Refactor QueryBuilder usage and implement *char join pushdown in PostgreSQL

[wendigo]

This is a preparatory step to be able to provide correct behavior when pushing down joins for case-(in)sensitive columns in the join condition comparisons

[grantatspothero]

Seems straightforward, as @hashhar said having an example implementation might help in visualizing why the change is needed.

[wendigo]

Supersedes #10435

[wendigo]

I've implemented *char join pushdown for PostgreSQL

[hashhar]

Nice. This would also allow us to accomodate other collation-sensitive pushdowns like predicates as well in a single place.

[wendigo]

@hashhar @ebyhr PTAL

[hashhar]

Maybe squash " Pass join relation aliases as a variable " and " Extract join condition formatting to separate method " since the " Pass join relation aliases as a variable " doesn't seem to do anything by itself.

[hashhar]

Squash " Move public methods to the top of the class " with " Open DefaultQueryBuilder methods for overloading "?

[hashhar]

Why is " Pass ConnectorSession to isSupportedJoinCondition " needed? It's unclear from the commit history where/if it's being used?

[hashhar]

The history is too fine-grained IMO. Makes following it a bit difficult.

The changes overall make sense.
I had a question - would just opening up the formatJoinCondition be sufficient for our case? The ability to provide different QueryBuilder seems like a hammer that can be easily misused/make code difficult to follow.

[wendigo]

Why is " Pass ConnectorSession to isSupportedJoinCondition " needed? It's unclear from the commit history where/if it's being used?

https://github.com/trinodb/trino/pull/10059/files#diff-f0eab13f31619586b4c64f21d633a031ef482c6cbf37d628d6cad76bc6178cf2R802

[hashhar]

Maybe BaseJdbcClient should provide a method to fetch the collation name that matches Trino semantics?

It'll make this method more re-usable across databases since not every database has a C collation.

Also same for the collation syntax.

Or maybe this should be left for future when we notice we are adding CollationAwareQueryBuilder2?

[hashhar]

Why is " Pass ConnectorSession to isSupportedJoinCondition " needed? It's unclear from the commit history where/if it's being used?

https://github.com/trinodb/trino/pull/10059/files#diff-f0eab13f31619586b4c64f21d633a031ef482c6cbf37d628d6cad76bc6178cf2R802

Ah, so it's needed for "Move collation-aware pushdown to query builder". Maybe it should be squashed together. 🙂

[wendigo]

@hashhar that's correct. That's why I've made it extensible

[hashhar]

LGTM % some history re-ordering (or some squashes)

@ebyhr PTAL around the QueryBuilder extension.

[ebyhr]

It would be nice to keep past commit history of DefaultQueryBuilder (ex-QueryBuilder). I will defer the final merge to @hashhar.

[hashhar]

@ebyhr Good point.

@wendigo Can you add a commit before "Pass QueryBuilder instance through the JdbcClient constructor" which just creates an unused exact copy of QueryBuilder as DefaultQueryBuilder so that the new file (DefaultQueryBuilder) has it's history preserved (since going forward it will be the new QueryBuilder impl.).

[wendigo]

PTAL @hashhar @ebyhr

[hashhar]

Just some checkstyle fixes in "Pass QueryBuilder instance through the JdbcClient constructor"

It looks good to me otherwise.

[hashhar]

CI hit #4948 and #10631.

cc: @grantatspothero Some Kudu flake?