Traceback on slow connections when using python3 about chopsticks

[lordmauve]

This is a very detailed report, and an interesting issue. I will look into it.

[lordmauve]

It looks as though the change in Python 3 is that an unbuffered file object opened with fdopen no longer returns exactly size bytes - instead it focuses on only doing one system call.

We don't actually need an unbuffered file object in general - when we create 'inpipe' in bubble.py, it is buffered. However, the problem is that if we make the sys.stdin file object in PYTHON_ARGS buffered, it could read extra data into its buffers, which is thrown away in bubble.py once we dup all the file descriptors around and open inpipe.

Here's a slightly odd solution that I think might work - only buffer the exact amount we need:

PYTHON_ARGS = [                                                                 
    '-usS',                                                                     
    '-c',                                                                       
    'import sys, os; _bsz = %d ;' % len(bubble) +                               
    'sys.stdin = os.fdopen(0, \'rb\', _bsz); ' +                                
                                                                                
    '__bubble = sys.stdin.read(_bsz); ' +                                       
    'exec(compile(__bubble, \'bubble.py\', \'exec\'))'                          
]  

Could you try this and let me know if it works for you?

[lordmauve]

I'm not very happy with that actually. What we really need to do is keep hold of whatever is in the buffer - either modify or re-use the file object. I guess the simplest thing would be to do the stdin dup() in PYTHON_ARGS and don't re-open stdin in bubble.py.

[amigrave]

Here's my attempt following your directives. Tested on python2.7 and python3.5