How to get endianness of numpy dtype

Question

Related to Determine the endianness of a numpy array

Given an array

x = np.arange(3)

I can get the byte order by doing

>>> x.dtype.byteorder
'='

How do I find out if this is big or little endian? I would like to get '<', '>' or '|' as the output, not '='.

To be clear, I am not hung up on what format the information comes in. I just want to know "big endian", "little endian" or "irrelevant", but I don't care if it's "native" or not.

I might be missing your question here, but just check the native byteorder with `sys.byteorder`? — miradulo, Apr 09 '18 at 19:45
@miradulo. I think you are understanding perfectly. I am very surprised that there is no way to get that from numpy directly, given that numpy has to access that information at some point, presumably. — Mad Physicist, Apr 09 '18 at 19:49

user2357112 · Accepted Answer · 2018-04-09T19:51:05.543

14

Probably just check sys.byteorder. Even the numpy.dtype.byteorder examples in the docs use sys.byteorder to determine what's native.

endianness_map = {
    '>': 'big',
    '<': 'little',
    '=': sys.byteorder,
    '|': 'not applicable',
}

edited Apr 09 '18 at 19:51

answered Apr 09 '18 at 19:49

user2357112

260,549
28
431
505

In that case, I don't have much choice unfortunately. – Mad Physicist Apr 09 '18 at 19:50
1

I was thinking of something more like `nativecode = '>' if sys.byteorder == 'big' else '<'; x.dtype.byteorder.replace('=', nativecode)` – Mad Physicist Apr 09 '18 at 19:53

score 2 · Answer 2 · answered Apr 09 '18 at 19:53

2

You can swap the endianness twice to make numpy reveal the true endianness:

dtype_nonnative = dtype.newbyteorder('S').newbyteorder('S')
dtype_nonnative.byteorder

answered Apr 09 '18 at 19:53

Eric

95,302
53
242
374

2

I think I understand why that happens under the hood, but seriously, whyyyyyyyyy? – Mad Physicist Apr 09 '18 at 19:54
How about a PR to add a property to `dtype` to always get `<`, `>` or `|`? – Mad Physicist Apr 09 '18 at 19:55
That's a surprise, considering that even explicitly constructing `dtype(' – user2357112 Apr 09 '18 at 19:55
2

@user2357112 A bug in which direction? Explicitly constructed dtypes should product explicit byteorders, or `newbyteorder` should toggle between `=` and one of `{<, >}`? – miradulo Apr 09 '18 at 20:09
1

@miradulo: I'd expect `newbyteorder` to produce `'='` when the resulting byte order is the same as the native byte order. (The fact that explicit `dtype(' – user2357112 Apr 09 '18 at 20:47
@user2357112 Seems sensible, I didn't notice the comment in the docs there. Thanks! – miradulo Apr 09 '18 at 21:09
Another fun fact: `x.newbyteorder().newbyteorder()` does not do the same thing, where `x` is the array, but it does when it is the `dtype`. – Mad Physicist Apr 09 '18 at 21:42

How to get endianness of numpy dtype

2 Answers2