Custom derive: can't use const as attribute value

[srijs]

I'm using serde to implement deserialization for a custom protocol. To keep the code dry, I've pulled all relevant strings into constants, but that creates a problem when trying to use the constants as attributes values in serde's derive macro.

Here's a playground with a minimal example to reproduce the problem: http://play.rust-lang.org/?gist=47230ae9671b24b312bef9540e520ddf&version=stable&mode=debug&edition=2015

The check that seems to produce the error is part of libsyntax:

rust/src/libsyntax/parse/attr.rs

Lines 261 to 263 in 0db03e6

	let found = self.this_token_to_string();
	let msg = format!("expected unsuffixed literal or identifier, found {}", found);
	Err(self.diagnostic().struct_span_err(lo, &msg))

With my very basic knowledge of the rust compiler, I can see two potential ways this could be made possible:

1. Run derive macros after the constants have been expanded. This would be transparent to the macros (they could treat the expanded constants as literals), however this might be a no-go due to the way that the passes are ordered within rustc?

2. Relax the check in libsyntax (and possibly other places) to allow for identifiers to be valid attribute values on the syntax level, so that custom derive macros could use the identifier as part of their code gen (this wouldn't work for all macros, but at least for serde it looks like it'd be good enough).

If possible, could you give an indication as to whether this is something that you'd like in the compiler (in which case I'd be interested in working on a PR or RFC for it), or whether you think this is either impossible or undesirable for any reason?

[webern]

I'm surprised this issue doesn't have more upvotes. Is this issue represented somewhere else where it has more of a following? What I want to do is this:

const SOME_CONST: &str = "foobar";

#[some_derive_macro(name = SOME_CONST)]
struct Whatever {

}

[thomcc]

Yeah, you can't put consts in attributes, but it would be very nice. I'm not sure if solving this requires an RFC, but it might.

EDIT: For proc macros, the macro could probably solve it in many cases, but would have to be changed.

[oldgalileo]

The include_str! macro seems related, but I'm curious if there's been any work unlinked here that enables the usage of a const _: &str in an attribute yet?