Introduce ConnectOptions to build a connection string from a builder

[mehcode]

As of 0.3

use sqlx::Connect;

let mut conn = PgConnection::connect("postgres://user:pass@localhost/database?sslmode=require");

Proposal

let options = ConnectOptions::new()
    .host("localhost")
    .port(5432)
    .username("user")
    .password("pass")
    .param("sslmode", "require");

let mut conn = PgConnection::connect(options).await?;
let mut conn = options.connect().await?;
let pool = PgPool::new(options).await?;
let pool = PgPool::builder().build(options).await?;

• Introduce ConnectOptions
• Implement TryFrom<&str> for ConnectOptions to parse URI and DSN connection strings
• Pool and Connection construction now take O: TryInto<ConnectOptions>
• The url crate should not be used for parsing; in particular, a schema should not be a required piece

---

Additional ideas:

• Introduce database-specific option extension traits to for parameters. E.g.. PgConnectOptionsExt would provide a ssl_mode method to set the sslmode parameter.

[mehcode]

Here are some notes.

https://gist.github.com/rust-play/8434ec3de09f776f4376972824c6190e

cc @g-s-k

[g-s-k]

One thought, though it may be off-base, is that associated types on a driver's Database impl are used pretty heavily here. Might that be a good pattern to use (rather than extension traits) for consistency?

So, for example...
https://gist.github.com/rust-play/1f9387bd0176da6e511c89dfd2afa2fb

[mehcode]

No, you're totally right. That is better. This would also keep it even more separate and allow for say SQLite to just use a PathBuf as storage for its filename. I didn't like how SqliteConnectOptions turned out in my gist.

---

I'm currently working on some refactoring efforts for postgres so I'll roll this into it. I wouldn't mind help on SQLite after I push up what I have.

[mehcode]

This is now been refactored to separate option types. We can accept additional database specific option knobs quite easily now.