Postgres Support

[koudelka]

Postgres Support

[jeremyjh]

I guess your tests are using cockroachdb, is why they passed? Cockroach is only compatible at the protocol level, it is not source or feature compatible with postgres. For some reason in the case of extract function they chose not to even be compatible with ANSI SQL.

[koudelka]

just pushed proper postgres support in 8134472 on master, try it out and let me know how it goes. :)

[jeremyjh]

@koudelka It works, thanks! Glad to have this in the main repository already.

I have a couple of suggestions based on the path I was headed down.

The SQL in your Postgres module looks like pure ANSI, no special Postgres features. I'd call the module "ANSI". I'd also suggest changing the option you pass to honeydew_fields as dialect rather than database. Dialects now would be :cockroach and :ansi, in the future maybe someone will add :cql for Cassandra etc.

Second, you could have a single example project support multiple database adapters and run the test suite on both of them by using a separate MIX_ENV. I had started to make a separate MIX_ENV=pg_test, which changed the adapter - you could also use it in the migration to select the database/dialect depending on the value of Mix.env. I didn't get that far since you mentioned you were working on this, but the first step of the mix.config I did here: smartmetals@e0edef0

If you take both these suggestions adding support for Maria is completely trivial I think, it would just be another MIX_ENV for test and I don't think would require any different SQL code but I didn't try it yet.

[koudelka]

Can you see a reason to want to select which dialect is used, outside of needing to manually select it when the adapter is ambiguous?

Thanks for the MIX_ENV idea! I really didn't like the idea of duplicating the example project. :)

[jeremyjh]

@koudelka No, I think if the macros can get access to the applications' Repo configuration you could save the user the trouble, but I think also if a single ANSI dialect will cover Postgres and MySQL (as written it would work with SQL Server and Oracle too, I think) you can just default to that, which will cover 95% of users.

[koudelka]

combined the example into a single project, thanks again for the suggestion. :)