[gfx]

If the double precision is not required, float32 is more compact than float64.

It is useful if the precisions of numbers do not matter.

[codecov-io]

Codecov Report

Merging #79 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #79      +/-   ##
==========================================
+ Coverage   98.15%   98.16%   +0.01%     
==========================================
  Files          15       15              
  Lines         919      927       +8     
  Branches      187      189       +2     
==========================================
+ Hits          902      910       +8     
  Misses         17       17

Impacted Files	Coverage Δ
src/Encoder.ts	97.9% <100%> (+0.07%)	⬆️
src/encode.ts	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c4a1ef5...91eeb98. Read the comment docs.

[michaelficarra]

This is more of "force" than "prefer". Why not just test whether the value can be losslessly represented as a float32 and use that, otherwise use float64?

[gfx]

@michaelficarra Thanks for your advice. s/prefer/force/ at 91eeb98.

Why not just test whether the value can be losslessly represented as a float32 and use that, otherwise use float64?

I think it should make encoding much slower than the forceFloat32 option because the floating point number is a complex data type that requires much more bit operations.