feat: Added warning when failing to update index cache

[ranger-ross]

What does this PR try to resolve?

Fixes #13712

Adds a warning if there is an error updating the index cache.
It also attempts to avoid warning spam as requested in this comment

Below is an example output

    Updating crates.io index
warning: failed to write cache, path: /home/ross/.cargo/registry/index/index.crates.io-1949cf8c6b5b557f/.cache/ba/se/base64, error: Permission denied (os error 13)
   Compiling serde v1.0.217
   Compiling r-efi v5.2.0
   Compiling base64 v0.22.1
   Compiling cargo-13712 v0.1.0 (/home/ross/projects/cargo-13712)
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.20s

How should we test and review this PR?

I tested this on my machine by manually changing the owner/permission of the index files

sudo chown root ~/.cargo/registry/index/.../.cache/r-
sudo chmod 700 ~/.cargo/registry/index/.../.cache/r-

# in a project that uses the `r-efi` crate
cargo build 

I wasn't quiet sure about the best way to add a test to the testsuite for this. 😅
If there is good way to create a test for this lmk

[rustbot]

r? @weihanglo

rustbot has assigned @weihanglo.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[weihanglo]

We have a similar test, maybe we can take reuse that test to assert their stderr?

cargo/tests/testsuite/registry.rs

Lines 3053 to 3098 in 2939e96

	fn readonly_registry_still_works() {
	Package::new("foo", "0.1.0").publish();
	let p = project()
	.file(
	"Cargo.toml",
	r#"
	[package]
	name = "a"
	version = "0.5.0"
	edition = "2015"
	authors = []
	[dependencies]
	foo = '0.1.0'
	"#,
	)
	.file("src/lib.rs", "")
	.build();
	p.cargo("generate-lockfile").run();
	p.cargo("fetch --locked").run();
	chmod_readonly(&paths::home(), true);
	p.cargo("check").run();
	// make sure we un-readonly the files afterwards so "cargo clean" can remove them (#6934)
	chmod_readonly(&paths::home(), false);
	fn chmod_readonly(path: &Path, readonly: bool) {
	for entry in t!(path.read_dir()) {
	let entry = t!(entry);
	let path = entry.path();
	if t!(entry.file_type()).is_dir() {
	chmod_readonly(&path, readonly);
	} else {
	set_readonly(&path, readonly);
	}
	}
	set_readonly(path, readonly);
	}
	fn set_readonly(path: &Path, readonly: bool) {
	let mut perms = t!(path.metadata()).permissions();
	perms.set_readonly(readonly);
	t!(fs::set_permissions(path, perms));
	}
	}

[ranger-ross]

thanks, I was able to use this to create a test based off of this in 9f2d0a3

[weihanglo]

Shameless copied from #15026 (comment).

We generally ask that PRs are split into

• One commit that adds the test, showing the currently broken behavior (ie it passes)
• The fix commit that updates the test to pass with the change

This makes the diff between the two commits show the change in behavior

Mind doing that?

[ranger-ross]

Sure, I re-ordered the commits so that the test was added before the fix.
Let me know if I did something incorrectly

[weihanglo]

My original idea is that both commits should pass CI. And the second commit contains changes in both the logic fix and the test, so the test change could reflect how the behavior has changed with the fix. See this PR as an example #15036. Anyway it doesn't really matter, and thank you!

[weihanglo]

Should we add more than one dependency, so we know it really emits only once?

[ranger-ross]

I was testing with this and notice the dependencies would get compiled in different orders causing .with_stderr_data() to fail sometimes.

But I think this can be worked around by using the [..] to ignore the names since they are not the point of this test.

    Package::new("foo", "0.1.0").publish();
    Package::new("fo2", "0.1.0").publish();

// ...

        .with_stderr_data(str![[r#"
[WARNING] failed to write cache, path: [ROOT]/home/.cargo/registry/index/-[HASH]/.cache/3/f/fo[..], [ERROR] Permission denied (os error 13)
[COMPILING] fo[..] v0.1.0
[COMPILING] fo[..] v0.1.0
[COMPILING] a v0.5.0 ([ROOT]/foo)
[FINISHED] `dev` profile [unoptimized + debuginfo] target(s) in [ELAPSED]s

"#]])

I think something like this should.
But let me know if you are aware of a smarter way to do this. 👍

[weihanglo]

We do have the unordered method to relax the snapshot comparsion a bit, though your approach is pretty great!