Why are slice indices the way they are in python?

[Steve Bergman]

A couple of off the wall questions.

It seems to me that there is usually a solid *reason* for things in
Python and I'm wondering about the rationale for the way slicing works:

my_string[2:5]

gets you the 3rd through the 3rd through the 5th character of the
string because indexing starts at 0 and you get everything up to, but
not including the second index.

Why? It doesn't seem intuitive to me. To me, it makes it harder, not
easier, to work with slices than if indexing started at 1 and the
above expression got you the 2nd throught the 5th character.

Another thing that I've been puzzling over is the pow() function.

pow(x,y) gives x**y. Fine.

But pow(x,y,z) gives (x**y) % c

I'm curious to know what the pressing reason for such a feature was.

[Terry Reedy]

"Steve Bergman" <st***@rueb.comwrote in message
news:11**********************@14g2000cws.googlegro ups.com...

>A couple of off the wall questions.

It seems to me that there is usually a solid *reason* for things in
Python and I'm wondering about the rationale for the way slicing works

Yes, see below.

my_string[2:5]

gets you the 3rd through the 3rd through the 5th character of the
string because indexing starts at 0 and you get everything up to, but
not including the second index.

Why?

len(s[2:5])== 5-2
s[2:5] + s[5:7] == s[2:7]

Another thing that I've been puzzling over is the pow() function.

pow(x,y) gives x**y. Fine.

But pow(x,y,z) gives (x**y) % c

you obvious meant (x**y) % z

>
I'm curious to know what the pressing reason for such a feature was.

Such exponential remainders are used in cryptography work, for instance,
with fairly large values, and can be efficiently computed, especially in C,
*without* computing a humongous x**y intermediate value.

Since this is an int function, it does not really belong in the floating
point math module.

Terry Jan Reedy

[Ant]

Steve Bergman wrote:

....

Why? It doesn't seem intuitive to me. To me, it makes it harder, not
easier, to work with slices than if indexing started at 1 and the
above expression got you the 2nd throught the 5th character.

Dijkstra has an article about this:
http://www.cs.utexas.edu/users/EWD/t...xx/EWD831.html

In practice it probably wouldn't make a great deal of difference. Here
at my current job, we have a situation where a bunch of Progress 4GL
programmers have been taught Java and let loose on the code base. we
now have a situation where there are a plethora of methods written
designed to be 1-based, and you have to trawl through the code to work
out if the method was written by a Java programmer or a 4GL programmer
to save on array index exceptions...

The infuriating thing is that if you are iterating over an array in
Java, and have to use one of the functions based on an array index, you
have to use function(i + 1). Drilling back through the code, and
invariably you have the function written:

def function(i):
return array[i-1]

!!! (at least you would if the Java was written in Python by 4GL
programmers, but you get the idea ;-) )

[ogbash]

Why? It doesn't seem intuitive to me. To me, it makes it harder, not
easier, to work with slices than if indexing started at 1 and the
above expression got you the 2nd throught the 5th character.

Zero-based indices and excluding last index often works nicer, it is
not the rule, but all my experience says that.

a[i:i] is very handy to have in many algorithms, so you don't have to
exclude this case explicitly.
Index starting at 1 or including N often adds +-1 expressions to the
code. I think that http://en.wikipedia.org/wiki/Modular_arithmetic
makes it more intuitive.

And if talking about dates, then I suggest NEVER use 2006-12-31
23:59:59 in data, always 2007-01-01 00:00:00 instead. If customer wants
to see former on the screen just substruct 1 millisecond from the data
and round it to what is needed. It makes all arithmetics, finding
period collapses, time differences, etc much more painless. You
sometimes don't even know if your database supports seconds,
milliseconds, microseconds and what your operating system and
programming language support. So if you are using inclusive end date
you may find yourself in trouble very soon.

Again, its not intuitive for ordinary people, but math people and
programmers often find 0-based and exlusive ends are nicer to work
with, at least me :)

Another thing that I've been puzzling over is the pow() function.

pow(x,y) gives x**y. Fine.

But pow(x,y,z) gives (x**y) % c

I'm curious to know what the pressing reason for such a feature was.

This makes cryptography code 100 times faster.

Oleg

[Michael Torrie]

On Thu, 2006-11-30 at 08:58 -0800, og****@gmail.com wrote:

And if talking about dates, then I suggest NEVER use 2006-12-31
23:59:59 in data, always 2007-01-01 00:00:00 instead.

Forgive me for my ignorance, but isn't "2006-12-31 23:59:59" actually
one entire second earlier than "2007-01-01 00:00:00"? Hence they are
two different dates are they not?

Michael

[Fredrik Lundh]

Michael Torrie wrote:

Forgive me for my ignorance, but isn't "2006-12-31 23:59:59" actually
one entire second earlier than "2007-01-01 00:00:00"? Hence they are
two different dates are they not?

given the context, maybe the poster meant that he's storing date
*ranges* as [start, stop) (i.e. including start but not including
stop).

</F>