r? @SimonSapin

Note that this is somewhat a wallpaper-ish patch, since the string would be lowercased if its length is less that the bindings.

We can disable bindings altogether otherwise I guess.

I think you’re right: we can’t have all three of lowercasing into a buffer on the stack to avoid allocation, support bindings in match arms, and get exactly the behavior of match &*input.to_ascii_lowercase() { /* … */ } (other than allocation). I’m also tempted to drop support for bindings in match arms.

@canaltinova What was your use case for bindings? Can you write your code without? Maybe replacing them like this:

-match_ignore_ascii_case! { &expect_keyword(),
+let keyword = expect_keyword();
+match_ignore_ascii_case! { &keyword,
     "foo" => Foo,
     "bar" => Bar,
-    keyword => Other(keyword.to_owned()),
+    _ => Other(keyword.to_owned()),
 }

will-change property accepts <custom-ident>. I was using this to convert to lowercase and exclude some unacceptable identifiers. Here's how I use this macro:

match_ignore_ascii_case! { &i.expect_ident()?,
    "will-change" | "none" | "all" | "auto" |
    "initial" | "inherit" | "unset" => Err(()),
    ident  => {
        Ok((Atom::from(ident)))
    }
}

Ah, <custom-ident> is actually case sensitive. I think I can live with that 😃

Ah, yes. If you want to get the lower-case version of an arbitrary string (of unbounded length), not just match it against a set of pre-defined patterns, then you should use s.to_ascii_lowercase() + match instead of match_ignore_ascii_case!.

Thanks for the PR @emilio but I think I’d rather remove support for match bindings. With this PR the bound value would sometimes be lower-cased and sometimes not, depending on its length. Closing, I’ll work on another PR now.