⚡ Faster decomposeFloat/Double

[zbjornson]

This replaces the (creative) loop-based implementation with a constant-time one. The number 1.234 is about 1100x faster and 1e30 is about 2300x faster.

If you're okay with this change I can do the same thing for double.

Category:

• ✨ Introduce new features
• 📝 Add or update documentation
• ✅ Add or update tests
• 🐛 Fix a bug
• 🏷️ Add or update types
• ⚡️ Improve performance
• Other(s): ...

Potential impacts:

• Generated values
• Shrink values
• Performance - faster
• Typings
• Other(s): ...

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

Latest deployment of this branch, based on commit 1c798e1:

Sandbox	Source
Vanilla	Configuration
@fast-check/examples	Configuration

[dubzzz]

Looks awesome. I'll spend some time checking the diff but I think we can go with a similar optim for doubles

[zbjornson]

Thanks! I'm happy to work on the double variant also.

It looks like I broke something. How do you get unit tests to run locally? If I run yarn workspace fast-check test, I get a ton of TS errors.

[dubzzz]

Perfect! It even works with node 8 which is the currently minimal supported version for fast-check (will be dropped soon, but so far v3 is supposed to support it).

[dubzzz]

You may try:

yarn # install all deps
yarn build:all # build all packages

# usually I changedir into /packages/fast-check
# but workspace commands also work
yarn test
yarn e2e # runs against prod bundle so need `yarn build` to be called before if you change things

You'll also need to declare the version bump (minor for fast-check and none for others):

yarn version:bump

I'd need to update my contributing guide with this setup

[zbjornson]

Thanks for the guidance there. Fixed the failing test, switched from DataView to TypedArray for slightly better perf, ran version:bump, and added another commit for decomposeDouble.

[codecov]

Codecov Report

Merging #4059 (f715047) into main (f1a86ce) will increase coverage by 0.03%.
The diff coverage is 100.00%.

❗ Current head f715047 differs from pull request most recent head 1c798e1. Consider uploading reports for the commit 1c798e1 to get more accurate results

@@            Coverage Diff             @@
##             main    #4059      +/-   ##
==========================================
+ Coverage   94.60%   94.64%   +0.03%     
==========================================
  Files         211      211              
  Lines        5893     5897       +4     
  Branches     1351     1353       +2     
==========================================
+ Hits         5575     5581       +6     
+ Misses        302      300       -2     
  Partials       16       16              

Flag	Coverage Δ
unit-tests	94.64% <100.00%> (+0.03%)	⬆️
unit-tests-14.x-Linux	94.64% <100.00%> (+0.03%)	⬆️
unit-tests-16.x-Linux	94.64% <100.00%> (+0.03%)	⬆️
unit-tests-18.x-Linux	94.64% <100.00%> (+0.03%)	⬆️
unit-tests-latest-Linux	94.64% <100.00%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
.../src/arbitrary/_internals/helpers/DoubleHelpers.ts	100.00% <100.00%> (+1.63%)	⬆️
...k/src/arbitrary/_internals/helpers/FloatHelpers.ts	100.00% <100.00%> (+2.12%)	⬆️
packages/fast-check/src/arbitrary/double.ts	100.00% <100.00%> (ø)
packages/fast-check/src/arbitrary/float.ts	100.00% <100.00%> (ø)

[dubzzz]

It seems that format is invalid. Can you please run yarn format on your PR?

[dubzzz]

Thank you so much for this awesome contribution 🥰

[zbjornson]

My pleasure, thanks for the fast review.