Spans from serde losing expansion info

[Manishearth]

https://github.com/Manishearth/scratch/tree/rustc_serde_spans

If you run cargo build in poc there, the span_lint shows:

   Compiling poc v0.1.0 (file:///home/manishearth/sand/bar/poc)
src/lib.rs:7:10: 7:19 error: this let-binding has unit value.
src/lib.rs:7 #[derive(Serialize)]
                      ^~~~~~~~~
src/lib.rs:4:9: 4:23 note: lint level defined here
src/lib.rs:4 #![deny(let_unit_value)]
                     ^~~~~~~~~~~~~~
src/lib.rs:7:10: 7:19 error: this let-binding has unit value.
src/lib.rs:7 #[derive(Serialize)]
                      ^~~~~~~~~
src/lib.rs:4:9: 4:23 note: lint level defined here
src/lib.rs:4 #![deny(let_unit_value)]
                     ^~~~~~~~~~~~~~
error: aborting due to 2 previous errors
Could not compile `poc`.

I'm running a nightly, and serde by default uses the plugin infrastructure, so I doubt it's syntex causing any issues here. (Also for that to happen I'd need a build script on my side)

What's unusual here is that the span_lints don't say anything about being part of a macro. They're actually catching a let binding from inside the macro (I verified this by running it on the expanded code) -- but the span says nothing about expansion and just points at derive(). I think the ExpnInfo has vanished somehow, but I'm not sure.

cc @eddyb @erickt

[oli-obk]

I don't think this is an issue anymore with custom derive stabilized?

[Mark-Simulacrum]

I feel like this is intentional, but not sure.

[estebank]

I'm closing this, as with proc-macro-derive library writers can specify spans to provide appropriate spans for errors.