Issue2062

Using the latest jython2.7, I'm not able to use bytearray with os.write.
In CPython2.7.5 I get:

>>> import os
>>> fd = os.open('/tmp/foo', os.O_RDWR)
>>> os.write(fd, bytearray('foobar'))
6

From jython 2.7b1 I get:

>>> import os
>>> fd = os.open('/tmp/foo', os.O_RDWR)
>>> os.write(fd, bytearray('foobar'))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: write(): 2nd arg can't be coerced to String

A quick-n-dirty fix with adding an overloaded method.

A complete solution, here and many other places where a String argument really means bytes, is to accept any PyObject supporting the Buffer API. Then this would work as it does on CPython:

>>> fd = os.open('x.tmp', os.O_CREAT | os.O_RDWR)
>>> a = bytearray("Stuff and nonsense. ")
>>> os.write(fd, a)
20
>>> os.write(fd, buffer(a))
20
>>> os.write(fd, memoryview(a))
20

It would be cool to find some automagic implementation, but I think it has to be a ByteUtil class (or some such), offering few well-crafted static methods, and a re-write of these methods one by one. Asking a buffer for its String is least work but goes completely against the idea of the buffer API as efficient in-place access, unless it's really a PyString after all.

Blocker for beta 4

Clean & polished fix, now in my local code base, behaves like this:
>>> import os
>>> s = "This is a test."
>>> m = memoryview(s)
>>> fd = os.open("x.tmp", os.O_WRONLY | os.O_CREAT)
>>> os.write(fd, s)
15
>>> os.write(fd, m)
15
>>> os.write(fd, bytearray(s))
15
>>> os.write(fd, buffer(s))
15
>>> os.write(fd, 1.5)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: write(): 2nd arg can't be coerced to org.python.core.BufferProtocol
>>> os.write(fd, m[1::2])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
BufferError: underlying buffer is not C-contiguous
>>> os.close(fd)
>>> os.write(fd, s)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: [Errno 9] Bad file descriptor 

I was surprised that I could declare this simply as:
public static int write(PyObject fd, BufferProtocol bytes)
and let Jython take care of the type test. (Indra Talip did this in binascii.) Previously I've accepted a PyObject, then tested it with instanceof. It means the error message is not quite the same as CPython when you hand it, say, a float in place of bytes, but the code is simpler.

Hmm ... a test: good idea. I'll add one to test_os_jy.

I assume we are still pending a test, but otherwise it is certainly fixed. I will keep this open for another week or so.

The test is here http://hg.python.org/jython/rev/0944779f5d1c, so I take your "unless" as a vote to close, and agree.