Float To String in OTP 25

01 Mar 2022 - Thomas Depierre

In OTP 25, you will have access to a new option for float_to_list/2 and float_to_binary/2. This was the result of work i contributed with the support of the EEF. Why did we do that and when should you use it ?


Starting in OTP 25, you should use the short option everytime you want to cast a float to a string, at least by default. Only use other options if you know you need them and what you are doing (you probably don’t). This would look like this

float_to_list(0.3, [short])
float_to_binary(0.3, [short])

This will be more precise, shorter, probably faster and more memory-efficient than any other option available.

The Long Reason

The BEAM natively handle floating point numbers as a binary IEEE 754 64-bits number. This is a well known format for floating points numbers, that have a lot of advantages. I am not going to go over them all today. But sometimes, we need to cast that number to a string. This is usually needed when exporting the number to an external system. JSON, CSV are two widely used format to exchange a high number of numeric data. CSS and SVG regularly need floating points, in a text form. And we do need regularly to cast floats into logs, stacktraces or even in the shell.

It happens that casting a float to a string properly, without loss of precision, quickly and efficiently is still a hard problem and one that eluded computer science for the past few decades. Luckily, someone found a way to do it a few years ago.

In OTP 25, thanks to the support of the Erlang Ecosystem Foundation, I contributed an implementation of that algorithm to OTP, allowing the whole community to benefit from an average 10x speedup in CPU time and average 100x reduction in memory use compared to previous best practices.

The different ways to print a float

  • e
  • f
  • g
  • shortest round trip correctly rounded