Research plugin hook for alternative database backends

[simonw]

I started exploring what Datasette would like running against PostgreSQL in #670 and @dazzag24 did some work on Parquet described in #657.

I had initially thought this was WAY too much additional complexity, but I'm beginning to think that the Database class may be small enough that having it abstract away the details of running queries against alternative database backends could be feasible.

A bigger issue is SQL generation, but I realized that most of Datasette's SQL generation code exists just in the TableView class that runs the table page. If this was abstracted into some kind of SQL builder that could be then customized per-database it might be reasonable to get it working.

Very unlikely for this to make it into Datasette 1.0, but maybe this would be the defining feature of Datasette 2.0?

[simonw]

DuckDB has a read-only mechanism: https://duckdb.org/docs/api/python

import duckdb
con = duckdb.connect(database="/tmp/blah.db", read_only=True)