`TryFrom` for now-incompatible type falls back to `From` error message

[jonathanGB]

I am working on a library which at some point converts a mysql_common::value::Value to a type DataType via DataType::try_from(<value>). I am using an impl TryFrom<mysql_common::value::Value> for DataType defined in a separate crate, which worked well initially. However, this other crate has since then bumped up the dependencies, such that my project was now using a version of Value incompatible to the one used by the other crate.

The compiler complained (see error message) that "the trait std::convert::From<mysql_common::value::Value> is not implemented for DataType", and that the conversion was infallible. I think it would have been clearer and easier to debug had the compiler error message rather said that TryFrom was implemented for a type by the same name but incompatible (different version), rather than fall back on a blanket implementation error message which was deceptive in this case.

Error message

error[E0277]: the trait bound `noria::data::DataType: std::convert::From<mysql_common::value::Value>` is not satisfied
   --> src/lib.rs:283:19
    |
283 |             match DataType::try_from(value) {
    |                   ^^^^^^^^^^^^^^^^^^ the trait `std::convert::From<mysql_common::value::Value>` is not implemented for `noria::data::DataType`
    |
    = help: the following implementations were found:
              <noria::data::DataType as std::convert::From<&'a nom_sql::common::Literal>>
              <noria::data::DataType as std::convert::From<&'a noria::data::DataType>>
              <noria::data::DataType as std::convert::From<&'a str>>
              <noria::data::DataType as std::convert::From<chrono::naive::datetime::NaiveDateTime>>
            and 11 others
    = note: required because of the requirements on the impl of `std::convert::Into<noria::data::DataType>` for `mysql_common::value::Value`
    = note: required because of the requirements on the impl of `std::convert::TryFrom<mysql_common::value::Value>` for `noria::data::DataType`

error[E0308]: mismatched types
   --> src/lib.rs:285:77
    |
285 |                 Err(e) => return Err(WriteProxyErr::MysqlToNoriaDatatypeErr(e)),
    |                                                                             ^ expected `&str`, found enum `std::convert::Infallible`

Meta

rustc 1.44.0-nightly (94d3463 2020-04-09)
binary: rustc
commit-hash: 94d3463
commit-date: 2020-04-09
host: x86_64-unknown-linux-gnu
release: 1.44.0-nightly
LLVM version: 9.0

[jonhoo]

To provide a little more context: the issue here is that this implementation exists:

impl TryFrom<a::SomeType> for b::MyType { /* .. */ }

for, say, a = "1.0". And then some crate that depends on b tries to write:

b::MyType::try_from(a::SomeType)

but it depends on, say, a = "2.0". Rust therefore concludes (correctly) that the impl in b does not apply, since a[1.0]::SomeType != a[2.0]::SomeType. It then instead falls back on the blanket implementation from std:

impl<T, U> TryFrom<U> for T where
    U: Into<T>, 

And gives the very unintuitive error

error[E0277]: the trait bound `b::MyType: std::convert::From<a::SomeType>` is not satisfied

It would be much more helpful if the compiler here said

The trait bound `b::MyType: TryFrom<a::SomeType>` is not implement for this version of `a`.

This is essentially a particular instance of #22750.

[Kixunil]

I'd argue that the From message being misleading is a bigger issue here than it being from another crate. (Or maybe there are two issues.)

Yes, I know it happens because in some cases this kind of reasoning is useful. And it is possible to understand it by reading the error message more carefully - the note: at the end.

I wonder if rearranging the message would be sufficient, or some other optimization is needed (or even possible).

[Enselic]

Triage: I think this might have been fixed by #146874. Can you see if that PR helped this use case? Thanks!