Performance Improvements for files with thousands of tests

[jamestalmage]

Using the emoji-aware test-per-character branch as a benchmark, and running time ava:

master:

  4363 passed

ava  22.61s user 0.54s system 109% cpu 21.093 total

this branch (only the Runner commits):

  4363 passed

ava  7.30s user 0.51s system 139% cpu 5.597 total

this branch (all the commits):

  4363 passed

ava  6.32s user 0.46s system 145% cpu 4.675 total

---

updated benchmark (with empower-core performance tweaks):

ava --verbose  5.05s user 0.32s system 138% cpu 3.872 total

[jamestalmage]

This relies on some changes to empower-core.

twada/empower-core#7

[jamestalmage]

Further splitting that test suite into multiple files (beaugunderson/emoji-aware#1):

  4363 passed

ava --verbose  9.28s user 0.65s system 458% cpu 2.164 total

[sindresorhus]

?

[jamestalmage]

With this change, destructuring would no longer work. This is not allowed anymore:

test({is, same} => {
  is(1 + 1, 2);
  same([1,2].slice(1), [2]);
});

You have to preserve the this reference:

test(t => {
  t.is(1 + 1, 2);
  t.same([1,2].slice(1), [2]);
});

[sindresorhus]

👍 Looks good! (ᵔᴥᵔ)

Too bad jamestalmage@4e23943 was slow. It was more readable.

[jamestalmage]

Too bad jamestalmage/ava@4e23943 was slow. It was more readable.

Yeah, unfortunately it is a big offender in this scenario.

[jamestalmage]

Also a technically breaking change (though not one I am really concerned about):

t.is(1 + 1, 2);

No longer returns a Promise.

Only t.throws and t.doesNotThrow will return a promise.

[sindresorhus]

though not one I am really concerned about

Me neither.

[vadimdemedes]

Results look promising, let's get it in!

[sindresorhus]

(つ◕౪◕)つ━☆ﾟ.*･｡ﾟ

[kentcdodds]

[jamestalmage]

@kentcdodds, et. all

Just for comparison - I also created a branch converting emoji-aware to mocha

AVA now ekes out slightly better performance:

ava --verbose  9.76s user   0.71s system    465% cpu    2.252 total
mocha          2.07s user   0.31s system     96% cpu    2.460 total   

However, it's obvious the only reason AVA is winning is that it is leveraging multiple processes. Hopefully this means there are still lots of opportunities to make AVA even faster.

[kentcdodds]

Totally. If we could make ava as fast as mocha on a per-file basis then ava would scream with speed. This is making me feel really positive about the potential for ava :D