tempfile.TemporaryFile and httplib incompatibility

@atsuoishimoto

BPO	15267
Nosy	@atsuoishimoto, @orsenthil, @vstinner, @tjguk, @bitdancer, @vadmium, @serhiy-storchaka, @demianbrecht
Files	issue15267.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = 'https://github.com/serhiy-storchaka'
closed_at = <Date 2015-05-16.16:07:44.809>
created_at = <Date 2012-07-07.00:24:34.769>
labels = ['type-bug', 'OS-windows']
title = 'tempfile.TemporaryFile and httplib incompatibility'
updated_at = <Date 2015-05-16.16:07:44.808>
user = 'https://bugs.python.org/tzs'

bugs.python.org fields:

activity = <Date 2015-05-16.16:07:44.808>
actor = 'serhiy.storchaka'
assignee = 'serhiy.storchaka'
closed = True
closed_date = <Date 2015-05-16.16:07:44.809>
closer = 'serhiy.storchaka'
components = ['Windows']
creation = <Date 2012-07-07.00:24:34.769>
creator = 'tzs'
dependencies = []
files = ['26595']
hgrepos = []
issue_num = 15267
keywords = ['patch', 'needs review']
message_count = 14.0
messages = ['164764', '165074', '165076', '165093', '165183', '166867', '189285', '232533', '232534', '240469', '240470', '240481', '243327', '243328']
nosy_count = 12.0
nosy_names = ['ishimoto', 'orsenthil', 'vstinner', 'tim.golden', 'r.david.murray', 'python-dev', 'Ramchandra Apte', 'martin.panter', 'serhiy.storchaka', 'tzs', 'lemanyk1', 'demian.brecht']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue15267'
versions = ['Python 2.7']

In httplib.py, there is this code to try to get the body length:

    def _set_content_length(self, body):
        # Set the content-length based on the body.
        thelen = None
        try:
            thelen = str(len(body))
        except TypeError, te:
            # If this is a file-like object, try to
            # fstat its file descriptor
            try:
                thelen = str(os.fstat(body.fileno()).st_size)
            except (AttributeError, OSError):
                # Don't send a length if this failed
                if self.debuglevel > 0: print "Cannot stat!!"

However, if the body is a temporary file created by tempfile.TemporaryFile(), the len(body) in the first try throws an AttributeError, not a TypeError, on Windows and so it is not caught and unhappiness ensues. It is fine on Macintosh, and I would presume also on Linux.

Windows behaves different because on the other systems, TemporaryFile() returns an actual file object, and len() on a file throws TypeError. On Windows, TemporaryFile() returns an object that wraps the underlying file, and calling len() on that objects invokes this from tempfile.py (around line 371):

    def __getattr__(self, name):
        # Attribute lookups are delegated to the underlying file
        # and cached for non-numeric results
        # (i.e. methods are cached, closed and friends are not)
        file = self.__dict__['file']
        a = getattr(file, name)
        if not issubclass(type(a), type(0)):
            setattr(self, name, a)
        return a

Since the underlying file does not have a __len__ method, the getattr fails and throws AttributeError.

I'm sorry I'm not submitting a patch, but I do not know enough about the design of these two libraries to know whether the correct fix is for httplib to be more broad in the exceptions that cause it to check for a file when len() fails, or if the object returned by TemporaryFile() should be more closely mimicking a true file object and so should be made to throw TypeError when someone tries to use len() on it.

Could you create a failing test, please, Tim S?

BTW, type(0) should be replaced with int in the code.

