Comments on Version 2, Draft Pep for Deprecating Builtins

[Huaiyu Zhu]

Jeff Epler <jepler at unpythonic.net> wrote:
>On Thu, May 09, 2002 at 06:30:36PM +0000, Huaiyu Zhu wrote:
>> I don't see the reason to prevent 2**2002.
>
>$ time python2.3 -c '2**2**20'
>86.28user 0.40system 1:36.95elapsed 89%CPU (0avgtext+0avgdata 0maxresident)k
>0inputs+0outputs (304major+300minor)pagefaults 0swaps
>... and AFAIK that's with the GIL held, so it means anybody who can
>r_eval an arbitrary string blocks your entire application for an
>arbitrary amount of time (or until the intermediate result no longer
>fits in RAM)
>
>Or try
>$ time python2.3 -c '[[0] * 2**x for x in range(15, 31)]'
>Traceback (most recent call last):
>  File "<string>", line 1, in ?
>MemoryError
>Command exited with non-zero status 1
>6.76user 2.69system 1:02.34elapsed 15%CPU (0avgtext+0avgdata 0maxresident)k
>0inputs+0outputs (38459major+97990minor)pagefaults 32247swaps
>... I thrashed the machine pretty good for a minute.  I could have
>thrashed it for longer if I'd tried.
>
>
>> But if there is such a reason,
>> maybe one should tweak r_eval to handle it?
>
>r_eval doesn't perform this level of inspection of code.

Oh, you are thinking about complexity analysis.  I'm afraid that can't be
done generally by any tool, unless it is really draconian, ie blocking many
perfectly valid strings like "2**2002".

Huaiyu