In response to Alexandre's comment on Rietveld. Access to a local variable is faster than to a global one and the current implementation uses this for struct.pack. I just use same trick for struct.unpack. Here is a microbenchmark which demonstrate some effect of this optimization. I got 0.6491418619989417, 0.6947214259998873, and 0.5394902769985492 for optimized, non-optimized and advanced optimized functions.

Of course, we can achieve even better effect if we will cache not only struct.pack, but struct.Struct('<i').pack, struct.Struct('B').pack, etc. I were considered this as a reason for other patch, but we can do it in this issue.