convert binary floating and doubles in IEEE-754 format.

Questions / Help
1

Hi. I’m new to elixir, so this could be obvious. I’m reading and writing to a device that uses a binary line format. The format is fixed. It also has integer fields, but they are easy to interpret. Floats and doubles are a bit harder, so I’m wondering and hoping that this is a problem that’s already been solved.

Here’s an example of what I’m looking at:

SSSSSSSSSSSSIIIIIIIIFFFFDDDDDDDD

Where S…S is a string descriptor.
IIII is an integer that I unpack with ((i1 <<< 32) + (i2 <<< 16) + (i3 <<< 8) + i4)

The ones I’m struggling with are FFFF (32 bit float) and DDDDDDDD (double prec float).

I know I can look at the IEEE-754 spec and figure this out, but I’m thinking there might be a solution already. For example in python given the FFFF, I can use struct.unpack(FFFF) to get the floating point value.

2

This should totally do it for you, unless you have to deal with NaN and infinties…

iex(1)> bin = <<1.22::float>>         
<<63, 243, 133, 30, 184, 81, 235, 133>>
iex(2)> <<x :: float>> = bin
<<63, 243, 133, 30, 184, 81, 235, 133>>
iex(3)> x
1.22
iex(4)> bin = <<1.22::float-size(32)>>
<<63, 156, 40, 246>>
2 Likes
3

Thanks. That’s just what I was looking for. NaN and infinity should be caught before they get on the line, so I think we’re good.

4

Just remember that binary pattern matching also supports integers, so instead of manual addition and shifts you can do:

<<describe::binary-size(12), i1::little-integer-32, i2::little-integer-32, f1::float-32, f2::float-64>> = input

That is why I love working with binary data in Erlang. It is just so enormously easy.

5 Likes
5

Thank you very much. That’s very helpful.

6

probably not an issue, but if things go south don’t forget to check the endianness (there are extra decorators for that if you need it).

1 Like
7

I figured out I needed big-integer instead of little. Is there a good list of all the decorators? I looked briefly and didn’t find one.

8

It is in docs for <<>>/1.

2 Likes