feat: use try_into in sqlx::query_as! conversions

[vidhanio]

closes #2648

[abonander]

This is a breaking change.

[vidhanio]

should the blanket TryFrom where From impl not prevent this from being breaking?

[bm-w]

IMO this would be a great improvement — @abonander, could you perhaps expand a little on your previous comment? If this is indeed a breaking change, are you outright rejecting this PR, or do you see a path forward to getting this merged, perhaps in an upcoming release that would permit breaking changes?

[abonander]

If I'm honest, it's mostly a gut feeling that a sweeping change like this should not sneak into a minor release. Even if it seems like it should be backward-compatible, there may be some subtle detail that we're not seeing that comes back to bite us later.

We should also be considerate of forward compatibility. If this change goes out in a minor release and someone updates their code to rely on it, then has to downgrade for whatever reason (e.g. to work around a bug), they're going to have a huge headache.

In all, my gut is telling me this is something that should be reserved for 0.8.0.

[jplatte]

Separately from the question of whether it's a breaking change, is this really the right thing to do? It would allow querying a table with an INTEGER column into a struct that has a field of type u8. While I guess it is convenient that it allows you to use unsigned types with postgres without extensions if you don't exceed i64::MAX and such, I really worry about this masking bugs where the types are unintentionally different. IMHO try_into decoding should be opt-in per query, not the default, if supported at all.

[abonander]

That is a fair concern as well. It has a lot of potential to turn compile-time errors into runtime errors, which is exactly the opposite of what we want.

[vidhanio]

Thank you for your comments, I definitely understand the concerns and agree about avoiding turning compile time errors into runtime errors. I also agree this is too large of a change to be a small patch.

Would a try_query_*! set of macros make more sense to solve this issue?

[abonander]

I think it would be better to be able to opt into this on a per-field basis, and require the user to explicitly specify the type to decode to ensure type-safety.

Ultimately, we want query_as!() to be able to work with #[derive(FromRow)], and I think that would be the best place to do this, e.g. #[sqlx(try_from = "<type to decode from>")]. That feature is being discussed in #514. It theoretically doesn't have any blockers but as I explain in my latest comments there, the support from the language is not quite where I want it to be yet.

I have an alternative, more immediately actionable idea which I'll elaborate on in the motivating issue, #2648. I'm going to close this PR as I'm not prepared to merge it as written because of the problems we've discussed.