large_enum_variant triggered in error-chain macro has unhelpful help message

[killercup]

You know how I like to suggest E-hard new lints? Sorry, this time I only have an E-hard L-enhancement to report.

warning: large size difference between variants
  --> src/main.rs:66:1
   |
66 | / error_chain! {
67 | |     foreign_links {
68 | |         Io(::std::io::Error);
69 | |         Handlebars(::handlebars::TemplateRenderError);
70 | |         Yaml(::yaml::Error);
71 | |     }
72 | | }
   | |_^
   |
   = note: #[warn(large_enum_variant)] on by default
   = help: for further information visit https://rust-lang-nursery.github.io/rust-clippy/v0.0.165/index.html#large_enum_variant
help: consider boxing the large fields to reduce the total size of the enum
   |
113| err : Box<$ foreign_link_error_path> ) {
   |       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   = note: this error originates in a macro outside of the current crate
   = issue author's remark: this help message is not helpful

Is there a way to (easily) print the enum variants with their sizes? This might be helpful even in rather obvious cases.

[oli-obk]

is there a point in reporting anything inside that macro? We can just turn it off in macros.

[killercup]

Well, the concern is valid: It's a large enum. Question is, if the programmer can do anything about it (in this case: probably). It's just the quoted code that's not really useful here.

[oli-obk]

can you tell error_chain to box the args?

Something like

error_chain! {
    foreign_links {
        Io(Box<::std::io::Error>);
    }
}

?

[oli-obk]

Is there a way to (easily) print the enum variants with their sizes? This might be helpful even in rather obvious cases.

That is trivial.

Figuring out how to produce nice error spans is a little involved, but sound like "just some detective work"

[killercup]

can you tell error_chain to box the args?

Doesn't look like it: https://play.rust-lang.org/?gist=0007d7af1676d861d6cb534f4fa735c1&version=stable

---

Sorry, this time I only have an E-hard L-enhancement to report.

oli-obk added E-easy E-medium L-enhancement T-middle

😞

[oli-obk]

error-chain already seems to box its contents: https://play.rust-lang.org/?gist=232208cd5ee6f1fcf95c4eb52562dfde&version=stable

[killercup]

Hahaha, sorry, hadn't had enough coffee yet. That's basically the quoted macro line from above! 😆

113| err : Box<$ foreign_link_error_path> ) {

[Arnavion]

@killercup

can you tell error_chain to box the args?

Doesn't look like it: https://play.rust-lang.org/?gist=0007d7af1676d861d6cb534f4fa735c1&version=stable

error-chain is totally fine with letting you have a foreign link of Box<io::Error>, since Box<io::Error> does impl std::error::Error that foreign links require.

The error in your link is because you told error-chain that the foreign link is over Box<io::Error>, so the EK impls From<Box<io::Error>>. But your code is trying to use ? with an unboxed io::Error for which there is no From impl.

---

@oli-obk

error-chain already seems to box its contents: https://play.rust-lang.org/?gist=232208cd5ee6f1fcf95c4eb52562dfde&version=stable

No, error-chain's codegen does not box foreign links. The size of foo::Error and bar::Error in your playground is the same because of the effect of the implicit Msg(String) member that occupies 24 bytes, leading to 32 bytes total to include the discriminant. See:

mod foo {
    pub enum ErrorKind {
        Msg(String),
        Io(Box<::std::io::Error>),
        #[doc(hidden)]
        __Nonexhaustive { },
    }
}

mod bar {
    pub enum ErrorKind {
        Msg(String),
        Io(::std::io::Error),
        #[doc(hidden)]
        __Nonexhaustive { },
    }
}

fn main() {
    println!("{}", std::mem::size_of::<foo::ErrorKind>()); // 32
    println!("{}", std::mem::size_of::<bar::ErrorKind>()); // 32
    println!("{}", std::mem::size_of::<String>()); // 24
    println!("{}", std::mem::size_of::<::std::io::Error>()); // 16
    println!("{}", std::mem::size_of::<Box<::std::io::Error>>()); // 8
}