Simplify match blocks into ? operator using SSR

[MarijnS95]

Extracted from #991

While addressing minor code inconsistency I stumbled upon a commit from Heftig using rerast for structural refactoring, and it turns out Rust-analyzer supports SSR as well. This made it possible to easily do a bulk of conversions to use ? instead of manual match blocks. See the relevant commits for the rules used.

One minor deficiency is the inability to repeat labels: it is impossible to specify Some($a) => $a, meaning that a manual check is necessary to see if the right hand side wasn't doing any extra conversion. These had to be addressed by hand, but weren't many.

[sdroege]

Why? It looks nicer to have it aligned properly and the output is the same

[MarijnS95]

I don't find \n\ particularly readable, but then again that allows to have leading spaces as indentation.

Would this work? (Notice the "\):

            "\
impl fmt::Display for {0} {{
\tfn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {{
\t\twrite!(f, \"{0}::{{}}\", match *self {{
",

That also allows \t to be substituted with four spaces, the native indentation.

[sdroege]

That would also work for me (also with four leading spaces).

Just it would be nice to have the code in the string consistently indented, otherwise it's confusing to read.

[MarijnS95]

also with four leading spaces

That would resurface the need for trailing \n\ again. I'll remove these changed from the PR as they're merely extra noise.

[sdroege]

Ok :) Maybe let's handle that separately then, @GuillaumeGomez probably also has opinions on this topic

[sdroege]

Not much, what matters is that it's readable inside the gir code. The autogenerated code itself will be cleaned up anyway, as you say.

[MarijnS95]

It seems that'd improve a lot with quote! and the way it can expand arrays or iterators, and in general colorize target code in the IDE. But it appears the need for // comments with nested /**/ completely breaks this unfortunately :(.

[GuillaumeGomez]

No, quote! didn't improve the readability at all.

[sdroege]

Yeah I tried to use quote here but it only made things worse in practice. Maybe I just used it wrongly though...

[MarijnS95]

I'd argue:

        let type_name = format_ident!("{}", enum_.name);

        let members = members.iter().map(|m| {
            let name = &m.name;
            let ident = format_ident!("{}", name);
            // TODO: version_condition_tokens(env, m.version, false, 3) -> Option<TokenStream>
            quote!(Self::#ident => #name)
        });

        quote! {
            impl fmt::Display for #type_name {
                fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
                    match *self {
                        #( #members, )*
                        _ => "Unknown",
                    }
                }
            }
        }

Is way easier to read than:

        writeln!(
            w,
            "impl fmt::Display for {0} {{\n\
             \tfn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {{\n\
             \t\twrite!(f, \"{0}::{{}}\", match *self {{",
            enum_.name
        )?;
        for member in &members {
            version_condition(w, env, member.version, false, 3)?;
            writeln!(w, "\t\t\t{0}::{1} => \"{1}\",", enum_.name, member.name)?;
        }
        writeln!(
            w,
            "\t\t\t_ => \"Unknown\",\n\
             \t\t}})\n\
             \t}}\n\
             }}\n"
        )?;

But then there are quite a few things that are not ergonomic:

• Not able to expand members in-line (ie. #(m.name));
• Not able to convert strings to identifiers within the macro.

Not to mention some of the issues you were finding in #919, which "simply" needs some extra tooling around.

These in turn make it impossible to have something like (pseudo-syntax) #( Self::#(/*turn into identifier*/ #(members).name) => #(#(members).name), )* directly in place of the map.

Other than that the ability to expand iterators and deal with Option<> in expansion is pretty neat.

[sdroege]

Looks good to me otherwise, thanks!

[sdroege]

The PR looks good for me now, let's worry about those string formatting details some other time.

@GuillaumeGomez ?

[GuillaumeGomez]

Thanks!