chore: Switch to using Node built-in test runner

[jsumners-nr]

This PR resolves #2379. It must target the next branch because Node.js 16 does not include the test runner.

[jsumners-nr]

I don't like that we have to rely on shell expansion of the globs in all of these scripts. But we don't run CI on Windows, and in-process expansion of the globs isn't supported until Node 22. So it is what it is.

[bizob2828]

I have notice the default reporter spec could pose problems in CI. It does not appear there is an equivalent to tap's classic reporter. If we change to dot reporter it will reduce I/O and seems to speed up the test runs as well. This may have to be a follow up because --test-reporter doesn't exist until 18

I did suggest some changes as I noticed the globs didn't account for tests in the direct child of a given test type. However, it also appears my proposed changes to include all tests by chaining globs segfaults in node 16, and does not work in 18 and 20.

I did compare the coverage is still the same, as expected because we're still relying on v8 coverage and c8 to report it.

[jsumners-nr]

I have notice the default reporter spec could pose problems in CI.

What problems might those be?

If we change to dot reporter it will reduce I/O and seems to speed up the test runs as well.

I don't particularly like the dot reporter because it is very cryptic. But putting the speed theory to the test:

# Unit test run without specifying a reporter:
ℹ tests 176
ℹ suites 0
ℹ pass 176
ℹ fail 0
ℹ cancelled 0
ℹ skipped 0
ℹ todo 0
ℹ duration_ms 31175.115292

real    0m34.861s
user    1m25.072s
sys     0m31.439s

# With the dot reporter:
................................................................................................................................................................................

real    0m29.611s
user    1m25.536s
sys     0m32.046s

I've done the test multiple times and either reporter is roughly the same amount of time.

One key difference is in error reporting. This is what the dot reporter does if there is a test failure:

❯ npm run unit

> [email protected] unit
> rm -f newrelic_agent.log && time c8 -o ./coverage/unit node --test-reporter dot --test test/unit/*.test.js test/unit/**/*.test.js

.......................................X........................................................................................................................................

real    0m28.944s
user    1m25.835s
sys     0m31.681s

I think that is really useless. I don't enjoy having to search all of the output of the standard reporter, but at least we can see which test(s) failed.

[jsumners-nr]

I think that is really useless. I don't enjoy having to search all of the output of the standard reporter, but at least we can see which test(s) failed.

🤔 we could report to stdout with the dot reporter but report to something like unit.txt via stderr with the spec reporter.

We could also write a simple reporter that outputs test failures when they occur, and nothing on success. In other words, mimic our versioned tests output. Example:

❯ npm run unit

> [email protected] unit
> rm -f newrelic_agent.log && time c8 -o ./coverage/unit node --test-reporter ./reporter.js --test test/unit/*.test.js test/unit/**/*.test.js

failed: /Users/jsumners/Projects/team-repos/node-newrelic/test/unit/attributes.test.js
real    0m26.942s
user    1m23.630s
sys     0m31.549s

I think we could write a much better reporter that shows offending line numbers and such, but the runner in v18 seems to omit a lot of that in my test case (just reporting line = 1; column = 1 for the induced failure when it should at least be line 35). But I think that report is enough to give us more to go on than the dot reporter and doesn't spam stdout with difficult to sift through text.

[bizob2828]

the update to globs still isn't running all the tests. i found in node 22 if you wrap each glob in single quotes it works. but that fails in 18 and 20

[bizob2828]

overall looks great. i do have some failing tests that i suspect are because of time outs though

[bizob2828]

i think a timeout has to be set. the facts.test.js fails

[bizob2828]

i think we need to restore the timeout. i have 1 test that fails

[bizob2828]

very nice. we can always add to this.