homepage

'rU' universal newline support is useless, because read lines end with
'\n' regardless of actual line end in the source file. Applications that
care about line ends still open file in binary mode and gather the stats
manually. 
So, to make this mode useful - the 'rbU' should be addded. Otherwise it
doesn't worth complication both in C code and in documentation.

The whole idea of universal newline mode is that the various possible
line endings ('\r', '\n' and '\r\n') are all mapped to '\n' precisely
so the user doesn't have to detect and fiddle with them. Using 'b' and
'U' together makes no sense.
* If you really want to see the line endings use 'rb'.
* If you don't care about the line endings regardless of source, use 'rU'.
* Otherwise use 'r'.

If you open file with 'r' - all line endings will be mapped precisely to
'\n' anyways, so it has nothing to do with 'U' mode.

> If you open file with 'r' - all line endings will be mapped precisely to
> '\n' anyways, so it has nothing to do with 'U' mode.
No they won't -- only the platform-specific newline will. On Unix, 'r'
and 'rb' are the same.

That's weird and the worst is that it is not documented. Manual says:
"If Python is built without universal newline support a mode with 'U' is
the same as normal text mode." 
but no information about what is "normal text mode" behaviour.
The way Python works that you describe is weird, but true. If developer
uses Windows platform - Unix and Windows files will be handled in the
same way, but not files from Mac platform. The worst that developer
can't know this, because he is unlikely to have any Mac files to test.
This behavior is like a long standing mine to collate Windows and Mac
Python users. Why not to fix it?

This behavior is inherited from the C-level fopen() and therefore
"normal text mode" is whatever that defines.
Is this really nowhere documented?

anatoly> If you open file with 'r' - all line endings will be mapped
 anatoly> precisely to '\n' anyways, so it has nothing to do with 'U'
 anatoly> mode.
Before 3.0 at least, if you copy a text file from, say, Windows to Mac, and
open it with 'r', you get lines which end in '\r\n'. Here's a simple
example:
 >>> open("dos.txt", "rb").read()
 'a single line\r\nanother line\r\n'
 >>> f = open("dos.txt")
 >>> f.next()
 'a single line\r\n'
 >>> f = open("dos.txt", "r")
 >>> f.next()
 'a single line\r\n'
 >>> f.next()
 'another line\r\n'
If, on the other hand, you open it with 'rU', the '\r\n' literal line ending
is converted, even though CRLF is not the canonical Mac line ending:
 >>> f = open("dos.txt", "rU")
 >>> f.next()
 'a single line\n'
 >>> f.next()
 'another line\n'
Skip

> This behavior is inherited from the C-level fopen() and therefore
> "normal text mode" is whatever that defines.
> Is this really nowhere documented?
Relation to fopen() function may be documented, but there is no
explanation of what "normal text mode" is. Is it really pythonic that a
script writer without former experience with C, stdio and fopen should
be aware of inherited fopen "behavior" when programming Python?

At least the 2.6 docs say
"The default is to use text mode, which may convert ``'\n'`` characters
to a platform-specific representation on writing and back on reading."

That's fine with me. I just need a 'rbU' mode to know in which format
should I write the output file if I want to preserve proper line endings
regardless of platform.
As for Python 2.6 note - I would replace "may convert" with "converts".

If you want to write your own line endings, read with "rU" and write
with "rb".

If lineends are mixed I would like to leave them as is.

Did you look at the io.open() function?
It's a new module in python2.6, but also the builtin "open" in py3k!
"""
 * On input, if newline is None, universal newlines mode is
 enabled. Lines in the input can end in '\n', '\r', or '\r\n', and
 these are translated into '\n' before being returned to the
 caller. If it is '', universal newline mode is enabled, but line
 endings are returned to the caller untranslated. If it has any of
 the other legal values, input lines are only terminated by the given
 string, and the line ending is returned to the caller untranslated.
"""
I suggest to try
 io.open(filename, newline="")

As I indicated in msg69679 if you want to see the line endings just open
the file in binary mode ('rb').

Thanks for the hints. It appeared that "universal text mode" is not for
crossplatform but for platform-specific programming. =)
So I gave it up and ended with my own 'rb' newlines counter and 'wb'
writer which inserts lines in required format.
As for 2.6 io.open()
http://docs.python.org/dev/library/io.html#module-io
- can anybody point what's the difference between text mode with
newlines='' and binary mode?
- the comment about newline=<string>
"If it is '', universal newline mode is enabled, but line endings are
returned to the caller untranslated. If it has any of the other legal
values, input lines are only terminated by the given string, and the
line ending is returned to the caller untranslated."
does it mean that if newline='\r\n' is specified all single '\n'
characters are returned inline?

> does it mean that if newline='\r\n' is specified all single '\n'
> characters are returned inline?
Yes.
Let's take a file with mixed newlines:
>>> io.open("c:/temp/t", "rb").read()
'a\rb\r\nc\nd\n'
rb mode splits only on '\r\n' (I'm on Windows)
>>> io.open("c:/temp/t", "rb").readlines()
['a\rb\r\n', 'c\n', 'd\n']
rU mode splits on every newline, and converts everything to \n
>>> io.open("c:/temp/t", "rU").readlines()
[u'a\n', u'b\n', u'c\n', u'd\n']
newline='' splits like rU, but does not translate newlines:
>>> io.open("c:/temp/t", newline='').readlines()
[u'a\r', u'b\r\n', u'c\n', u'd\n']
newline='\r\n' only splits on the specified string:
>>> io.open("c:/temp/t", newline='\r\n').readlines()
[u'a\rb\r\n', u'c\nd\n']

This '\r' makes things worse. I am also on Windows and didn't thought
that "rb" processes '\r\n' linefeeds as a side-effect of '\n' being the
last character. Thanks.
newline='' is just what I need. I guess there is no alternative to it in
2.5 series except splitting lines returned from binary read manually.
What about file.newlines attribute - is it preserved in 2.6/Py3k?
BTW, it would be nice to have this example in manual.

Please read
http://docs.python.org/dev/library/io.html#io.TextIOBase.newlines