Feature: allow buck2 test to test pure Starlark logic

[thoughtpolice]

I just had to implement this function in Starlark:

# starlark has no 'rjust', and 'format' does not support padding with
# zeros, so we have to do this manually.
def rjust(s: str, pad: str, width: int) -> str:
    if len(s) >= width:
        return s
    return pad * (width - len(s)) + s

What I'd like to do is write a test for this in terms of buck2 test, but I don't really think I can. I don't see why it shouldn't be possible, though?

For example you could imagine a BUILD file that does something like this:

test_starlark_assert(
   name = 'test-example',
   expected = '0003',
   actual = rjust('3', '0', 4),
)

test_starlark_assert could then just be a rule that returns some provider containing a pass/fail status, and the logic itself can actually be in the rule (this would allow you to have things like expected failures and write them on your own). Because these tests are actually pure you could have a family of providers you return:

def _test_starlark_assert_impl(...):
    # ...
    if expected  == actual:
        pass_fail_info = PureTestPassResult('passed')
    else:
        pass_fail_info = PureTestFailureResult(f'expected value {expected} but got value {actual}')
    return [ DefaultInfo(), pass_fail_info ]

I understand the TestInfo v2 thing won't be done for a while now, but this seems like a rather independent addition to the API.

[ndmitchell]

I just do:

def _test():
    assert_eq(rjust('3', '0', 4), '0003')

_test()

At the bottom of the file. Means these tests run in every single run, but they are typically millisecond duration, and if they fail, the consequences are often serious. As an example,

buck2/prelude/utils/directory_fold.bzl

Line 45 in f68a1d2

def _test_tree_functions():