diesel: Examine our compiler errors, and see if we can improve them

So I discussed this with some folks at the all hands this year, and rustc_on_unimplemented has gotten a lot more functionality since the last time we looked at it. It still won’t solve all our problematic cases, but I think it may be enough to help us in a few of them. I’m going to try to look into this further to give the compiler folks more feedback on where it could help, and where we still need better hooks.

This is something I’d like to make a major focus in 2019, hopefully looking into using a lint to help cases that we can’t fix with other compiler hooks. A lot has changed since the last activity on this issue, so I’d like to renew the call for “please give us your bad error messages”. If you can provide a small, single file script to reproduce the problem (e.g. something we’d be able to add as a compile test to Diesel), that would be much appreciated.

@pickfire

Even with adding #[table_name = “users”]

error[E0433]: failed to resolve: use of undeclared crate or module `users` 
--> src/models.rs:8:16
  |
8 | #[table_name = "users"]
  |                ^^^^^^^ use of undeclared crate or module `users`

But this could be fixed with use super::schema::users;, this one is faster for me since I accidentally saw this on the guide when checking but I wish the errors could be as good as the rust compiler errors.

Generating a better error message for this case is unfortunately not possible as rust does not provide any API for proc macros to check which items are in scope and which are not in scope.

I was getting a weird error when missing table_name, I thought it will show some good errors but somehow I forgot to insert it. Then when I run it, I got confused with the error message, at the end it took me some time to realize it’s related to missing table_name.

error[E0433]: failed to resolve: use of undeclared crate or module `new_users`
 --> src/models.rs:8:12
 |
8 | pub struct NewUser {
 |            ^^^^^^^ use of undeclared crate or module `new_users`

If there is no #[table_name = "…"] attribute all of our derives default to using the plural (just append a s) snake case name of the current struct as table_name. As removing this behavior is a rather large change I’m not able to see how to improve that behavior here.

In the meantime I got another weird error.

error[E0308]: mismatched types
 --> src/bin/init.rs:8:9
  |
8  |       let users = vec![
  |  _________^^^^^___-
  | |         |
  | |         expected struct `Vec`, found struct `hello::schema::users::table`
9  | |         NewUser { name: String::from("John") },
10 | |         NewUser { name: String::from("Jane") },
11 | |     ];
  | |_____- this expression has type `Vec<hello::models::NewUser>`
  |
  = note: expected struct `Vec<hello::models::NewUser>`
             found struct `hello::schema::users::table`

This was because I mixed users variable and users::table below, weird namespace crash, I thought it should be type namespace and won’t happen.

You likely had an use schema::users::dsl::* somewhere which reexports schema::users::table as schema::users::dsl::users. schema::users::table is a zero sized struct, which means it’s both a value and a type, so something with that name exists in both namespaces. Again this is nothing we can fix on diesels side, but I think the rustc error message could be improved there giving a hint that there may be a name collision there. Could you open a issue in the rustc repo about that?