UploadedFile doesn't iterate over lines with \r line endings

[Tai Lee]

UploadedFile.__iter__ loops over chunks and converts each chunk to a StringIO object and then iterates over that. Iterating over StringIO objects only splits on \n, and I'm not sure why we need StringIO at all. Couldn't we just do for line in chunk.splitlines() (which does split on \r and \n).

[Tai Lee]

fix iteration of UploadedFile objects on newlines.

[Tai Lee]

Looks like using splitlines will strip the line endings and break binary uploads. Here's a new patch that iterates through a call to re.findall with a simple regular expression instead of a StringIO object.

[Tai Lee]

Here's a use-case for this change. Upload a CSV file (which can have \r or \r\n line endings), and pass the UploadedFile object to csv.reader and you get Error: newline inside string when you try to iterate over it. When working with real files on the local file system you can open them with universal newline support, which is not available with StringIO or cStringIO objects.

[Tai Lee]

Updated for r8654 and fixed regular expression to not match empty strings.

[Malcolm Tredinnick]

Universal newline handling only makes sense for files generated/stored on the local system, where the newline convention is known. The UploadedFile class, pretty much by definition, is for data that has come from, and been generated on, a remote system that we know nothing about. So trying to use the appropriate line-ending handling is an unattainable goal for that situation. #8656 is the ticket to make sure this is documented.

[Jacob]

Milestone 1.0 deleted

[Tai Lee]

I'm re-opening this ticket because I have found a much simpler implementation, and this is still a problem for anyone who needs to parse uploaded text files line-by-line (e.g. CSV) when they can't know what line terminators end-users are using in their files. Excel 2011 on OS X is at least one commonly used program that still saves CSV data with \r line terminators by default.

Originally, there was a discussion between Malcolm and Jacob on IRC and the strongest objection at the time to fixing this was the ugly regular expression in my previous patch. This was also during a busy time, and it was decided that this should be a documentation issue.

I don't think the explanation given above when this ticket was wontfixed actually makes much sense. Uploaded file data which has come from and been generated on a remote system that we know nothing about is precisely the reason why we *need* universal newline support when iterating those files line-by-line.

I'm not sure how I missed it before, but when I said that we couldn't just use splitlines() because it consumed the line terminators, it seems I was wrong. splitlines() has a keepends argument that will actually do exactly what we need without any ugly regular expressions. Perhaps this wasn't the case back then, when we still had to support Python 2.3 and 2.4. I couldn't find confirmation in the Python 2.3 and 2.4 docs, but I did confirm that splitlines(True) works in Python 2.5 which is now our minimum version number.

I've fixed this in a branch on GitHub, and opened a pull request. All tests pass under OS X Lion with Python 2.7.2 and SQLite.

​https://github.com/thirstydigital/django/tree/tickets/8149-universal-newline-uploads

​https://github.com/django/django/pull/320

[Claude Paroz]

This iter method appears to be untested.

[Tai Lee]

Sorry. Don't know why I thought this didn't need tests. New pull request with test.

​https://github.com/django/django/pull/323

[Tai Lee]

Updated versionchanged note to 1.6. Any feedback?

[Tim Graham]

The test doesn't pass on Python 3.

[Jon Dufresne]

I've been running into this issue as well. Most notably when parsing uploaded files as CSV from Mac.

I added a submitted a pull request that works with Python 2 and 3: ​https://github.com/django/django/pull/3291

[Jon Dufresne]

Sorry in advance if this is incorrect usage of this flag.

I am removing "Patch needs improvement" as my latest pull request does work with Python 3 and there are no outstanding requests from a review.

[Tim Graham]

@jdufresne - that's correct usage. In fact, since tchaumeny reviewed it, he could mark the ticket as "ready for checkin".

[Tim Graham <timograham@…>]

In eb4f6de980c5148ba48d4ed67f31cca27dd132a8:

Fixed #8149 -- Made File.iter() support universal newlines.

The following are recognized as ending a line: the Unix end-of-line
convention '\n', the Windows convention '\r\n', and the old
Macintosh convention '\r'.

​http://www.python.org/dev/peps/pep-0278

Thanks tchaumeny for review.