add SQLx driver

[rust-db]

add SQLx driver

[jxs]

hi, and thanks for looking into this! 👍

The current API for SQLx does a ping-pong between connection and transaction. When you begin a transaction, it takes ownership of the connection object. This makes writing an impl AsyncTransaction for sqlx::PgConnection very hard... it may only be possible with the current APIs/Traits to implement for a PgPool... then the migration system can grab a connection from the pool, and run through the whole process.

that is ok i think, mysql_async support is only implemented for Pool at the moment too.

regarding the other two issues, i'd rather wait for sqlx to update to 0.23, and the compiler bugs to be fixed, what do you think?
Meanwhile i am going to provide migrations from a config file instead of connection, so that we support usecases where the driver is not yet supported like this one

[jxs]

hi @JamesHinshelwood, have the issues mentioned by @belak on #57 (comment) been addressed?

[thomaseizinger]

libsqlite-sys has a version mismatch between rusqlite and sqlx. This can be resolved by migrating back to rusqlite 0.22 until sqlx has upgraded.

This is already resolved in latest SQLx master, however, there is a big release coming up which seems to change things fundamentally so it might be good to wait for 0.6 to be released.

[belak]

At least for now, this is harder than you'd expect for multiple reasons:

• The current API for SQLx does a ping-pong between connection and transaction. When you begin a transaction, it takes ownership of the connection object. This makes writing an impl AsyncTransaction for sqlx::PgConnection very hard... it may only be possible with the current APIs/Traits to implement for a PgPool... then the migration system can grab a connection from the pool, and run through the whole process.
• There are compiler bugs around <Connection as HasRow>::Row vs PgRow - it has issues inferring that they're the same thing. This can be worked around by using a macro with the concrete types to implement the same functionality everywhere. That may actually be better in the long term. The developer is working on some tweaks which may improve this soon.
• libsqlite-sys has a version mismatch between rusqlite and sqlx. This can be resolved by migrating back to rusqlite 0.22 until sqlx has upgraded.

I may continue this later with Pool rather than the direct Connections if there's still interest.

[belak]

I don't know how soon the compiler bugs will be fixed, but I think waiting for a new SQLx version to work around them would be good enough.

One other question: how would the API for this work? I've got the basic driver done but I'm not as sure about the other necessary changes. Since there would be multiple types which could provide migrations for the different database types it complicates things a bit.

[jxs]

I don't know how soon the compiler bugs will be fixed, but I think waiting for a new SQLx version to work around them would be good enough.

ok awesome 👍

One other question: how would the API for this work? I've got the basic driver done but I'm not as sure about the other necessary changes. Since there would be multiple types which could provide migrations for the different database types it complicates things a bit.

I am not sure, but what you are saying is that, through SQLx traits we can't have a unified API for all of their supported drivers?

[JamesHinshelwood]

Has anything changed on the status of this issue since these previous comments?

I'm happy to take a look at implementing something if there aren't any blockers..?

[jxs]

also, since sqlx now also offers migration features, this might be redundant no?