> Bad wording perhaps, but:
> 
> +                    if not final:
> +                        n += 9  # next frame header
>                      write = self.file_write
>                      write(FRAME)
>                      write(pack("<Q", n))
> 
> does change how the frame length is calculated and emitted in the pickle
> stream.

Of course (as any optimizer). It produces more optimal pickled data which can 
be parsed by existing unpicklers.

> This is not compliant with how the PEP defines it (the frame size doesn't
> include the header of the next frame):
> http://www.python.org/dev/peps/pep-3154/#framing

"How the pickler decides to partition the pickle stream into frames is an 
implementation detail."

> > All tests are passed with this optimization
> 
> Well, perhaps there are not enough tests :-) But the protocol is
> standardized so that other people can implement it. The reference
> implementation can't do something different than the PEP does.

Could you write tests which exposes behavior difference without sticking 
implementation details?