Path: csiph.com!usenet.pasdenom.info!dedibox.gegeweb.org!gegeweb.eu!nntpfeed.proxad.net!proxad.net!feeder1-2.proxad.net!usenet-fr.net!nerim.net!novso.com!newsfeed.xs4all.nl!newsfeed3a.news.xs4all.nl!xs4all!post.news.xs4all.nl!not-for-mail
MIME-Version: 1.0
In-Reply-To: <5384c539$0$29978$c3e8da3$5496439d@news.astraweb.com>
References: <jn415bx6uf.ln2@news.ducksburg.com> <mailman.10220.1400764235.18130.python-list@python.org> <05c15bxrpj.ln2@news.ducksburg.com> <mailman.10223.1400768058.18130.python-list@python.org> <k9f15bxoql.ln2@news.ducksburg.com> <mailman.10225.1400772863.18130.python-list@python.org> <sck35bxdc5.ln2@news.ducksburg.com> <mailman.10239.1400842778.18130.python-list@python.org> <amme5bxc3v.ln2@news.ducksburg.com> <5384c539$0$29978$c3e8da3$5496439d@news.astraweb.com>
Date: Wed, 28 May 2014 05:16:24 +1000
Subject: Re: hashing strings to integers
From: Chris Angelico <rosuav@gmail.com>
Cc: "python-list@python.org" <python-list@python.org>
Content-Type: text/plain; charset=UTF-8
Precedence: list
Newsgroups: comp.lang.python
Message-ID: <mailman.10372.1401218186.18130.python-list@python.org>
Lines: 40
NNTP-Posting-Host: 2001:888:2000:d::a6
Xref: csiph.com comp.lang.python:72123

On Wed, May 28, 2014 at 3:02 AM, Steven D'Aprano
<steve+comp.lang.python@pearwood.info> wrote:
> But I know that Python is a high-level language with
> lots of high-level data structures like dicts which trade-off time and
> memory for programmer convenience, and that I'd want to see some real
> benchmarks proving that my application was too slow before giving up that
> convenience with a complicated strategy like this.

And they trade off that time and memory on the basis of X years of
development expertise. A while ago (about ten years or so, now... wow,
that's quite a while) I wrote a C++ program that needed an
ever-growing array; for simplicity, I went with a very basic system of
doubling the size every time, from a base of something like 1024 or
8192. (Note that it was storing and moving around only pointers, so
it's comparable to Python's list.) That means it has an average 25%
slack space at all times, more until its first allocation, and every
now and then it has a huge job of copying a pile of pointers into a
new array. (Imagine it's now at 16777216 and it needs to add the
16,777,217th string to the array. Bye-bye CPU caches.) These
boundaries became *user-visible pauses*, fortunately at predictable
points, but on a not-terrible computer it could cause a >1s pause just
copying heaps of pointers. How do you think a Python list will
perform, under the same workload (periodic appending of single strings
or small groups of strings, very frequent retrieval based on index -
probably about a 20:1 read:write ratio)? Not only would it be far more
convenient, it's probably going to outperform my hand-rolled code -
unless I'm so brilliant (or lucky) that I can stumble to something as
good as can be achieved with years of dedicated development. Yeah, I
don't think so.

Same goes for hashing algorithms. I can at least boast that I've never
written one of those... although I have several times written a binary
tree of one form or another, in order to provide comparable features
to a dict. And once again, now that I know how convenient and
performant high level languages can be (which may or may not have been
true 15 years ago, but I certainly didn't know it), I don't rewrite
what can be done better by just tossing in a couple of curly braces
and saying "here, map this to that".

ChrisA