unicode

[7stud]

Based on this example and the error:

-----
u_str = u"abc\u9999"
print u_str

UnicodeEncodeError: 'ascii' codec can't encode character u'\u9999' in
position 3: ordinal not in range(128)
------

it looks like when I try to display the string, the ascii decoder
parses each character in the string and fails when it can't convert a
numerical code that is higher than 127 to a character, i.e. the
character \u9999.

In the following example, I use encode() to convert a unicode string
to a regular string:

-----
u_str = u"abc\u9999"
reg_str = u_str.encode("utf-8")
print repr(reg_str)
-----

and the output is:

'abc\xe9\xa6\x99'

1) Why aren't the characters 'a', 'b', and 'c' in hex notation? It
looks like python must be using the ascii decoder to parse the
characters in the string again--with the result being python converts
only the 1 byte numerical codes to characters. 2) Why didn't that
cause an error like above for the 3 byte character?

Then if I try this:

---
u_str = u"abc\u9999"
reg_str = u_str.encode("utf-8")
print reg_str
---

I get the output:

abc<some chinese character>

Here it looks like python isn't using the ascii decoder anymore. 2)
What determines which decoder python uses?

 

[Erik Max Francis]

Based on this example and the error:

-----
u_str = u"abc\u9999"
print u_str

UnicodeEncodeError: 'ascii' codec can't encode character u'\u9999' in
position 3: ordinal not in range(128)
------

it looks like when I try to display the string, the ascii decoder
parses each character in the string and fails when it can't convert a
numerical code that is higher than 127 to a character, i.e. the
character \u9999.

If you try to print a Unicode string, then Python will attempt to first
encode it using the default encoding for that file. Here, it's apparent
the default encoding is 'ascii', so it attempts to encode it into ASCII,
which it can't do, hence the exception. The error is no different from
this:
Traceback (most recent call last):
File "<stdin>", line 1, in ?
UnicodeEncodeError: 'ascii' codec can't encode character u'\u9999' in
position 3: ordinal not in range(128)

In the following example, I use encode() to convert a unicode string
to a regular string:

-----
u_str = u"abc\u9999"
reg_str = u_str.encode("utf-8")
print repr(reg_str)
-----

and the output is:

'abc\xe9\xa6\x99'

1) Why aren't the characters 'a', 'b', and 'c' in hex notation? It
looks like python must be using the ascii decoder to parse the
characters in the string again--with the result being python converts
only the 1 byte numerical codes to characters. 2) Why didn't that
cause an error like above for the 3 byte character?

Since you've already encoded the Unicode object as a normal string,
Python isn't trying to do any implicit encoding. As for why 'abc'
appears in plain text, that's just the way repr works:
'\x99'

repr is attempting to show the string in the most readable fashion. If
the character is printable, then it just shows it as itself. If it's
unprintable, then it shows it in hex string escape notation.

Then if I try this:

---
u_str = u"abc\u9999"
reg_str = u_str.encode("utf-8")
print reg_str
---

I get the output:

abc<some chinese character>

Here it looks like python isn't using the ascii decoder anymore. 2)
What determines which decoder python uses?

Again, that's because by already encoding it as a string, Python isn't
doing any implicit encoding. So it prints the raw string, which happens
to be UTF-8, and which your terminal obviously supports, so you see the
proper character.

 

[Sander Steffann]

Hi,

If you try to print a Unicode string, then Python will attempt to first
encode it using the default encoding for that file. Here, it's apparent
the default encoding is 'ascii', so it attempts to encode it into ASCII,
which it can't do, hence the exception.

If you want to change the default encoding of your stdout and stderr, you
can do something like this:

import codecs, sys
sys.stdout = codecs.getwriter('utf-8')(sys.stdout)
sys.stderr = codecs.getwriter('utf-8')(sys.stderr)

After doing this, print u_str will work as expected (when using an utf-8
terminal)

- Sander

 

[7stud]

If you try to print a Unicode string, then Python will attempt to first
encode it using the default encoding for that file. Here, it's apparent
the default encoding is 'ascii', so it attempts to encode it into ASCII,
which it can't do, hence the exception. The error is no different from
this:

Traceback (most recent call last):
File "<stdin>", line 1, in ?
UnicodeEncodeError: 'ascii' codec can't encode character u'\u9999' in
position 3: ordinal not in range(128)


Since you've already encoded the Unicode object as a normal string,
Python isn't trying to do any implicit encoding. As for why 'abc'
appears in plain text, that's just the way repr works:

'\x99'

repr is attempting to show the string in the most readable fashion. If
the character is printable, then it just shows it as itself. If it's
unprintable, then it shows it in hex string escape notation.


Again, that's because by already encoding it as a string, Python isn't
doing any implicit encoding. So it prints the raw string, which happens
to be UTF-8, and which your terminal obviously supports, so you see the
proper character.

