Best way to convert string to bytes in Python 3?

Question

TypeError: 'str' does not support the buffer interface suggests two possible methods to convert a string to bytes:

b = bytes(mystring, 'utf-8')

b = mystring.encode('utf-8')

Which method is more Pythonic?

Solution 1

If you look at the docs for bytes, it points you to bytearray:

bytearray([source[, encoding[, errors]]])

Return a new array of bytes. The bytearray type is a mutable sequence of integers in the range 0 <= x < 256. It has most of the usual methods of mutable sequences, described in Mutable Sequence Types, as well as most methods that the bytes type has, see Bytes and Byte Array Methods.

The optional source parameter can be used to initialize the array in a few different ways:

If it is a string, you must also give the encoding (and optionally, errors) parameters; bytearray() then converts the string to bytes using str.encode().

If it is an integer, the array will have that size and will be initialized with null bytes.

If it is an object conforming to the buffer interface, a read-only buffer of the object will be used to initialize the bytes array.

If it is an iterable, it must be an iterable of integers in the range 0 <= x < 256, which are used as the initial contents of the array.

Without an argument, an array of size 0 is created.

So bytes can do much more than just encode a string. It's Pythonic that it would allow you to call the constructor with any type of source parameter that makes sense.

For encoding a string, I think that some_string.encode(encoding) is more Pythonic than using the constructor, because it is the most self documenting -- "take this string and encode it with this encoding" is clearer than bytes(some_string, encoding) -- there is no explicit verb when you use the constructor.

I checked the Python source. If you pass a unicode string to bytes using CPython, it calls PyUnicode_AsEncodedString, which is the implementation of encode; so you're just skipping a level of indirection if you call encode yourself.

Also, see Serdalis' comment -- unicode_string.encode(encoding) is also more Pythonic because its inverse is byte_string.decode(encoding) and symmetry is nice.

Solution 2

It's easier than it is thought:

my_str = "hello world"
my_str_as_bytes = str.encode(my_str)
print(type(my_str_as_bytes)) # ensure it is byte representation
my_decoded_str = my_str_as_bytes.decode()
print(type(my_decoded_str)) # ensure it is string representation

you can verify by printing the types. Refer to output below.

<class 'bytes'>
<class 'str'>

Solution 3

The absolutely best way is neither of the 2, but the 3rd. The first parameter to encode defaults to 'utf-8' ever since Python 3.0. Thus the best way is

b = mystring.encode()

This will also be faster, because the default argument results not in the string "utf-8" in the C code, but NULL, which is much faster to check!

Here be some timings:

In [1]: %timeit -r 10 'abc'.encode('utf-8')
The slowest run took 38.07 times longer than the fastest. 
This could mean that an intermediate result is being cached.
10000000 loops, best of 10: 183 ns per loop

In [2]: %timeit -r 10 'abc'.encode()
The slowest run took 27.34 times longer than the fastest. 
This could mean that an intermediate result is being cached.
10000000 loops, best of 10: 137 ns per loop

Despite the warning the times were very stable after repeated runs - the deviation was just ~2 per cent.

---

Using encode() without an argument is not Python 2 compatible, as in Python 2 the default character encoding is ASCII.

>>> 'äöä'.encode()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 0: ordinal not in range(128)

Solution 4

Answer for a slightly different problem:

You have a sequence of raw unicode that was saved into a str variable:

s_str: str = "\x00\x01\x00\xc0\x01\x00\x00\x00\x04"

You need to be able to get the byte literal of that unicode (for struct.unpack(), etc.)

s_bytes: bytes = b'\x00\x01\x00\xc0\x01\x00\x00\x00\x04'

Solution:

s_new: bytes = bytes(s, encoding="raw_unicode_escape")

Reference (scroll up for standard encodings):

Python Specific Encodings

View more solutions

Comments

[Lennart Regebro]

Use encode/decode is more common, and perhaps clearer.

[m3nda]

@LennartRegebro I dismiss. Even if it's more common, reading "bytes()" i know what its doing, while encode() don't make me feel it is encoding to bytes.

[Lennart Regebro]

@erm3nda Which is a good reason to use it until it does feel like that, then you are one step closer to Unicode zen.

[m3nda]

@LennartRegebro I feel good enough to just use bytes(item, "utf8"), as explicit is better than implicit, so... str.encode( ) defaults silently to bytes, making you more Unicode-zen but less Explicit-Zen. Also "common" is not a term that i like to follow. Also, bytes(item, "utf8"), is more like the str(), and b"string" notations. My apologies if i am so noob to understand your reasons. Thank you.

[Mark Ransom]

@erm3nda if you read the accepted answer you can see that encode() doesn't call bytes(), it's the other way around. Of course that's not immediately obvious which is why I asked the question.

[m3nda]

Doh, sorry. Anyway, what i said applies too for some_string.encode(encoding), being as example "string".encode("utf8") which returns type bytes. For me, using the term bytes() makes much more sense. I tend to think that encode/decode is more charset related than data type. Again, i may be so much noob to think like that... but i love explicit, and there not "byte" refer into "some".encode("utf8"). Thank you, i've checked that str.encode() just doesnt't default to anyting.

[mtraceur]

@erm3nda Doesn't the very meaning of the word encode in the context of text include "to bytes", because encoding text is the taking of abstract text data and turning it into some actual concrete byte representation?

