struct on WiPy gives different output than on PC

[alexpul]

I'm sending serial messages that consist of float values encoded using struct between my WiPy and PC (python 2.7). When I encode a decimal value, like 3.4, on the WiPy and then send it to the PC and decode it I get 3.4000000953674316. I noticed that the output of struct.pack('<d',3.4) is slightly different on the WiPy and on the PC.

Code on WiPy:

x_wipy= struct.pack('d', 3.4)      #encode value
print(binascii.hexlify(x_wipy))    #display encoding
>>> b'0000004033330b40'

Code on PC (python 2.7):

x_pc = struct.pack('d', 3.4)     #encode value
print(binascii.hexlify(x_pc))    #display encoding
>>> 3333333333330b40

Does anyone know why they differ? Could it be different versions of python?

Other observations:

• WiPy decodes both x_wipy and x_pc correctly to 3.4
• PC decodes x_pc to 3.4 and x_wipy to 3.4000000953674316
• WiPy returns bytes, PC returns str -> this is due to different python versions

[robert-hh]

@alexpul Indeed, but both pack to the same bytes object.
PC:

>>> struct.pack("f", 3.4)
b'\x9a\x99Y@'
>>> struct.unpack("f", b'\x9a\x99Y@')
(3.4000000953674316,)
>>> 

Pycom:

>>> struct.pack("f", 3.4)
b'\x9a\x99Y@'

[alexpul]

Thanks for the great reply @robert-hh!

For any other readers that may be having similar problems: I'm sending the encoded data from the wipy to my PC and I found that even if I encoded it using a single precision float ('f') the value still showed in python as a 64 bit number (hence the really long decimal value). I found that if I forced python to see the number as 32 bit using numpy.float32(x_wipy) then the decimal was correctly represented.

[robert-hh]

@alexpul Micropython does supports only 32 bit floats with, giving a ~7 digit precision, and 3.4 being not a binary fraction cannot be encoded w/o precision loss into a float. That is different for numbers, which can be represented as binary fraction, like 3.5. PC Python uses 64 bit floats.
So what you see are rounding errors in different occurences.