Python default string encoding

Question

When, where and how does Python implicitly apply encodings to strings or does implicit transcodings (conversions)?

And what are those "default" (i.e., implied) encodings?

For example, what are the encodings:

• of string literals?

    s = "Byte string with national characters"
    us = u"Unicode string with national characters"
  

• of byte strings when type-converted to and from Unicode?

    data = unicode(random_byte_string)
  

• when byte- and Unicode strings are written to/from a file or a terminal?

    print(open("The full text of War and Peace.txt").read())
  

Answer 1

There are multiple parts of Python's functionality involved here: reading the source code and parsing the string literals, transcoding, and printing. Each has its own conventions.

Short answer:

• For the purpose of code parsing:
  • str (Py2) -- not applicable, raw bytes from the file are taken
  • unicode (Py2)/str (Py3) -- "source encoding", defaults are ascii (Py2) and utf-8 (Py3)
  • bytes (Py3) -- none, non-ASCII characters are prohibited in the literal
• For the purpose of transcoding:
  • both (Py2) -- sys.getdefaultencoding() (ascii almost always)
    • there are implicit conversions which often result in a UnicodeDecodeError/UnicodeEncodeError
  • both (Py3) -- none, must specify encoding explicitly when converting
• For the purpose of I/O:
  • unicode (Py2) -- <file>.encoding if set, otherwise sys.getdefaultencoding()
  • str (Py2) -- not applicable, raw bytes are written
  • str (Py3) -- <file>.encoding, always set and defaults to locale.getpreferredencoding()
  • bytes (Py3) -- none, printing produces its repr() instead

---

First of all, some terminology clarification so that you understand the rest correctly. Decoding is translation from bytes to characters (Unicode or otherwise), and encoding (as a process) is the reverse. See The Absolute Minimum Every Software Developer Absolutely, Positively Must Know About Unicode and Character Sets (No Excuses!) – Joel on Software to get the distinction.

Now...

Reading the source and parsing string literals

At the start of a source file, you can specify the file's "source encoding" (its exact effect is described later). If not specified, the default is ascii for Python 2 and utf-8 for Python 3. A UTF-8 BOM has the same effect as a utf-8 encoding declaration.

Python 2

Python 2 reads the source as raw bytes. It only uses the "source encoding" to parse a Unicode literal when it sees one. (It's more complicated than that under the hood, but this is the net effect.)

> type t.py
# Encoding: cp1251
s = "абвгд"
us = u"абвгд"
print repr(s), repr(us)
> py -2 t.py
'\xe0\xe1\xe2\xe3\xe4' u'\u0430\u0431\u0432\u0433\u0434'

<change encoding declaration in the file to cp866, do not change the contents>
> py -2 t.py
'\xe0\xe1\xe2\xe3\xe4' u'\u0440\u0441\u0442\u0443\u0444'

<transcode the file to utf-8, update declaration or replace with BOM>
> py -2 t.py
'\xd0\xb0\xd0\xb1\xd0\xb2\xd0\xb3\xd0\xb4' u'\u0430\u0431\u0432\u0433\u0434'

So, regular strings will contain the exact bytes that are in the file. And Unicode strings will contain the result of decoding the file's bytes with the "source encoding".

If the decoding fails, you will get a SyntaxError. Same if there is a non-ASCII character in the file when there's no encoding specified. Finally, if unicode_literals future is used, any regular string literals (in that file only) are treated as Unicode literals when parsing, with all what that means.

Python 3

Python 3 decodes the entire source file with the "source encoding" into a sequence of Unicode characters. Any parsing is done after that. (In particular, this makes it possible to have Unicode in identifiers.) Since all string literals are now Unicode, no additional transcoding is needed. In byte literals, non-ASCII characters are prohibited (such bytes must be specified with escape sequences), evading the issue altogether.

Transcoding

As per the clarification at the start:

• str (Py2)/bytes (Py3) -- bytes => can only be decoded _{(directly, that is; details follow)}
• unicode (Py2)/str (Py3) -- characters => can only be encoded

Python 2

In both cases, if the encoding is not specified, sys.getdefaultencoding() is used. It is ascii _{(unless you uncomment a code chunk in site.py, or do some other hacks which are a recipe for disaster)}. So, for the purpose of transcoding, sys.getdefaultencoding() is the "string's default encoding".

Now, here's a caveat:

• a decode() and encode() -- with the default encoding -- is done implicitly when converting str<->unicode:

  • in string formatting (a third of UnicodeDecodeError/UnicodeEncodeError questions on Stack Overflow are about this)
  • when trying to encode() a str or decode() a unicode (the second third of the Stack Overflow questions)

Python 3

There's no "default encoding" at all: implicit conversion between str and bytes is now prohibited.

• bytes can only be decoded and str -- encoded, and the encoding argument is mandatory.
• converting bytes->str (incl. implicitly) produces its repr() instead (which is only useful for debug printing), evading the encoding issue entirely
• converting str->bytes is prohibited

Printing

This matter is unrelated to a variable's value but related to what you would see on the screen when it's printed -- and whether you will get a UnicodeEncodeError when printing.

Python 2

• A unicode is encoded with <file>.encoding if set; otherwise, it's implicitly converted to str as per the above. (The final third of the UnicodeEncodeError SO questions fall into here.)
  • For standard streams, the stream's encoding is guessed at startup from various environment-specific sources, and can be overridden with the PYTHONIOENCODING environment variable.
• str's bytes are sent to the OS stream as-is. What specific glyphs you will see on the screen depends on your terminal's encoding settings (if it's something like UTF-8, you may see nothing at all if you print a byte sequence that is invalid UTF-8).

Python 3

The changes are:

• Now files opened with text vs. binary mode natively accept str or bytes, correspondingly, and outright refuse to process the wrong type. Text-mode files always have an encoding set, locale.getpreferredencoding(False) being the default.
• print for text streams still implicitly converts everything to str, which in the case of bytes prints its repr() as per the above, evading the encoding issue altogether

Answer 2

Implicit encoding as internal format to store strings/arrays: you should not care about the encoding. In fact, Python decodes characters in a Python internal way. It is mostly transparent. Just image that it is a Unicode text, or a sequence of bytes, in an abstract way.

The internal coding in Python 3.x varies according the "larger" character. It could be UTF-8/ASCII (for ASCII strings), UTF-16 or UTF-32. When you are using strings, it is like you have a Unicode string (so abstract, not a real encoding). If you do not program in C or you use some special functions (memory view), you will never be able to see the internal encoding.

Bytes are just a view of actual memory. Python interprets is as unsigned char. But again, often you should just think about what the sequence it is, not on internal encoding.

Python 2 has bytes and string as unsigned char, and Unicode as UCS-2 (so code points above 65535 will be coded with two characters (UCS-2) in Python 2, and just one character (UTF-32) in Python 3).