fix(forge): Don't ignore config.toml when running invariant tests for coverage

[0xmp]

forge coverage ignores the config.toml file for invariant tests.

Invariant testing in coverage mode thus uses the default 256 runs which is small, making coverage for big invariant tests noisy/useless. This is not the case for regular (non-invariant) tests.

Motivation

This would bring us closer to having invariant tests as contributing fully to coverage (#4007).
Fixes #6565.

Solution

The coverage runner now takes the correct config given when running tests.

This is my first time writing Rust ever, so do feel free to ignore/rewrite completely.

[Evalir]

Hey! thanks for this!

So, I don't see an issue with the code itself—however, invariants right now don't contribute to coverage at all, so this would only really make coverage slower for users in the meanwhile :s. I don't think we wanna get this in before having a proper plan on how to revamp coverage and make invariant testing work with it.

[0xmp]

invariants right now don't contribute to coverage at all

That's weird, I'm running an invariant test and getting some coverage info following this Twitter thread by using

forge coverage --report lcov --report-file test/myfile.lcov --match-contract myInvariantTest
genhtml ./test/myfile.lcov -o output

and then checking the output html.

The coverage does seem to increase on my test when I increase the number of runs. Maybe I'm getting "fooled by randomness" and it's just a question of seed or something?
Edit: I am using a Handler if that changes anything

[Evalir]

Interesting thread—thanks for sharing. However we do not officially support coverage for invariant tests 😅 so while this might actually be working I'm really wary about this affecting CI to people. Wdyt @mds1 ?

[mds1]

Not sure if this is a common opinion, but:

1. You should have good coverage even with 1 fuzz run and no invariant runs. This gives a stricter measure of your test coverage by ensuring you aren't relying on fuzzing/randomness.
2. It is still useful to be able to see how much coverage many fuzz runs or many invariant runs provided, to gauge the effectiveness of your fuzz and invariant tests.

As a result, my preference is two coverage modes:

1. One mode ignores config settings, sets fuzz runs to 1, ignores invariant tests.
2. Second mode respects config settings,

I think people's expectation is that (2) is the default, and perhaps (1) becomes something like forge coverage --strict

I'm not sure where this PR fits into that potential design/UX

[Evalir]

let's go with this for now—we'll respect the invariant settings, and we can add the --strict mode later.