Infer multi-file binaries like src/bin/server/main.rs by convention

[matklad]

Currently cargo infers binaries from files in src/bin. This works if you binary fits in one file, but if you need more than one file, you have to write custom [[bin]] specification in Cargo.toml.

Looks like a reasonable convention for multifile binary projects is this:

src/
  bin/
    server/
        main.rs
        aux.rs
    client/
        main.rs
        foo.rs

So let's infer src/bin/foo/main.rs as a binary named foo?

This should be mostly backwards compatible change. The only failure case is when someone has src/bin/foo/main.rs which is not a binary, and no [[bin]] sections are present, and this seems unlikely.

A potential edge case is

src/
  bin/
    foo.rs
    foo/
      main.rs

This should be an error.

[alexcrichton]

Sounds great to me!

[matklad]

If anyone wants to fix it, the place to look at is this line

cargo/src/cargo/util/toml.rs

Line 54 in 263548d

try_add_files(&mut bins, root_path.join("src").join("bin"));

and this function

cargo/src/cargo/util/toml.rs

Line 503 in 263548d

fn inferred_bin_targets(name: &str, layout: &Layout) -> Vec<TomlTarget> {

.

The tests should probably go into this file: https://github.com/rust-lang/cargo/blob/master/tests/build.rs

[msehnout]

Is anybody working on this?

[alexcrichton]

@msehnout not that I know of, feel free to take it!

[crumblingstatue]

#4214 enabled this for src/bin, but how about also enabling this pattern for examples (and maybe integration tests)?

[matklad]

@crumblingstatue that sounds plausible! Right now I am refactoring the implementation of this stuff in #4259 though, so, if you'd like to send a PR, it's better to wait for a while :)

[rwakulszowa]

Hi @matklad, I think I could try to work on this one. The PR you mentioned is merged already. Is there anything I should be aware of before I start?

[matklad]

Awesome @rwakulszowa ! I think toml/targets.rs is the only file that should me modified, besides the test.

[rwakulszowa]

Hey, I have a few questions

• the discussion here mentions tests and examples. What about banchmarks?
• is there a nice way to test for existence of test binaries? Examples end up in the examples folder, but tests only produce artifacts in the main folder, decorated with hashes; is there a way to pass a regex / prefix string to the assert_that(..., existing_file())?

[matklad]

the discussion here mentions tests and examples. What about banchmarks?

Yeah, the benchmarks could use the same logic

is there a nice way to test for existence of test binaries?

I would probably check that cargo test --test foo works.

[sunjay]

Did support for examples with a main.rs file get added as part of this or should that be a separate issue?

[rwakulszowa]

Yes - examples, benches and tests follow the same logic and allow multi-file directories, as long as they contain a main.rs file. Did you encounter a problem with that? 19.11.2017 17:01 "Sunjay Varma" <notifications@github.com> napisał(a):

…

Did support for examples with a main.rs file get added as part of this or should that be a separate issue? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4086 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKQgyOsIslY-urII1LDbBZqRuXkU31cdks5s4F7zgaJpZM4Ni3Ta> .

[sunjay]

@rwakulszowa Yes, it didn't work. Here's a minimal example:

From the base project created by cargo new --lib <project name>, I added a file examples/dir/main.rs and a file examples/foo.rs.

.
├── Cargo.lock
├── Cargo.toml
├── examples
│   ├── dir
│   │   └── main.rs
│   └── foo.rs
└── src
    └── lib.rs

Both files contain the simple hello world program. When I try to run them, this is what happens:

$ cargo run --example dir
error: no example target named `dir`

Did you mean `foo`?

$ cargo run --example foo
    Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs
     Running `target/debug/examples/foo`
Hello, world!

The flat file worked, but nesting a file within a directory and having a file called "main.rs" did not work.

[rwakulszowa]

Are you sure you are using a recent cargo build? This change got merged in September, but I'm not sure which cargo release contains it ( @matklad ?) I have limited network access at the moment, so I'll check the official release tomorrow, but I tested the feature locally and it works fine. Maybe the feature didn't make it to the official release yet, or some other changes modified the behavior. I'll take a look. 19.11.2017 19:42 "Sunjay Varma" <notifications@github.com> napisał(a):

…

@rwakulszowa <https://github.com/rwakulszowa> Yes, it didn't work. Here's a minimal example: From the base project created by cargo new --lib <project name>, I added a file examples/dir/main.rs and a file examples/foo.rs. . ├── Cargo.lock ├── Cargo.toml ├── examples │ ├── dir │ │ └── main.rs │ └── foo.rs └── src └── lib.rs Both files contain the simple hello world program. When I try to run them, this is what happens: $ cargo run --example dir error: no example target named `dir` Did you mean `foo`? $ cargo run --example foo Finished dev [unoptimized + debuginfo] target(s) in 0.0 secs Running `target/debug/examples/foo` Hello, world! The flat file worked, but nesting a file within a directory and having a file called "main.rs" did not work. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4086 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKQgyEWAvRZWzF4_6nFTQlDQ5X2CNDGPks5s4ISxgaJpZM4Ni3Ta> .

[sunjay]

Maybe the feature didn't make it to the official release yet, or some other
changes modified the behavior. I'll take a look.

Yes it seems so. I tried running it with cargo +nightly run --example dir and it worked. Looks like this will probably be out in Rust 1.22 😄

[rwakulszowa]

Alright! 19.11.2017 20:45 "Sunjay Varma" <notifications@github.com> napisał(a):

…

Maybe the feature didn't make it to the official release yet, or some other changes modified the behavior. I'll take a look. Yes it seems so. I tried running it with cargo +nightly run --example dir and it worked. Looks like this will probably be out in Rust 1.22 😄 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#4086 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKQgyGFszm-gsWsYgHdRo9wYHW2a2AVNks5s4JNggaJpZM4Ni3Ta> .