Document.get_attachment() fails with binary data and when write_to is not None

[toddreed]

Using 2.0.0b2 with Python 3.5.0 and connecting to CouchDB 1.6.1.

I'm trying to access an attachment with this code:

image = Document(database, result['id'])
image_filename = '{0}.jpg'.format(image['_id'])

with open(image_filename, 'wb') as file:
    image.get_attachment('image', write_to=file, attachment_type='binary')

This produces the following exception:

  File "[redacted]/src/python/lemonaid.py", line 97, in seedannotations
    image.get_attachment('image', write_to=file, attachment_type='binary')
  File "[redacted]/lib/python3.5/site-packages/cloudant/document.py", line 396, in get_attachment
    write_to.write(resp.raw)
TypeError: a bytes-like object is required, not 'HTTPResponse'

In the debugger, the type of resp.raw is reported as <class 'requests.packages.urllib3.response.HTTPResponse'>.

Also, passing binary for attachment_type is not supported as documented. The get_attachment() method attempts to convert binary data to Unicode:

# from get_attachment...
        if attachment_type == 'json':
            return resp.json()
        return unicode_(resp.content)

When the response is binary, the last line above produces this exception:

  File "[redacted]/python3.5/site-packages/cloudant/document.py", line 400, in get_attachment
    return unicode_(resp.content)
  File "[redacted]/lib/python3.5/site-packages/cloudant/_2to3.py", line 95, in unicode_
    return astr.decode(ENCODING) if hasattr(astr, 'decode') else astr
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 0: invalid start byte

I would also argue that the attachment_type argument is ill-conceived. What if the caller does not know whether the attachment is binary or text. Ideally, the method should expose the content-type of the attachment rather than presume the caller knows. (In my case, my image attachment could be either PNG or JPEG, but since get_attachment() does not provide the content-type I'm guessing in my code right now.)

[alfinkel]

This does appear to be a bug. We'll take a closer look and resolve.

[alfinkel]

The write_to functionality has been fixed and the attachment will be handled based on the response Content-Type now. The attachment_type argument is now optional so that the caller can override the attachment handling logic if necessary.

[alfinkel]

Re: #106 (comment)

If I understand correctly, the functionality you are looking for is actually already available. The content type of any attachment is stored in a document's "_attachments" metadata. So you should either already have that information before calling get_attachment() or you will have that information as part of the call to get_attachment().

The Document content would look something like:

{
  "_id": "foo",
  "_rev": ...,
  ...
  "_attachments": {
    "bar": {
      "content_type": "image/jpeg",
      ...
    }
  }
}

If you already have a fetched version of the Document locally then you could find the the content_type and have your code react accordingly.

As in:

doc = db['foo']
content_type = doc['_attachments']['bar']['content_type']
# Handle your file settings and call get_attachment
...

If you are working with a Document object that has not been fetched from the server then get_attachment() performs a fetch of that document as an initial step. In this case you would not use the write_to argument and instead set the attachment_type='binary' which will return the raw data of the attachment. Then your code can interrogate the _attachments content_type and react accordingly to process the raw data returned by the method:

As in:

doc = Document(db, 'foo')
data = doc.get_attachments('bar', attachment_type='binary')
content_type = doc['_attachments']['bar']['content_type']
# Handle your file settings and process "data" as necessary
...

I hope this helps.

[toddreed]

Yes, that helps. Thanks for showing how the content type can be obtained.