A critic of Guido's blog on Python's lambda

[Alex Martelli]

Ken Tilton <kentilton at gmail.com> wrote:

> Martin P. Hellwig wrote:
> > Bill Atkins wrote:
> > <cut>
> > 
> >>
> >> How do you define scalability?
> >>
> > http://www.google.com/search?hl=en&q=define%3Ascalability&btnG=Google+Search
> > 
> 
> Damn! Google can do that?! Omigod!!! Not joking, I never knew that,a 

You're welcome; we do have several little useful tricks like that.

> lways used dictionary.com. Thx! I meant:
> 
> > The ability to add power and capability to an existing system without
> > significant expense or overhead. www.yipes.com/care/cc_glossary.shtml

Excellent -- just the definition of "scalability" that Google and its
competitor live and die by ((OK, OK, I'm _not_ implying that such issues
as usability &c don't matter, by no means -- but, I live mostly in the
world of infrastructure, where scalability and reliability reign)).


> The number of definitions explains why most respondents should save 
> their breath. Natural language is naturally ambiguous. Meanwhile Usenet
> is the perfect place to grab one meaning out of a dozen and argue over
> the implications of that one meaning which of course is never the one
> originally intended, as any reasonable, good faith reader would admit.

However, you and I are/were discussing exactly the same nuance of
meaning, either by a funny quirk of fate or because it's the one that
really matters in large-scale programming (and more generally,
large-scale systems).  E.g., if your existing system can gracefully
handle your current traffic of, say, a billion queries of complexity X,
you want to be able to rapidly add a feature that will increase the
average query's complexity to (X+dX) and attract 20% more users, so
you'll need to handle 1.2 billion queries just as gracefully: i.e., you
need to be able to add power and capability to your existing system,
rapidly and reliably, just as that definition says.

When this is the challenge, your choice of programming language is not
the first order of business, of course -- your hardware and network
architecture loom large, and so does the structuring of your
applications and infrastructure software across machines and networks.
Still, language does matter, at a "tertiary" level if you will. Among
the potential advantages of Lisp is the fact that you could use Lisp
across almost all semantic levels ("almost" because I don't think "Lisp
machines" are a realistic option nowadays, so lower levels of the stack
would remain in C and machine language -- but those levels may probably
be best handled by a specialized squad of kernel-level and device-driver
programmers, anyway); among the potential advantages of Python, the fact
that (while not as suited as Lisp to lower-level coding, partly because
of a lack of good solid compilers to make machine language out of it),
it brings a powerful drive to uniformity, rather than a drive towards a
host of "domain-specific" Little Languages as is encouraged by Lisp's
admirably-powerful macro system.

One key axis of scalability here is, how rapidly can you grow the teams
of people that develop and maintain your software base?  To meet all the
challenges and grasp all the opportunities of an exploding market,
Google has had to almost-double its size, in terms of number of
engineers, every year for the last few years -- I believe that doing so
while keeping stellar quality and productivity is an unprecedented feat,
and while (again!) the choice of language(s) is not a primary factor
(most kudos must go to our management and its approaches and methods, of
course, and in particular to the strong corporate identity and culture
they managed to develop and maintain), it still does matter.  The
uniformity of coding style and practices in our codebase is strong.

We don't demand Python knowledge from all the engineers we hire: for any
"engineering superstar" worth the adjective, Python is really easy and
fast to pick up and start using productively -- I've seen it happen
thousands of times, both in Google and in my previous career, and not
just for engineers with a strong software background, but also for those
whose specialties are hardware design, network operations, etc, etc. The
language's simplicity and versatility allow this. Python "fits people's
brains" to an unsurpassed extent -- in a way that, alas, languages
requiring major "paradigm shifts" (such as pure FP languages, or Common
Lisp, or even, say, Smalltalk, or Prolog...) just don't -- they really
require a certain kind of mathematical mindset or predisposition which
just isn't as widespread as you might hope.  Myself, I do have more or
less that kind of mindset, please note: while my Lisp and scheme are
nowadays very rusty, proficiency with them was part of what landed me my
first job, over a quarter century ago (microchip designers with a good
grasp of lisp-ish languages being pretty rare, and TI being rather
hungry for them at the time) -- but I must acknowlegde I'm an exception.

Of course, the choice of Python does mean that, when we really truly
need a "domain specific little language", we have to implement it as a
language in its own right, rather than piggybacking it on top of a
general-purpose language as Lisp would no doubt afford; see
<http://labs.google.com/papers/sawzall.html> for such a DSLL developed
at Google. However, I think this tradeoff is worthwhile, and, in
particular, does not impede scaling.


Alex