The argument is essential to avoid slowdown with delayed ACKs and the Nagle algorithm, so it should be documented. It was added when fixing issue 4336. I believe that it's new in Python 2.7.

I added in docs for the method from the actual method docstring from the http.client module.

This is fixed the following changesets.

changeset a3f2dba93743
changeset 1ed413b52af3
changeset 277688052c5a

Thanks for the patch, Ben Hayden.

The 2.7 documentation should mention the version in which the argument was added. I believe it was 2.7.

I also left some comments on the review page that should be addressed.

New changeset befa7b926aad by Senthil Kumaran in branch '3.2':
Issue  #13073 - Address the review comments made by Ezio.
http://hg.python.org/cpython/rev/befa7b926aad

New changeset a7b7ba225de7 by Senthil Kumaran in branch 'default':
merge from 3.2. Issue  #13073 - Address the review comments made by Ezio.
http://hg.python.org/cpython/rev/a7b7ba225de7

New changeset 64fae6f7b64c by Senthil Kumaran in branch '2.7':
Issue13073 - Address review comments and add versionchanged information in the docs.
http://hg.python.org/cpython/rev/64fae6f7b64c

I believe, I have addressed all the comments. Closing this report.

It is IMO a source of confusion that the doc talk about a string instead of “a bytes object” (3.x) or “a string (str)” (2.x, unless unicode is supported too).

Yes, I agree. I think, it can be clarified at that point too. Because.
in 2.7 the string is being checked and in 3.3 the message_body is
checked if it's instance of bytes.  But, I think, it should be
carefully worded (aligned with how other socket message args are
mentioned).