How to test I/O errors

[brson]

I just posted a fix for a file system error that I can't test for. rustup error handling is important but hard to test. In this case I would want to trigger out-of-disk during an installation. I'd prefer to be able to test this stuff.

I see two options:

• Write a fuse filesystem to create the errors. Requires a lot of work, only works for file system access, not network.
• Add mocking to the rustup-utils I/O functions, so that the test runner can e.g. set environment variables telling the I/O layer to generate specific errors. Downside is that it's ugly and isn't exercising big parts of the I/O stack.

I'm inclined toward the latter. @alexcrichton do you have opinions?

[brson]

As a pipe dream: wouldn't it be awesome if I could inject behavior into std's I/O stack? With std-aware cargo I can foresee replacing the I/O crate with an I/O crate that can introduce mock behavior.

[alexcrichton]

We may only need great coverage here on at least one platform, so I wonder if we could use linux containers in one way or another here? I think that would allow us to basically intercept all system calls and set custom quotas for things. The tests might be brittle over time but perhaps that's not so bad?

I agree that of those two options the latter is better, but I wouldn't be too thrilled about a mocking framework there :(