The article https://atleastfornow.net/blog/not-all-bytes/ says that bytes.join(sequence) is slower on Python 3 compared to Python 2. I compared Python 2 and Python 3 code: the main different seems to be that Python 3 uses the Py_buffer API to support more types than only bytes, and that it allocates a buffer (on the C stack or on the heap memory) of Py_buffer objects. Attached patch makes bytes.join(sequence) up to 29% faster. The patch avoids the Py_buffer API and the allocation of the temporary array of Py_buffer if all items are bytes or bytearray. I'm not 100% sure that it's worth it to optimize bytes.join(). On Python 2, bytes += bytes uses an hack in Python/ceval.c to optimize this instruction. On Python 3, the optimization is only applied to str += str (unicode), bytes += bytes is inefficient. To get best performances on Python 2 and Python 3, bytes.join(sequence) is better than bytes+=bytes. Microbenchmark commands: $ ./python -m perf timeit -s "sep=b''; seq=(b'hello', b'world')" 'sep.join(seq)' $ ./python -m perf timeit -s "sep=b''; seq=(b'hello', b'world', b'. ') * 100" 'sep.join(seq)' Python 3.6 => patched Python3.6: * 2 items: 92.1 ns +- 1.8 ns => 90.3 ns +- 3.1 ns (-2%) * 300 items: 3.11 us +- 0.07 us => 2.22 us +- 0.11 us (-29%) -- I'm not sure that Python 3 is really slower than Python 2 :-/ Python 3.5 is 10 ns slower tha Python 2.7 for sequence of 2 items, but it's 6% faster for 300 items. So the question is if it's worth it to optimize bytes.join(). Python 2: * 2 items: 87.7 ns +- 3.7 ns * 300 items: 3.25 us +- 0.11 us Python 3.5: * 2 items: 97.4 ns +- 9.0 ns * 300 items: 3.06 us +- 0.20 us
Tests in the article don't look reliable. For example bytes_percent() and bytes_plus() test nothing, because b"%s %s" % (b"hi", b"there") and b"hi" + b" " + b"there" are evaluated at compile time. Yes, bytes.join(sequence) is a little slower on Python 3 for short sequences. But for long sequences Python 3 is faster. The code for bytes.join() is already too complex, and the proposed optimization makes it more complicated. And the optimization decreases performance on my netbook: $ ./python -m timeit -s "sep=b' '; seq=(b'hello', b'world')" -- 'sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq); sep.join(seq)' Python 2.7: 100000 loops, best of 3: 7.24 usec per loop Python 3.6 unpatched: 100000 loops, best of 3: 8.62 usec per loop Python 3.6 patched: 100000 loops, best of 3: 9.11 usec per loop
