Currently it only accept a filename (and writestr only accept bytes).

Here is a tentative implementation.

Feel free to review it.

zipfile.ZipFile takes a file arg that can be a 'path to a file (a string) or a file-like object'. The .write() method takes a filename arg. I would think that the proposal should be what the title says, to expand that to a file arg, either a path or 'file', as with ZipFile itself. I cannot think of any reason not to do that.

The patch adds a separate .writefp method with the file and arcname switched and the compression_type omitted. The code is mostly a duplicated subset of .write. While is proves the concept, none of the differences strike me as good. I think instead there should just be one method that branches, where appropriate, on the type of 'file', just as ZipFile.__init__ does. I admit that I do not understand what .write is supposed to do or does do when filename is a directory.

The patch includes a test. It still needs doc and News patches, but those come after the basic patch is done.

A more general feature is implemented in bpo-26039. Now you can write to ZIP archive the content of opened file object with following two lines:

        with zipf.open('file.txt', 'wb') as target:
            shutil.copyfileobj(source, target)