Deepcopying a byte string is quicker than copying it - problem?

[Peter Otten]

Frank Millman wrote:

> Hi all
> 
> I noticed this a little while ago, but dismissed it as a curiosity. On
> reflection, I decided to mention it here in case it indicates a problem.
> 
> This is with python 3.3.2.
> 
> C:\>python -m timeit -s "import copy" "copy.copy('a'*1000)"
> 100000 loops, best of 3: 6.91 usec per loop
> 
> C:\>python -m timeit -s "import copy" "copy.deepcopy('a'*1000)"
> 100000 loops, best of 3: 11.8 usec per loop
> 
> C:\>python -m timeit -s "import copy" "copy.copy(b'a'*1000)"
> 10000 loops, best of 3: 79.9 usec per loop
> 
> C:\>python -m timeit -s "import copy" "copy.deepcopy(b'a'*1000)"
> 100000 loops, best of 3: 11.7 usec per loop
> 
> As you can see, deepcopying a string is slightly slower than copying it.
> 
> However, deepcopying a byte string is orders of magnitude quicker than
> copying it.
> 
> Actually, looking closer, it is the 'copy' that is slow, not the
> 'deepcopy' that is quick..
> 
> Expected, or odd?

Definitely odd. For some reason deepcopy() special-cases the bytes type 
while copy() does not. If you fix that:

$ python3.4 -m timeit -s 'import copy; b = b"a"*1000' 'copy.copy(b)'
10000 loops, best of 3: 21.2 usec per loop

$ python3.4 -m timeit -s 'import copy; copy._copy_dispatch[bytes] = 
copy._copy_immutable; b = b"a"*1000' 'copy.copy(b)'
1000000 loops, best of 3: 0.971 usec per loop

I think this an oversight rather than intentional. Please report to 
bugs.python.org.