Skip to content

float64.md

Latest commit

 

History

History
30 lines (21 loc) · 1.47 KB

float64.md

File metadata and controls

30 lines (21 loc) · 1.47 KB

Float64

Every serialization format we surveyed supports double-precision floating-point format IEEE-754.

For this reason we support Float64 as one of the Kinds and Token values.

data Token =
  ...
  | Float64 Float64 -- '.' (IEEE-754)
  ...

We also found that all programming languages we surveyed supports IEEE-754 floats. The one exception is Erlang, which supports IEEE-754, except for NaN and Inf.

Comparing NaN

IEEE-754 specifies a double-precision floating point number. One of the floating point numbers supported is NaN (not a number). The specification also says that NaN != NaN. This is important when calculating different NaNs, but this part of the spec is annoying when testing serializing and deserializing a NaN. There is no way to test that the serialized NaN is equal to the deserialized NaN, without creating an exception to this rule. For testing and validation purposes we recommend creating an exception to this rule where NaN = NaN. When testing serialization and deserialization, please remember to not use default equality for Float64, rather use an equality that includes the exception and rather assumes that NaN = NaN.

References