Avoid processing features on unconfigured crates

[jseyfried]

Fixes #34932, a regression caused by #34272.
r? @nrc

[retep998]

Why a compile fail test? Isn't the idea that disabling the crate with #![cfg] should allow it to compile despite use of #![feature]?

[jseyfried]

@retep998 feature doesn't prevent compilation on the nightlies since it is only disabled on stable/beta.

This test still tests that feature on an unconfigured crate is not processed. On stable/beta, processing feature causes an error (since feature is disabled), but on nightlies, it prevents gated feature errors.

[retep998]

For example

#![cfg(feature = "nightly")]
#![feature(some_feature)]
#[test] fn foo() {
    // code which relies on some_feature
}

If the nightly feature isn't enabled, it will compile on the current stable. However it will not compile in the current beta (or at least whatever is on the current playground). The regression is that the stability check for features is done even when the entire crate is disabled with a cfg.

So the correct test would need to make sure that this example would still compile in a stable environment where features are disabled. And compile-fail is supposed to make sure that the code fails to compile, right? Or am I crazy and it actually tests to make sure that it does compile?

[jseyfried]

@retep998 I know, this fixes that regression. The test I added is equivalent since features are stability checked when they are processed, so testing that features are not processed also tests that they are not stability checked.

If you know of a more direct way to test this, let me know -- I'm not sure how to add a test with stable/beta feature stability checking.

[eddyb]

// compile-flags:--test
// rustc-env:RUSTC_BOOTSTRAP_KEY=

#![cfg(feature="does-not-exist")]
#![feature(iter_arith_traits)]

// nothing below actually matters, but I added it anyway
#[test]
fn dummy() {
    let () = "this should not reach type-checking";
    panic!("this should not run");
}

At least when building beta/stable, this test will require no feature's to be observed, in order to compile.
Although the feature being used should be something we know we aren't going to ever stabilize (rustc_attrs?).

[retep998]

You could add a #![deny(unused_features)] to make sure it is tested on nightly as well, since currently on nightly it passes with a warning about an unused feature.

[jseyfried]

@eddyb @retep998 I amended the regression test with your suggestions.

[eddyb]

LGTM. cc @alexcrichton Is this the best we can do about testing?

[alexcrichton]

I don't have too many opinions, so for me if it works it works!

[eddyb]

You'll also need (found by Travis):

// ignore-pretty : (#23623) problems when  ending with // comments

[jseyfried]

Done.

[nrc]

@bors: r+

[bors]

📌 Commit 64d36cc has been approved by nrc

[bors]

⌛ Testing commit 64d36cc with merge bc81a3f...

[bors]

💔 Test failed - auto-win-gnu-32-opt

[alexcrichton]

@bors: retry

On Sun, Jul 24, 2016 at 11:24 PM, bors [email protected] wrote:

💔 Test failed - auto-win-gnu-32-opt
https://buildbot.rust-lang.org/builders/auto-win-gnu-32-opt/builds/5057

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#34969 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAD95OpRVtuUcZ7tagwxEz2XH4p5ieWaks5qZFaLgaJpZM4JSTZy
.

[alexcrichton]

@bors: retry

[bors]

⌛ Testing commit 64d36cc with merge 362d04d...

[bors]

💔 Test failed - auto-win-msvc-64-opt

[alexcrichton]

@bors: retry

On Tue, Jul 26, 2016 at 3:36 AM, bors [email protected] wrote:

💔 Test failed - auto-win-msvc-64-opt
https://buildbot.rust-lang.org/builders/auto-win-msvc-64-opt/builds/5074

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#34969 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAD95CIJfzLkYEJD09uJsE5m2uR6CzoAks5qZeMcgaJpZM4JSTZy
.

[nikomatsakis]

Accepting for backport: regression, small.