition.slt https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/pushdown_plan.slt https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/pushdown_range.slt https://github.com/tensorchord/pgvecto.rs/blob/main/tests/sqllogictest/reindex.slt

Some observations:

• https://github.com/pgvector/pgvector?tab=readme-ov-file#vector-type doesn't support NaN or Infinity
• unable to create an index on the partitioned table directly, but can use ONLY and then create indexes on each partitions
• rabbithole only supports the f32 (pgvector type) index
• rabbithole doesn't have fn/op like sphere, <<->>, etc. failed when calling <<->>
• pgvector sparsevec index from 1
• pgvector bit use text format like '010101110'
• pgvector uses l2 distance instead of l2 squared distance

Some of the function (partition index/partial index/reindex concurrently) needs test verification. I don't know whether we support it for now. Fine to fail the test, but need verification

So null, pushdown_range, pushdown_plan failed.

[usamoi]

unable to create an index on the partitioned table directly, but can use ONLY and then create indexes on each partitions

I think this should work?

rabbithole doesn't have fn/op like sphere, <<->>, etc

It does have.

pgvector uses l2 distance instead of l2 squared distance

be fixed by #44

[kemingy]

unable to create an index on the partitioned table directly, but can use ONLY and then create indexes on each partitions

I think this should work?

There will be:

ERROR:  operator class "vector_dot_ops" does not exist for access method "rabbithole"

Though the search_path is correctly configured.

rabbithole doesn't have fn/op like sphere, <<->>, etc

It does have.

Yes, it has. But there will be an error when using the <<->>:

ERROR:  calling `type_oid` is never expected

[usamoi]

Though the search_path is correctly configured.

Are you testing with PostgreSQL 17? It's a bug of PostgreSQL 17.

https://www.postgresql.org/message-id/18637-f51e314546e3ba2a%40postgresql.org

[kemingy]

Though the search_path is correctly configured.

Are you testing with PostgreSQL 17? It's a bug of PostgreSQL 17.

https://www.postgresql.org/message-id/18637-f51e314546e3ba2a%40postgresql.org

Weird.

While CREATE INDEX is running, the search_path is temporarily changed to pg_catalog, pg_temp.

Is that true? I tried public.vector_dot_ops in pg17 but also failed.

[usamoi]

Is that true? I tried public.vector_dot_ops in pg17 but also failed.

I think you should try rabbithole.vector_dot_ops?

[kemingy]

Is that true? I tried public.vector_dot_ops in pg17 but also failed.

I think you should try rabbithole.vector_dot_ops?

Fixed.

Is there a bug in <<->> op?

[usamoi]

Is there a bug in <<->> op?

Yes. It's not expected behavior.

[VoVAllen]

Can you try also rabbithole.<<->>?

[kemingy]

Can you try also rabbithole.<<->>?

I guess you mean OPERATOR(rabbithole.<<->>). No, it doesn't work. This is totally different problems.

[kemingy]

@VoVAllen review?