"use of undeclared lifetime name `'static`" is misleading

[solson]

Code: https://is.gd/6L8tEl

fn foo<'a: 'static>() {}

Output:

rustc 1.17.0-nightly (b1e31766d 2017-03-03)
error[E0261]: use of undeclared lifetime name `'static`
 --> <anon>:1:12
  |
1 | fn foo<'a: 'static>() {}
  |            ^^^^^^^ undeclared lifetime

Expectation:

<'a: 'static> is unnecessary (you can just use 'static in place of 'a), but maybe we should accept it? At any rate, the error message is misleading - 'static is the one globally declared lifetime. If we want to reject this code, the error message should not imply 'static is undefined, but rather that you just don't need the 'a lifetime.

[jseyfried]

I believe we'll want to check for 'static here.

[eugene-bulkin]

The right move here would be to reject 'static here because 'a is superfluous in that scenario, right? I'm not entirely sure what the performance implications would be of just accepting it since it's the same as doing foo<'static>.

Since it would be unnecessary, would be more appropriate to reject the code telling the user to just use 'static instead?

[adamransom]

I'd like to work on this, if it is still open (and/or available for mentoring)!

[jseyfried]

@adamransom
Yeah, go ahead! Feel free to ask questions here or ping me on IRC (jseyfried in #rustc or #rust-internals).

[jseyfried]

@eugene-bulkin

it's the same as doing foo<'static>

fn foo<'static>() {} isn't allowed.

would [it] be more appropriate to reject the code telling the user to just use 'static instead?

I think it would appropriate to reject the code and tell the user not to use a lifetime parameter at all -- there's never a need for a parameter that must always be 'static.

That being said, we also might want to allow it for consistency, perhaps with a warning (lint).
cc @rust-lang/lang @petrochenkov

[eugene-bulkin]

fn foo<'static>() {} isn't allowed.

Oh, well in that case it makes even more sense to just warn the user that the parameter is unnecessary; an error might be overkill, but a warning would seem reasonable.

[adamransom]

@jseyfried Great, I'll work on adding a warning as we discussed in IRC.

[petrochenkov]

I think it would appropriate to reject the code and tell the user not to use a lifetime parameter at all -- there's never a need for a parameter that must always be 'static.
That being said, we also might want to allow it for consistency, perhaps with a warning (lint).

Both seems reasonable.
Maybe this warning can be shoehorned into some existing lint? Creating a new one just for this seems wasteful. (Maybe this should be an error just to avoid adding a new lint.)