[mckenzm]

Encode and decode are always preferred as chaining is easier to read than nesting. e.g. ebcdic=passed.decode('utf-8').encode('ibm500')

[gerardw]

The 'utf-8' is the default, so the simplest answer is b = mystring.encode( )

[Serdalis]

+1 for having a good argument and quotes from the python docs. Also unicode_string.encode(encoding) matches nicely with bytearray.decode(encoding) when you want your string back.

[hamstergene]

bytearray is used when you need a mutable object. You don't need it for simple str↔bytes conversions.

[agf]

@EugeneHomyakov This has nothing to do with bytearray except that the docs for bytes don't give details, they just say "this is an immutable version of bytearray" so I have to quote from there.

[agf]

He knows how to do it, he's just asking which way is better. Please re-read the question.

[Mike]

FYI: str.decode(bytes) didn't work for me (Python 3.3.3 said "type object 'str' has no attribute 'decode'") I used bytes.decode() instead

[jfs]

@Mike: use obj.method() syntax instead of cls.method(obj) syntax i.e., use bytestring = unicode_text.encode(encoding) and unicode_text = bytestring.decode(encoding).

[Ted]

Mike and shenshin fixed the errors in the answer -- it is working now for py 3.6

[Vladimir Shebuniayeu]

You should be very carefull because encode create bytes but class will still be str, bytes method create bytes class.

[Uyghur Lives Matter]

This answer looks more like a comment to me. How does this actually answer the question?

[holdenweb]

Just a cautionary note from Python in a Nutshell about bytes: Avoid using the bytes type as a function with an integer argument. In v2 this returns the integer converted to a (byte)string because bytes is an alias for str, while in v3 it returns a bytestring containing the given number of null characters. So, for example, instead of the v3 expression bytes(6), use the equivalent b'\x00'*6, which seamlessly works the same way in each version.

[Antti Haapala -- Слава Україні]

... i.e. you're needlessly making an unbound method, and then calling it passing the self as the first argument

[Kellen Stuart]

@agf who cares? it helps people who come to this page looking to perform this operation

[abarnert]

@KolobCanyon The question already shows the right way to do it—call encode as a bound method on the string. This answer suggests that you should instead call the unbound method and pass it the string. That's the only new information in the answer, and it's wrong.

[abarnert]

There's only a sizable difference here because (a) the string is pure ASCII, meaning the internal storage is already the UTF-8 version, so looking up the codec is almost the only cost involved at all, and (b) the string is tiny, so even if you did have to encode, it wouldn't make much difference. Try it with, say, '\u00012345'*10000. Both take 28.8us on my laptop; the extra 50ns is presumably lost in the rounding error. Of course this is a pretty extreme example—but 'abc' is just as extreme in the opposite direction.

[Antti Haapala -- Слава Україні]

@abarnert true, but even then, there is no reason pass the argument as a string.

[iggy12345]

Just a note, that if you are trying to convert binary data to a string, you'll most likely need to use something like byte_string.decode('latin-1') as utf-8 doesn't cover the entire range 0x00 to 0xFF (0-255), check out the python docs for more info.

[techkuz]

tl;dr would be helpful

[MasterControlProgram]

@agf Even though it may seem that he/she knows (whatsoever), this is the most compact outline here. Thanks!

[hmijail]

According to this, the default arguments are always "absolutely the best way" to do things, right? This kind of speed analysis would feel like a probable exaggeration if this was about discussing C code. In an interpreted language, it leaves me speechless.

[Antti Haapala -- Слава Україні]

@hmijail you win nothing by explicitly typing the default argument values - more keystrokes, larger code and it is slower too.

[Mark Ransom]

The Zen of Python declares that explicit is better than implicit, which means that an explicit 'utf-8' parameter is to be preferred. But you've definitely shown that leaving off the parameter is faster. That makes this a good answer, even if it isn't the best one.

[Kade]

This was actually just what I was looking for. I could not figure out how to better phrase my question. :) Thank you @Brent!

[Orwellophile]

This was the answer I needed, coming from a google search of "python 3 convert str to bytes binary" this was the top result and looked promising. There are more interesting questions -- like how to convert a unicode string into a regular string (python 2.7) :p

[Antti Haapala -- Слава Україні]

@MarkRansom then how many times have you actually used int(s, 10) ;-)

[Mark Ransom]

Never, but the default for int has ALWAYS been 10. Not so for encode.

[CalfCrusher]

@hasanatkazmi you made my day!

[Antti Haapala -- Слава Україні]

@MarkRansom it has always been the default in Python 3 :P And there are no other Pythons.

[Mark Ransom]

Despite Python 2 no longer being supported, I suspect there will be people dealing with some legacy code for a very long time to come; if for no other reason than to upgrade it to the latest version of Python! I'm glad you didn't remove your warning for Python 2 users at the end.

[Ryan Codrai]

The saving of ~50ns is not a good reason to replace self-declarative code with ambiguous code.

[Hubert Grzeskowiak]

Encoding and decoding can mean many different things, but if you tell me you are encoding a string to utf-8, I would think we are talking about a binary representation of the string or some kind of encoding conversion.

[Mark Ransom]

All you're doing is adding another layer on top of what was suggested in the question. I can't see how that's useful at all.

[SunnyPro]

Some examples with output would be helpful.