homepage

This is an example, last assert return an error :
f = open('data.txt', 'w')
f.write("""line 1
line 2
line 3
line 4
line 5
line 6
line 7
line 8
line 9
line 10
line 11
""")
f.close()
f = open('data.txt', 'r')
assert f.readline() == 'line 1\n'
assert f.read() == """line 2
line 3
line 4
line 5
line 6
line 7
line 8
line 9
line 10
line 11
"""
f.close()
import codecs
f = codecs.open('data.txt', 'r', 'utf8')
assert f.read() == """line 1
line 2
line 3
line 4
line 5
line 6
line 7
line 8
line 9
line 10
line 11
"""
f.close()
f = codecs.open('data.txt', 'r', 'utf8')
assert f.readline() == 'line 1\n'
# this assert return a ERROR
assert f.read() == """line 2
line 3
line 4
line 5
line 6
line 7
line 8
line 9
line 10
line 11
"""
f.close()
Regards,
Stephane

Hi Stephane,
I think you're seeing different buffering behavior, which I suspect is correct according to docs.
codecs.open should default to line buffering[1], while open uses the system default[2].
The read() where the assert fails is returning the remaining buffer from the readline (which read 72 chars).
Asserting e.g. "f.read(1024) == ..." will give you the expected result.
[1] http://docs.python.org/library/codecs.html#codecs.open
[2] http://docs.python.org/library/functions.html#open 

Buffering applies when writing, not when reading a file.
There is indeed a problem in codecs.py: after a readline(), read() will return the content of the internal buffer, and not more.
The "size" parameter is a hint, and should not be used to decide whether the character buffer is enough to satisfy the read() request.
Patch is attached, with test.

Amaury Forgeot d'Arc wrote:
> 
> Amaury Forgeot d'Arc <amauryfa@gmail.com> added the comment:
> 
> Buffering applies when writing, not when reading a file.
> 
> There is indeed a problem in codecs.py: after a readline(), read() will return the content of the internal buffer, and not more.
> 
> The "size" parameter is a hint, and should not be used to decide whether the character buffer is enough to satisfy the read() request.
> Patch is attached, with test.
Agreed.
The patch looks good except the if-line should read:
if chars >= 0 and len(self.charbuffer) >= chars:
 ...
Thanks,
-- 
Marc-Andre Lemburg
eGenix.com
________________________________________________________________________
::: Try our new mxODBC.Connect Python Database Interface for free ! ::::
 eGenix.com Software, Skills and Services GmbH Pastor-Loeh-Str.48
 D-40764 Langenfeld, Germany. CEO Dipl.-Math. Marc-Andre Lemburg
 Registered at Amtsgericht Duesseldorf: HRB 46611
 http://www.egenix.com/company/contact/ 

Updated patch.
[I also tried to avoid reading the underlying file if len(self.bytebuffer)>=size, but it does not work with multibytes chars when size=1]

I applied the diff to test_codecs in py3k, removed the u prefixes and ran: failure. I applied the fix and the test passed.

Up, I think this patch isn't applied in Python 3.3a0.

According to this ticket it hasn't been applied anywhere yet (a message will be posted here when it is).

See also #12446.

I think the patch is wrong or is not optimal for case when chars is -1, but size is not.
If we want to read all data in any case, then we should call self.stream.read() without argument if chars < 0 or size < 0.
If we want to read no more than size bytes, then all loop code should be totally rewritten.
Perhaps I am wrong.

As showed in issue12446, issue14475 and issue16636 there are different methods to reproduce this bug (read(size, chars) + readlines(), readline() + readlines()). All this cases should be tested.

Here is revised patch.
* Behavior is changed less. read() is less greedy and uses characters from the buffer when read() is called with only one argument (size). It is now a little closer to io stream's read() than with previous patch.
* Added tests for cases of issue12446 and issue16636.
* Fixed read() for for the TransformCodecTest.test_read test added in 3.4. Actually the uu_codec and zlib_codec are broken.

Patch looks good to me, but if any specific features are needed to work around misbehaving codecs (as per issue 20132), a comment in the appropriate place referencing that issue would be helpful.
And if that workaround means we can remove the special casing from the test_readlines test for the binary transform, cool :)

Actually this patch doesn't work around misbehaving codecs. It just makes 
specific tests (one readline, one read) be passed. More complex tests which use 
multiple readline's or read's still can fail with these misbehaving codecs.

New changeset e24265eb2271 by Serhiy Storchaka in branch '2.7':
Issue #8260: The read(), readline() and readlines() methods of
http://hg.python.org/cpython/rev/e24265eb2271
New changeset 9c96c266896e by Serhiy Storchaka in branch '3.3':
Issue #8260: The read(), readline() and readlines() methods of
http://hg.python.org/cpython/rev/9c96c266896e
New changeset b72508a785de by Serhiy Storchaka in branch 'default':
Issue #8260: The read(), readline() and readlines() methods of
http://hg.python.org/cpython/rev/b72508a785de