A flag to unset all pre-existing variables

[tabokie]

It would make tests more reproducible.

[vmx]

That's a good idea. Though I don't think we need a new flag, this can be done already today with something along the lines of:

let vars_to_unset = std::env::vars().map(|(key, _value)| (key, None)).collect();
temp_env::with_vars(
    [
        vars_to_unset,
        vec![("MY_VAR".to_string(), Some("my_value"))],
    ]
    .concat(),
    || {
        println!("vmx: env without pre-existing vars: {:?}", std::env::vars());
    },
);

Does that solve your problem?

[tabokie]

Calling std::env::vars() directly will again be interfered by other test cases. E.g. test A unsets some variables, test B collects vars_to_unset missing those variables, then test A finishes and test B enters temp_env.

But I agree it'll work if everyone is purely adding additional variables instead of unsetting some.

[vmx]

Calling std::env::vars() directly will again be interfered by other test cases. E.g. test A unsets some variables, test B collects vars_to_unset missing those variables, then test A finishes and test B enters temp_env.

If I understand it correctly, you would set the vars_to_unset once (with a OnceCell` or so) and the other tests would re-use them.

[tabokie]

But that would force every tests to run the setup even if they don't necessarily need a clean environment. Any one test missing the setup && unset some variables in temp_env will cause the problem I described earlier.

[fabian-braun]

I have a draft here. Is that what you had in mind?

I'm a bit hesitant about it, because this will also unset all of these variables and that may cause surprising effects in tests, for example backtraces not working properly within the tests because RUST_BACKTRACE is unset or macros that rely on CARGO_PKG_NAME will suddenly not produce the expected result.

[tabokie]

Is that what you had in mind?

Yes it is, appreciate the work!

may cause surprising effects in tests

True, I didn't think about this before. It isn't just the behavior inside temp_env is altered, but all other unrelated tests running at the same time.

I think a better description of what temp_env should provide, is the isolation against all other tests running temp_env. Currently it means we can read variables unmodified by them. But we can't write variables (after the construction of the env).

Say we provide a method called temp_env::set_in_scope(key, value), the desired functionality can be implemented like this:

temp_env::new(
    || {
        for (key, var) in std::env::vars() {
          temp_env::set_in_scope(key, None);
        }
    },
);