"cargo metadata" no longer displays "example" kind if example specifies crate-type

[ehuss]

As of /pull/3556, the cargo metadata command does not output the "kind" as "example" if the example specifies a crate-type. Sample Cargo.toml:

[package]
name = "scratch"
version = "0.0.0"

[[example]]
name = "elib1"
path = "examples/elib1.rs"
crate-type=["staticlib"]

I am working on a tool that uses the metadata output to determine which command-line arguments to pass to Cargo in order to build a target. Without knowing this is an example, it is impossible to know to pass --example.

I am unsure of what the correct solution is. Some ideas:

• "kind" can display "example" even if it is an ExampleLib.
• Include an "is_example" field.
• Include an "args" field that indicates how to build this target.

[jethrogb]

Duplicate of #3572 Edit: wait. This seems to be the inverse of #3572?

[KalitaAlexey]

@alexcrichton,
I think kind should still show "example".
If it is right, I will change that. It is easy change.

Do we have tests for cargo-metadata?
None of tests failed.
I can write some tests.

[jethrogb]

I think metadata should show both type of output and the information required to determine the correct cargo cmd line.

[KalitaAlexey]

@jethrogb,
You can't execute cargo run --example if an example is a library so we might need something new, i.e., examplelib.

[jethrogb]

But you can execute build, rustc, or test with that flag.

[KalitaAlexey]

But I would rather prefer to add a field example which is either true or false.

[jethrogb]

I think that would be sufficient?

[KalitaAlexey]

@jethrogb,
Yes it would since it allows to determine whether it is an example or not and to determine what it produces.

[alexcrichton]

I posted on the PR as well but I think we should revert the breaking change here and just extend the metadata format to otherwise accommodate examples-as-libraries in metadata.

@KalitaAlexey would you be willing to prepare such a PR?

[KalitaAlexey]

@alexcrichton,
I will send a PR today (in 3 hours).

[KalitaAlexey]

@alexcrichton,
It is not a breaking change because we didn't accept crate-type for an example.
It seems that cargo metadata just reads crate-type.
I can specify crate-type = ["e"] and it prints it.
So I didn't break anything.
This is why I think we should add a field "is_example" of boolean.
"kind" is what specified in crate-type.
What do you think about it?

[jethrogb]

Sure it is a breaking change. Before it was kind: ["example"], now it's not.

[KalitaAlexey]

@jethrogb,
If you don't specify any crate-type it will be "kind": ["example"].

[alexcrichton]

@KalitaAlexey yeah this was a breaking change from before because existing code has broken. I think "kind" was probably a misnomer from before, so we should leave "kind" as it was and then add a separate "crate-types" field which has the information about crate types.

[KalitaAlexey]

@alexcrichton,
Okay.
I will find the solution.