--
Erik Max Francis && (e-mail address removed) && http://www.alcyone.com/max/
San Jose, CA, USA && 37 20 N 121 53 W && AIM, Y!M erikmaxfrancis
Let us not seek the Republican answer or the Democratic answer but
the right answer. -- John F. Kennedy

So let me see if I have this right:

Here is some code:
-----
print "print unicode string:"
#print u"abc\u9999" #error
print repr(u'abc\u9999')
print

print "print regular string containing chars in unicode syntax:"
print 'abc\u9999'
print repr('abc\u9999')
print

print "print regular string containing chars in utf-8 syntax:"
#encode() converts unicode strings to regular strings
print u'abc\u9999'.encode("utf-8")
print repr(u'abc\u9999'.encode("utf-8") )
-----

Here is the output:
-------
print unicode string:
u'abc\u9999'

print regular string containing chars in unicode syntax:
abc\u9999
'abc\\u9999'

print regular string containing chars in utf-8 syntax:
abc<chinese character>
'abc\xe9\xa6\x99'
------




1) If you print a unicode string:

*print implicitly calls str()*

a) str() calls encode(), and encode() tries to convert the unicode
string to a regular string. encode() uses the default encoding, which
is ascii. If encode() can't convert a character, then encode() raises
an exception.

b) repr() calls encode(), but if encode() raises an exception for a
character, repr() catches the exception and skips over the character
leaving the character unchanged.

2) If you print a regular string containing characters in unicode
syntax:

a) str() calls encode(), but if encode() raises an exception for a
character, str() catches the exception and skips over the character
leaving the character unchanged. Same as 1b.

b) repr() similar to a), but repr() then escapes the escapes in the
string.


3) If you print a regular string containing characters in utf-8
syntax:

a) str() outputs the string to your terminal, and if your terminal can
convert the utf-8 numerical codes to characters it does so.

b) repr() blocks your terminal from interpreting the characters by
escaping the escapes in your string. Why don't I see two slashes like
in the output for 2b?

 

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

1) If you print a unicode string:

*print implicitly calls str()*

No. print does nothing if the object is already a string or unicode
object, and calls str() only otherwise.

a) str() calls encode(), and encode() tries to convert the unicode
string to a regular string. encode() uses the default encoding, which
is ascii. If encode() can't convert a character, then encode() raises
an exception.

Yes and no. This is what str() does, but str() isn't called. Instead,
print inspects sys.stdout.encoding, and uses that encoding to encode
the string. That, in turn, may raise an exception (in particular if
sys.stdout.encoding is "ascii" or not set).

b) repr() calls encode(), but if encode() raises an exception for a
character, repr() catches the exception and skips over the character
leaving the character unchanged.

No. repr() never calls encode. Instead, each type, including unicode,
may have its own __repr__ which is called. unicode.__repr__ escapes
all non-ASCII characters.

2) If you print a regular string containing characters in unicode
syntax:

No. There is no such thing:

py> len("\u")
2
py> "\u"[0]
'\\'
py> "\u"[1]
'u'

In a regular string, \u has no meaning, so \ stands just for itself.

a) str() calls encode(), but if encode() raises an exception for a
character, str() catches the exception and skips over the character
leaving the character unchanged. Same as 1b.

No. Printing a string never invokes .encode(), and no exception occurs
at all. Instead, the \ just gets printed as is.

b) repr() similar to a), but repr() then escapes the escapes in the
string.

str.__repr__ escapes the backslash just in case, so that it won't have
to check for the next character; in that sense, it generates a normal
form.

3) If you print a regular string containing characters in utf-8
syntax:

a) str() outputs the string to your terminal, and if your terminal can
convert the utf-8 numerical codes to characters it does so.

Correct. In general, you should always use the terminal's encoding
when printing to the terminal. That way, you can print everything
just fine what the terminal can display, and get an exception if
you try to print something that the terminal would be unable to
display.

b) repr() blocks your terminal from interpreting the characters by
escaping the escapes in your string. Why don't I see two slashes like
in the output for 2b?

str.__repr__ produces an output that is legal Python syntax for a string
literal. len(u'\u9999'.encode('utf-8')) is 3, so this Chinese character
really encodes as three separate bytes. As these are non-ASCII bytes,
__repr__ choses a representation that is legal Python syntax. For that
characters, only \xe9, \xa6 and \x99 are valid Python syntax (each
representing a single byte). For a backslash, Python could have
generated \x5c or \134 as well, which are all different spellings
of "backslash in a string literal". Python chose the most legible
one, which is the double-backslash.

HTH,
Martin

 

[7stud]

Hi,

Thanks for the detailed response.

Yes and no. This is what str() does, but str() isn't called. Instead,
print inspects sys.stdout.encoding, and uses that encoding to encode
the string. That, in turn, may raise an exception (in particular if
sys.stdout.encoding is "ascii" or not set).

Is that the same as print calling encode(u_str, sys.stdout.encoding)

 

[7stud]

Hi,

Thanks for the detailed response.




Is that the same as print calling encode(u_str, sys.stdout.encoding)

ooops. I mean is that the same as print calling
u_str.encode(sys.stdout.encoding)?

 

[=?ISO-8859-1?Q?=22Martin_v=2E_L=F6wis=22?=]

ooops. I mean is that the same as print calling

u_str.encode(sys.stdout.encoding)?

Almost. It's rather

u_str.encode(sys.stdout.encoding or sys.getdefaultencoding())

(in case sys.stdout.encoding isn't set)

Regards,
Martin