Gate shell_completions tests on availability

[ehuss]

This changes the shell_completions tests to only run if the corresponding external program is installed. Generally, cargo's tests shouldn't require any external commands to be installed.

This also uses the ignore attribute for disabling on macos. It is quite confusing when running on macos to see a successful test when it isn't running. This ensures that the test output indicates that it is ignored and the reason.

cc @shannmu

[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

[ehuss]

Hm, so this approach isn't going to work. It's too late for me to continue looking at this, will maybe look at it tomorrow or the next few days.

[weihanglo]

error: failed to install component: 'llvm-tools-preview-x86_64-unknown-linux-gnu', detected conflict: 'lib/rustlib/x86_64-unknown-linux-gnu/bin/llc'

@bors try

[bors]

⌛ Trying commit fdf4f26 with merge fe6e210...

[bors]

💔 Test failed - checks-actions

[epage]

I recommend we be free with reverts when a code change is blocking. We can always fix this in another pass.

[weihanglo]

rust-lang/rust#130292 is merged. This shouldn't be blocked anymore.

@bors r+

[bors]

📌 Commit fdf4f26 has been approved by weihanglo

It is now in the queue for this repository.

[bors]

⌛ Testing commit fdf4f26 with merge 543b910...

[ehuss]

@bors r-

This isn't going to work, since not all jobs have the shells installed.

I'm looking at a different approach where the jobs will be identified as being "extended" to test these extra things that might not be available.

[ehuss]

And BTW, this will block updating in rust-lang/rust, since those won't have the shells installed either. I'm poking around now, but I'm a little confused why the fish test seems to be randomly failing.

[shannmu]

The Fish shell test failures might be due to the test depending on the completest crate, which tests the completion results based on an interactive shell. We cannot be certain when the completion results are fully generated, so there is a built-in timeout parameter. If the completion results are not generated within the timeout period, it may cause the test to fail.

[ehuss]

Hmm. That's unfortunate.

Let's just disable the tests for now. I have opened #14546 to do that, and #14545 lists the problems that I noticed while looking at this.

I'm probably not going to look at this further for now.