Here is a program that demonstrates the problem:

    import httplib
    import tempfile

    f = tempfile.TemporaryFile()
    f.write("Hello, Temporary File!")
    f.seek(0)

    c = httplib.HTTPConnection('bugs.python.org')
    c.request('POST', '/', f, {'Content-type' : 'application/octet-stream'})

    r = c.getresponse()
    print r.status, r.reason

The expected output is "200 OK", and that's what happens on Mac and almost certainly on Linux. On Windows, this is the result:

    Traceback (most recent call last):
      File "bad.py", line 9, in <module>
        c.request('POST', '/', f, {'Content-type' : 'application/octet
      File "C:\Python27\lib\httplib.py", line 958, in request
        self._send_request(method, url, body, headers)
      File "C:\Python27\lib\httplib.py", line 989, in _send_request
        self._set_content_length(body)
      File "C:\Python27\lib\httplib.py", line 964, in _set_content_len
        thelen = str(len(body))
      File "C:\Python27\lib\tempfile.py", line 383, in __getattr__
        a = getattr(file, name)
    AttributeError: 'file' object has no attribute '__len__'

Changing the lines:

    f = tempfile.TemporaryFile()
    f.write("Hello, Temporary File!")
    f.seek(0)

to:

    f = open("temp.file", "w+")
    f.write("Hello, Less Temporary File!")
    f.seek(0)

makes it work on Windows.

I can confirm that this isn't a problem for 3.x. No-one's listed in the experts list for the httplib module but Senthil appears to be the de factor maintainer so I've added him to the call for his opinion. My feeling would be to do the simplest thing possible and add AttributeError to the list of trapped exceptions.

patch contains fix and test for 2.7.
With this patch, AttibuteError is captured as Tim sujested.

i think it is not good, len(fd) must raise AttributeError, not TypeError

Catching TypeError on len() looks as an ugly test to check if body is a string. Why not doing the opposite: first to call fileno() and call AttributeError? Or even use hasattr(body, "fileno")?

bpo-15267.patch: I would feel more confortable if test_send_tempfile() ensures that the socket contains the file content.

See also bpo-23740, about cleaning up the types for computing Content-Length in 3.5.

More general and simple solution is to make tempfile.NamedTemporaryFile new-style class.

Old-style class:

>>> import tempfile
>>> f = tempfile.NamedTemporaryFile()
>>> len(f)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/serhiy/py/cpython-2.7/Lib/tempfile.py", line 391, in __getattr__
    a = getattr(file, name)
AttributeError: 'file' object has no attribute '__len__'

New-style class:

>>> import tempfile
>>> f = tempfile.NamedTemporaryFile()
>>> len(f)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: object of type '_TemporaryFileWrapper' has no len()

We don't do classic to new style class changes for bug fixes without a compelling reason; we've been bitten by unexpected breakage doing that in the past.

New changeset c34513c2a894 by Serhiy Storchaka in branch '2.7':
Issue bpo-15267: HTTPConnection.request() now is compatibile with old-style
https://hg.python.org/cpython/rev/c34513c2a894

Catching TypeError or AttributeError on len() is a hack, but it is compatible with Python 3 and older code. We could handle this in bpo-23740.

Changing TemporaryFile would solve this issue, but only for TemporaryFile. len() raises AttributeError for other file-like classic classes. So I committed slightly modified Atsuo's patch (extended test as Victor suggested).

As for converting classic classes to new-style classes, see for example bpo-14399. It is not so harmless, as expected, unfortunately. But perhaps it was done multiple times in bug-fix releases.

Navigation Menu

Repository navigation

tempfile.TemporaryFile and httplib incompatibility #59472

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

tempfile.TemporaryFile and httplib incompatibility #59472

Description

Activity

tzs commented on Jul 7, 2012

tjguk commented on Jul 9, 2012

RamchandraApte commented on Jul 9, 2012

tzs commented on Jul 9, 2012

tjguk commented on Jul 10, 2012

atsuoishimoto commented on Jul 30, 2012

lemanyk1 commented on May 15, 2013

vstinner commented on Dec 12, 2014

vstinner commented on Dec 12, 2014

vadmium commented on Apr 11, 2015

serhiy-storchaka commented on Apr 11, 2015

bitdancer commented on Apr 11, 2015

python-dev commented on May 16, 2015

serhiy-storchaka commented on May 16, 2015

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions