Re: OS X python.el completion issue

[Stefan Monnier]

>>> With a problem case, not only does read-from-string result in nil, but
>>> python-send-receive results in nil as well.  When debugging
>>> python-send-receive in the same fashion, I run into the phenomenon
>>> where accept-process-output always times out and returns nil, even when
>>> given input known to work. Perhaps this is a result of edebug vs.
>>> regular execution?
>> 
>> Yes, edebugging with things like accept-process-output is
>> always problematic.  I generally resort to sprinkling the code with
>> (message "here var=%S" var).
>> 
>>> I seem to recall issues with OS X and ptys; could this be one of them?
>> 
>> Could be.  Can you try to change process-connection-type to avoid the use of
>> ptys, and see if it helps?

> That had no effect on the behavior.  I guess it must actually be
> an issue with the size of the incoming data vs. some buffer size
> in the C code ... does this sound plausible?

It sounds possible, your experiments seem to indicate the boundary could be
1024 bytes.

> I will write up some scripts to incrementally generate and test Python
> attrs and see what the maximum working length is, for lack of any insight
> into the root cause.

Maybe you could add a `message' at the beginning of python-preoutput-filter,
to see what the process filter actually receives.  Or maybe even at the
beginning of comint-output-filter.


        Stefan