Replace hardcoded forward slash with path::MAIN_SEPARATOR

[laumann]

Fixes#40149

[rust-highfive]

r? @BurntSushi

(rust_highfive has picked a reviewer for you, use r? to override)

[BurntSushi]

@bors r+

[bors]

📌 Commit cb4f536 has been approved by BurntSushi

[BurntSushi]

@laumann Thanks!

[laumann]

@BurntSushi Thank you :-)

Also came across https://github.com/rust-lang/rust/blob/master/src/libsyntax/parse/parser.rs#L5307 - can it be fixed here or is it too late?

[BurntSushi]

@bors r-

[BurntSushi]

@laumann Probably not a good idea to add commits after a PR has been r+'d. I think it will rightyfully cause the PR to fail to merge.

The new changes look OK. Could you please squash them down to a single commit please?

[laumann]

@BurntSushi OK, can I fix it here, or should I open a new PR?

[BurntSushi]

@laumann You can fix it here by squashing your branch locally (probably with a rebase) and then force pushing.

[laumann]

OK, no problem, thanks :-)

[laumann]

Rebased. Something went wrong with https://travis-ci.org/rust-lang/rust/jobs/212496260, but I can't really tell what.

[bors]

☔ The latest upstream changes (presumably #40346) made this pull request unmergeable. Please resolve the merge conflicts.

[BurntSushi]

@laumann You need to merge changes from master back into this PR because someone else has made changes since submitting your PR that conflict with your changes.

You can do this by checking out your master branch and re-syncing it with master on rust-lang/rust, usually with git pull origin master --ff-only (assuming origin points to rust-lang/rust and not your fork). Then checkout your PR branch, git checkout slash-in-diagnostics-path, and rebase it on to master.

[laumann]

@BurntSushi Rebased.

[BurntSushi]

@laumann Thanks!

[BurntSushi]

@bors r+

[bors]

📌 Commit 4f0141b has been approved by BurntSushi

[frewsxcv]

I think this is causing a test failure on Windows:

https://ci.appveyor.com/project/rust-lang/rust/build/1.0.2460/job/kfyqxp0uxi22y3lk

@bors r-

[laumann]

@frewsxcv You are right - it looks like test/parse-fail/mod_file_not_exists.rs which makes sense :-) I'll get it fixed!

Why is that test in parse-fail and not compile-fail?

[laumann]

It seems the issue can be fixed several ways, but which one would be preferred?

1. Duplicate the test in parse-fail/mod_file_not_exists_windows.rs and add:

// ignore-gnu
// ignore-android
// ignore-bitrig
// ignore-macos
// ignore-dragonfly
// ignore-freebsd
// ignore-haiku
// ignore-ios
// ignore-linux
// ignore-netbsd
// ignore-openbsd
// ignore-solaris
// ignore-emscripten

and add ignore-windows to parse-fail/mod_file_not_exists.rs

2. Convert the test to a run-make test (suggested by @jseyfried), to make it easier to switch on OS

But I'm wondering what the preferred approach is...

I tried fiddling with #[cfg(windows)] and something like:

#[cfg(windows)]
mod not_a_real_file; //[windows]~ ERROR file not found for module `not_a_real_file`
//[windows]~^ HELP name the file either not_a_real_file.rs or not_a_real_file\mod.rs inside the directory
#[cfg(not(windows))]
mod not_a_real_file; //[not(windows)]~ ERROR file not found for module `not_a_real_file`
//[not(windows)]~^ HELP name the file either not_a_real_file.rs or not_a_real_file/mod.rs inside the directory
fn main() {
assert_eq!(mod_file_aux::bar(), 10);
}

But that trick didn't work :-)

[laumann]

@BurntSushi Should I prefer to amend the existing commit, or should I just add new ones (to fix the test)?

[BurntSushi]

@laumann Amend the existing commit. We shouldn't have broken commits. :-)

[laumann]

@BurntSushi The build failed, but I can't really tell why - is there some way to try again?

[BurntSushi]

@laumann I kicked it.

[laumann]

@BurntSushi Yay! Thanks :-)

[laumann]

@frewsxcv It should be fixed now, not sure if I need to do anything else.

[BurntSushi]

It doesn't seem OK to ignore this test on Windows.

@alexcrichton What should we do here? This PR is changing a few places to use the native path separator when printing file paths, so I assume this test is failing because the error messages don't compare equal. Is there a way to do conditional testing based on platform?

[BurntSushi]

I now see that there is another test in this PR, mod_file_not_exist_windows.rs, which essentially does this. But it's done by exhausting every non-Windows platform. Is that the right way to do this?

[laumann]

@BurntSushi I considered different approaches - this could maybe be turned into a run-make test instead.

I went for this solution, because I looked at src/test/codegen/dllimports/main.rs (and others) that do windows-specific things by exhausting the other options.

[BurntSushi]

Precedence is good enough for me.

[BurntSushi]

Thanks!

[laumann]

Thank you :-)

Ah, of course, you have to r+ it! I was wondering why it wasn't being tested again...

[BurntSushi]

@bors r+

[bors]

📌 Commit 25130da has been approved by BurntSushi

[bors]

🔒 Merge conflict

[laumann]

@BurntSushi Should I rebase again?

[BurntSushi]

Yup. Looks like something else got merged that conflicts with this, so you'll need to rebase to resolve conflicts.

[laumann]

Will do

[laumann]

Seems to be good now. 🤞

[BurntSushi]

@bors retry

[alexcrichton]

@bors: r=BurntSushi

[bors]

📌 Commit b376386 has been approved by BurntSushi

[bors]

⌛ Testing commit b376386 with merge 474f7a9...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: BurntSushi
Pushing 474f7a9 to master...