macros: Remove 'r#' prefix from raw identifiers in field names

[dtolnay]

Motivation

Currently a call like info!(..., r#type = "...") ends up emitting a field named "r#type", which is never desirable. Previous work in #2925 made the desired behavior expressible as info!(..., type = "..."), and even before that PR one could write it as info!(..., "type" = "...").

However, when dealing with macro-generated use of tracing, we do still sometimes end up with raw identifiers. For example, if user-provided identifiers are used for both declaring (or instantiating, or accessing) a struct, and also for tracing. Something like this:

example!(name, r#type);

// expands to:

...
info!(..., name = GLOBAL.name, r#type = GLOBAL.r#type);
...

If we ask the user of example! to pass type (unraw), there is no way for the macro to access GLOBAL.r#type from that. But if we ask the user of example! to pass r#type (raw), there is no straightforward way for the macro to trace the field name as "type". The best option would have been something like:

info!(
    ...
    $(
        {
            (const {
                if let Some((b"r#", unraw)) = stringify!($field).as_bytes().split_first_chunk() {
                    unsafe { core::str::from_utf8_unchecked(unraw) }
                } else {
                    stringify!($field)
                }
            })
        } = GLOBAL.$field,
    )*
);

but this gets even more ridiculous when dealing with field names that are not just a single identifier. (prefix.r#keyword.suffix).

Solution

r#keyword = $value is made equivalent to keyword = $value. The field will be emitted with the name "keyword", not "r#keyword" as before.

[kadenlnelson]

Interesting... I'm ignorant to this so forgive me if I'm understanding this wrong.

Assuming that example! is an external macro, is it possible that this behavior could be solved within the example macro? Entertain me for a minute, but it appears that the tracing macros (ie, info!) are behaving as expected. Since the example macro renders what is ultimately fed into info!, that is where I feel the behavior should be addressed.

Now, that is my 2c when viewing this as a bug, and I'm still not certain whether I would personally classify this as a bug or a UX enhancement.