fix: logWithExit_ should exit on Just

[elldritch]

Overview

Fixes a bug surfaced by d4f6bef. See the unexpected test failure at https://github.com/fossas/spectrometer/pull/277/checks?check_run_id=3000639861.

Previously, programs that ran a successful DiagnosticsC action under logWithExit_ would (incorrectly) continue evaluation. These are not the intended semantics. Instead, a successful run should exit successfully, much like how an unsuccessful run exits with failure.

This bug affects commands with steps after logWithExit_ e.g. fossa test always exits non-zero even on success.

Acceptance criteria

Actions run with logWithExit_ should exit zero on success.

Testing plan

Run a fossa test on a revision with no issues (I'm using the latest Spectrometer revision). See that it succeeds. Check that the exit code is zero.

cabal run fossa -- test --project https://github.com/fossas/spectrometer --revision 4f69104e2dcdb557a870c6621639750605261ca4 && echo OK

Risks

This significantly changes the semantics of logWithExit_. I believe these were the original intended semantics given that this is only ever used as a wrapper for commands (see grep logWithExit_), but other code that uses this will now abort surprisingly.

Checklist

• I added tests for this PR's change (or confirmed tests are not viable).
• I linted and formatted (via haskell-language-server) any haskell files I touched in this PR.
• If this PR introduced a user-visible change, I added documentation into docs/.
• I updated Changelog.md if this change is externally facing. If this PR did not mark a release, I added my changes into an # Unreleased section at the top.
• I linked this PR to any referenced GitHub issues, if they exist.

[elldritch]

@cnr what's a good way to test this?

Is there a convenient way to do a single file with a small testMain, where I can check that the program fails to reach a line that causes it to exit non-zero?

[cnr]

@liftM because exitSuccess/exitFailure really just throw IO Exceptions (throwIO ExitSuccess), you can test it with the hspec builtin shouldThrow:

  describe "foo" $ do
    it "should do the thing" $ do
      myAction `shouldThrow` (== ExitSuccess)

[cnr]

LGTM - did a cursory skim of all of the places we use logWithExit_ and didn't notice anything problematic. we only ever use it for program entrypoints

[elldritch]

Given that tests passed for 6dd5045, and that the integration test failure is due to a bad release (we should really build the integration test CLI from source, and have a separate test for the installer), I'm going to force-merge this into the mainline to get the release out.