Crates using placement syntax behind a cfg broken

[pietroalbini]

Some crates were using <- behind a feature flag, so they worked on stable even if the placement syntax was nightly-only. After removing it though they fail to compile even on stable with a syntax error, because <- is not recognized anymore.

• mbox regressed from stable to beta (build log) cc @kennytm
• rayon-hash regressed from stable to beta (build log) cc @cuviper

[kennytm]

It is already fixed in mbox 0.5.0 and 0.4.4.

[nikomatsakis]

Oh, that's annoying. We should patch rayon-hash.

[petrochenkov]

This is not the first time this is happening, removal of impl Trait for .. {} caused same consequences (see #46480).
We should probably continue parsing <- and report the error later in AST validation.

[nikomatsakis]

Discussed in the @rust-lang/compiler meeting. I think we reached this consensus:

• in this case, yes, we should do as @petrochenkov suggests: keep parsing, error in AST validation
• if in the future, if it happens that we want to repurpose the syntax, we deal with it then: best of course is to use an edition for it, but maybe by then there no longer exist crates using it
• going forward, it would be ideal if we could do some kind of "pre-cfg / expansion" feature check to try and avoid this in the future; that may or may not be feasible though, and we couldn't do it retroactively for all existing things anyway

[nikomatsakis]

Some discussion of the final at or around here in the IRC logs

https://botbot.me/mozilla/rustc/2018-05-17/?msg=100165163&page=3

[nikomatsakis]

triage: P-high

[nikomatsakis]

Would anyone have time to whip up a change to the parser?

cc @rust-lang/compiler @aidanhs

[nikomatsakis]

For reference, the PR that reverted the syntax was #48333

[cuviper]

I am working on rayon-hash. The quick hack would be to just nuke the placement stuff, but I'm trying to do a proper update to catch up to std's code -- fun with git subtree.

But frankly, I don't know of anyone really using rayon-hash anyway.

[cuviper]

Fixed in rayon-hash 0.3.0.

[nikomatsakis]

I'll handle this.

[nikomatsakis]

Restored in #51052

[leoyvens]

We could try calling more attention to lib authors about this through the API guidelines.

[estebank]

The two regressions where about <-, but we also reverted the in syntax as well, which has degraded back a couple of errors, like the following:

fn main() {
    let i = ();
    for x in in 0..1 {}
    i;
}

Would it be ok to create a PR removing the in parsing while keeping the current <- handling for a while longer?

CC #36611 #52964 #51602