ReadFile - Change to support overlapped IO

[exvito]

In #115 I mentioned what I believe are current limitations to support overlapped IO in ReadFile.

I investigated a bit and, maybe, if...

• ReadFile instantiates a Python bytearray (which is mutable) as the read buffer.
• Uses ffi.from_buffer(bytearray) as the lpBuffer for ReadFile
• Returns that bytearray as the result

...then the OS will be able to write the data there and the caller, now holding a reference to that mutable bytearray, can access it (after GetOverlappedResult or equivalent signals its ok, of course).

I'm just not sure about:

• Memory ownership / leaks that can result from this approach (needs thinking).
• How Pythonic such an approach is: call ReadFile and get a muttable buffer which will change under your feet, once the overlapped IO is completed.

[opalmer]

Memory ownership / leaks that can result from this approach (needs thinking).

I think that's managed by cffi, but it's probably something we could test.

How Pythonic such an approach is: call ReadFile and get a muttable buffer which will change under your feet, once the overlapped IO is completed.

I wouldn't call it Pythonic but I would said it's probably fairly standard from a Windows perspective. Considering pywincffi is wrapping Windows functions it's probably better to be more windows-like in cases like this than Pythonic (which should make the behavior more predictable).