[Tutor] Const on Python

[Andreas Kostyrka]

Am Donnerstag, den 06.03.2008, 08:35 -0500 schrieb Kent Johnson:
> C++ is extremely complex. The good side of this is it gives you 
> tremendous control - final, const, pass by reference or value, memory 
> allocation, etc, etc. The bad side is that it is a lot to think about - 
> should this parameter be const? who is going to deallocate this object? 
> and there is a lot of room for error, whole books have been written on 
> the topic (e.g. Effective C++ which even has a sequel).

Concur. What is worse, it's to complicated for the average "commercial"
developer. And the claim that you can program C++ effectively without
knowing the whole language does not work in practice. (Hint: Yes, you
can program without understanding the whole language. And no, you have
no chance of debugging your program if you don't understand what that
complicated library is doing. So in practice you end with senior
developers doing the debugging, while the newbies try to understand what
you are doing.)

As a saving grace to C++ I must note that there are a number of
"features" in Python that can produce something comparativly
newbie-unfriendly as C++. But you will notice that most Python
developers shrink away from using these features, so in effect this is
not a burning issue. Somehow these "advanced" topics don't deter newbies
from writing and debugging Python programs. (E.g. you can program for
years Python without learning the fact that it's not the class that does
the self bounding, instead it's function.__get__ that does it ;) )

> 
> Java is less complex and flexible than C++. At first I missed the 
> control of C++, then I realized that it wasn't really buying me much and 
> that the cost was too high - coding Java is much less work than coding C++.

Java has automatic memory management, is safe (meaning that typically
the worst thing that happens is traceback), and has an object model that
nearer to a dynamic language like Python or Smalltalk than to C++.
Basically, while it looks much like C++ on a first glance, the
underpinnings come from more dynamic/runtime oriented languages.

And because it does all these irrelevant details for the developer
automatically, the IT industry can get enough souls to program it. ;)

> 
> Then Python. Python takes away even more low-level control. And I don't 
> miss it at all. My Python programs work just fine without static typing, 
> constants, etc.

Worse, something that one likes to forget. Practically in any software
system the amount of code that is heavily exercised, the "inner loop" is
tiny. Meaning that after having written your Python code (usually in a
fraction of the time planned for the C++ implementation), you can
improve the algorithms and speed up these critical sections. (By redoing
them in Python, Pyrex, C/C++.)

That's the underlying truth to "Python faster than C++" message. Not
because Python is faster than C++ (in execution speed). Of the primitive
operations around 98-98% are slower in Python than in C/C++. But it
allows you to implement your application faster than in C++. The initial
implementation will be slower than a C++ implementation. But it might be
already "fast enough". But the point is, that in most cases, your
friendly C++ coder in the next room is still implementing his initial
implementation while the Python coder is already benchmarking/tuning his
app. Or spends time thinking about the algorithms.

So the following stands, usually, and you can usually even replace the
languages with Perl, Ruby, PHP, Smalltalk, Lisp, Java, ...:

C++ is faster than Python (if the resources in money and time are
unlimited: that's why many C++ gurus in academy have a strong different
feeling).

Python is faster than C++ (in time to market, if resources are limited)
Python is faster than C++ (in execution time, if resources are limited
enough so that the C++ version is an unoptimized first try, while the
Python version is usually a 2nd or 3rd generation implementation).


In commercial practice sometimes "political" considerations make for
crazy decisions, e.g. witnessed by me:

1.) a new web service needs to be developed.
2.) the policy of the company is that all developers need to know C
++/Java.
3.) the project lead calls in the next free developer to his office, and
as it happens it's one of the local C++ gurus.
4.) The C++ guru, being slightly bored, tells the project lead that in
his own estimation, it would be faster to do the service in Python, even
with him learning Python first. It would make even more sense
technically to do the webservice in Erlang, as it has to interface with
an Erlang server. He'd be happy to learn more Erlang too ;)
5.) The project lead makes him to do it in C++ anyway. Would make the
scheduling of the group way to complicated if you could not assign any
developer to every ticket coming in.

That's the moment where you start scratching your head.

> Python is simple and flexible and just does what I want with no fuss. 
> Java is a heavy straightjacket. C++ is inconceivably complex.

Concur. It's just that most people in the IT industry don't grasp it. Or
the fact that especially C++ comes like an iceberg with the worst under
the waterline: Debugging that stuff. (Even with the nicest coolest
debugging tools, debugging C++ still takes an experienced developer with
intuition.)

> So relax and enjoy it, the only risk is that you will never want to 
> touch Java and C++ again.

Well, at least not without getting an premium for all the pain ;)

Andreas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Dies ist ein digital signierter Nachrichtenteil
Url : http://mail.python.org/pipermail/tutor/attachments/20080306/462f0fd3/attachment.pgp