cargo init's boilerplate doesn't work with ?

[jsha]

cargo init currently produces boilerplate that looks like this:

fn main() {
    println!("Hello, world!");
}

However, if someone starts from this boilerplate and pastes in examples from documentation that use ?, they'll get an error. main doesn't return Result, so ? can't be used. The API guidelines recommend that examples use ? rather than unwrap.

As an example, if someone tries to paste in this small File example:

use std::fs::File;
use std::io::prelude::*;

fn main() {
    let mut file = File::create("foo.txt")?;
    file.write_all(b"Hello, world!")?;
}

They get some rather confusing errors (what's FromResidual?):

error[E0277]: the `?` operator can only be used in a function that returns `Result` or `Option` (or another type that implements `FromResidual`)
   --> src/main.rs:5:43
    |
4   | / fn main() {
5   | |     let mut file = File::create("foo.txt")?;
    | |                                           ^ cannot use the `?` operator in a function that returns `()`
6   | |     file.write_all(b"Hello, world!")?;
7   | | }
    | |_- this function should return `Result` or `Option` to accept `?`
    |
    = help: the trait `FromResidual<Result<Infallible, std::io::Error>>` is not implemented for `()`
note: required by `from_residual`
   --> /home/jsha/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/try_trait.rs:339:5
    |
339 |     fn from_residual(residual: R) -> Self;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

error[E0277]: the `?` operator can only be used in a function that returns `Result` or `Option` (or another type that implements `FromResidual`)
   --> src/main.rs:6:37
    |
4   | / fn main() {
5   | |     let mut file = File::create("foo.txt")?;
6   | |     file.write_all(b"Hello, world!")?;
    | |                                     ^ cannot use the `?` operator in a function that returns `()`
7   | | }
    | |_- this function should return `Result` or `Option` to accept `?`
    |
    = help: the trait `FromResidual<Result<Infallible, std::io::Error>>` is not implemented for `()`
note: required by `from_residual`
   --> /home/jsha/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/try_trait.rs:339:5
    |
339 |     fn from_residual(residual: R) -> Self;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

For more information about this error, try `rustc --explain E0277`.

This is a problem I remember running into when I was first learning the language. I suspect it may be an issue for other language learners too.

I propose changing the boilerplate to:

use std::error::Error;

fn main() -> Result<(), Box<dyn Error>> {
    println!("Hello, world!");
    Ok(())
}

This compiles cleanly and passes clippy.

[jfrimmel]

I have mixed feelings above this. Personally I use plain fn main() more often than fn main() -> Result<_, _>. Therefore the current cargo template suits my style more than the proposed code. But I'm fine both ways.

But: I think we should improve the current diagnostics to output something like that:

error[E0277]: the `?` operator can only be used in a function that returns `Result` or `Option` (or another type that implements `FromResidual`)
   --> src/main.rs:6:37
    |
4   | / fn main() {
5   | |     let mut file = File::create("foo.txt")?;
6   | |     file.write_all(b"Hello, world!")?;
    | |                                     ^ cannot use the `?` operator in a function that returns `()`
7   | | }
    | |_- help: add a fallible return type: `fn main() -> Result<(), Box<dyn std::error::Error>`

[camelid]

I also rarely use fn main() -> Result<_, _>. Personally, I think it'd be preferable to just improve the error rather than changing the output of cargo new. I appreciate that the current output of cargo new is quite simple and, at least for binaries, doesn't add extra boilerplate that I don't need.

[jsha]

My experience as a beginning Rust dev was that I often needed to cut-and-paste examples from documents, and I was often frustrated by them not working until I remembered to add a -> Result to my main. I also found that the error messages at the time led me down a path of writing Results with a concrete error type rather than a dyn Error. Looking through my juvenilia I see:

fn main() -> io::Result<()> {
fn main() -> Result<(), Box<dyn error::Error>> {
fn main() -> Result<(), Box<dyn std::error::Error>> {
fn main() -> Result<(), Box<dyn std::error::Error + Send + Sync + 'static>> {
fn main() -> Result<(), String> {

I think it's really useful for new users if cargo new offers a simple solution that "just works" in almost all cases. Or to put it another way: as experienced devs, we know we can remove the-> Result boilerplate if we don't want it; or we can just write our own boilerplate, since it's so short. The time spent by experienced devs deleting an unwanted -> Result is outweighed by the frustration experienced by new devs when their copy-pasted code doesn't compile. Learning Rust can be a frustrating experience, and if there's a common error that almost all beginners are likely to hit, I think it's really valuable to remove that error from their experience entirely.

[camelid]

I didn't hit this as a beginner Rust programmer. I would manually handle my errors and then call process::exit if there was an error. I also don't think I copied and pasted code that would have caused this error.

So maybe many other people did hit this when they were beginners, but I myself don't have the perspective of hitting this as a beginner.

[camelid]

Also, Box<dyn Error> can add its own confusion in my experience since I sometimes have to cast manually with Box::from or similar to fix inference errors, which even as an experienced user I find confusing.

[weihanglo]

This should be moved to rust-lang/cargo, no?

Anyway, Cargo has this semi-related issue about cargo-template #5151.

[Enselic]

Triage: Yep, this should be moved to cargo. Can someone with sufficient privileges do that please?

[epage]

@estebank Independent of what we do for cargo init, I feel like this is something a new user could quickly run into and could be made a bit more user friendly. If nothing else, I could see detecting main() and suggesting to change the return type to be Box<dyn Error>.

Thoughts?

[estebank]

@epage yes, I agree we should do that. I am unsure if there's a ticket for it already. We should create one in the compiler's repo for improving this output. We already detect fn main() for to not suggest changing the return type in E0308.

[epage]

I went through about 3 pages of diagnostic issues with the error code in them and didn't see something related, so I created rust-lang/rust#125997