Basic question about speed/coding style/memory

[Jan Riechers]

Hello Pythonlist,

I have one very basic question about speed,memory friendly coding, and
coding style of the following easy "if"-statement in Python 2.7, but Im
sure its also the same in Python 3.x

Block
#----------------------------------
if statemente_true:
doSomething()
else:
doSomethingElseInstead()

#----------------------------------

versus this block:
#----------------------------------
if statement_true:
doSomething()
return

doSomethingElseInstead()

#----------------------------------


I understand the first pattern that I tell the interpreter to do:
Check if the conditional is true, run "doSomething()" else go inside the
else block and "doSomethingElseInstead()".

while the 2nd does only checks:
doSomething() if statement_true, if not, just go directly to
"doSomethingElseInstead()


Now, very briefly, what is the better way to proceed in terms of
execution speed, readability, coding style?
Letting out the fact that, in order to prevent
"doSomethingElseInstead"-Block to execute, a return has to provided.

Thank you for reading and hope someone brings light into that.

Your fellow python programmer
Jan

[Andrew Berg]

On 7/21/2012 2:33 AM, Jan Riechers wrote:
> Block
> ...
> versus this block:
> ...

> Now, very briefly, what is the better way to proceed in terms of
> execution speed, readability, coding style?

Using if/else is the most readable in the general sense. Using return
(or break or continue as applicable) in this manner would indicate (at
least to me) that it's an exceptional or otherwise special case and that
the function can't do what it's supposed to. In that case, I would try
to catch an exception rather than use if/else whenever possible. I
highly doubt there is a significant performance difference between them.
--
CPython 3.3.0b1 | Windows NT 6.1.7601.17803

[Jan Riechers]

Hello Andrew,

Your answer is right, in other circumstances I also would stick to
try/except, break-statements in loops and so forth.
But the question was a bit more elementary.

Cause, as I understand the interpreter chooses either the "else" (1st
block) or just proceeds with following code outside the if.

So if there is some overhead in some fashion in case we don't offer the
else, assuming the interpreter has to exit the evaluation of the
"if"-statement clause and return to a "normal parsing code"-state
outside the if statement itself.

I hope this explanation makes more sense in what I want to ask ;)

Jan

[Chris Angelico]

On Sat, Jul 21, 2012 at 5:33 PM, Jan Riechers <janp...@freenet.de> wrote:
> Block
> #----------------------------------
> if statemente_true:
> doSomething()
> else:
> doSomethingElseInstead()
>
> #----------------------------------

This means, to me, that the two options are peers - you do this or you do that.

> versus this block:
> #----------------------------------
> if statement_true:
> doSomething()
> return
>
> doSomethingElseInstead()
>
> #----------------------------------

This would be for an early abort. Don't bother doing most of this
function's work, just doSomething. Might be an error condition, or
perhaps an optimized path.

Definitely for error conditions, I would use the second option. The
"fail and bail" notation keeps the entire error handling in one place:

def func(x,y,z):
if x<0:
y+=5
return
if y<0:
raise PEBKAC("There's an idiot here somewhere")
# ... do the rest of the work

Note the similarity between the control structures. Raising an
exception immediately terminates processing, without polluting the
rest of the function with an unnecessary indentation level. Early
aborting through normal function return can do the same thing.

But this is purely a matter of style. I don't think there's any
significance in terms of processing time or memory usage, and even if
there is, it would be dwarfed by considerations of readability. Make
your code look like what it's doing, and let the execution take care
of itself.

ChrisA

[Andrew Berg]

On 7/21/2012 3:13 AM, Jan Riechers wrote:
> Cause, as I understand the interpreter chooses either the "else" (1st
> block) or just proceeds with following code outside the if.

If none of the if/elif statements evaluate to something true, the else
block is executed.

> So if there is some overhead in some fashion in case we don't offer the
> else, assuming the interpreter has to exit the evaluation of the
> "if"-statement clause and return to a "normal parsing code"-state
> outside the if statement itself.

I really don't understand. You can look into the dis module if you want
to look at how CPython bytecode is executed and the timeit module to
measure speed. In any case, I don't see how there would be any
significant difference.

http://docs.python.org/py3k/library/dis.html
http://docs.python.org/py3k/library/timeit.html

[Steven D'Aprano]

On Sat, 21 Jul 2012 10:33:27 +0300, Jan Riechers wrote:

> Hello Pythonlist,
>
> I have one very basic question about speed,memory friendly coding, and
> coding style of the following easy "if"-statement in Python 2.7, but Im
> sure its also the same in Python 3.x

I assume that the following is meant to be inside a function, otherwise
the return in the second example is illegal.

But in general, you're worrying too much about trivia. One way or the
other, any speed difference will be trivial. Write whatever style reads
and writes most naturally, and only worry about what's faster where it
actually counts.

To give it an analogy that might be clear, this question is not too far
from worrying about whether your car will be faster with the radio aerial
up or down. Yes, technically the car will be slower with the aerial up,
due to air resistance, but you'd have a job measuring it, and it makes no
difference whether you are zooming down the highway at 120mph or stuck in
traffic crawling along at 5mph.


Here's a minimal example:


def with_else(x):
if x:
a = x
else:
a = x+1
return a


def without_else(x):
if x:
a = x
return a
a = x+1
return a


Notice that I try to make each function do the same amount of work, so
that we're seeing only the difference between "else" vs "no else".

Now let's test the speed difference with Python 2.7. Because this is
timing small code snippets, we should use the timeit module to time the
code:

from timeit import Timer
setup = "from __main__ import with_else, without_else"
t1 = Timer("for i in (0, 1): result = with_else(i)", setup)
t2 = Timer("for i in (0, 1): result = without_else(i)", setup)

Each snippet calls the function twice, once to take the if branch, then
to take the else branch.

Now we time how long it takes to run each code snippet 1000000 times. We
do that six times each, and print the best (lowest) speed:

py> min(t1.repeat(repeat=6))
0.9761919975280762
py> min(t2.repeat(repeat=6))
0.9494419097900391

So there is approximately 0.03 second difference per TWO MILLION
if...else blocks, or about 15 nanoseconds each. This is highly unlikely
to be the bottleneck in your code. Assuming the difference is real, and
not just measurement error, the difference is insignificant.

So, don't worry about which is faster. Write whichever is more natural,
easier to read and write.

> Block
> #----------------------------------
> if statemente_true:
> doSomething()
> else:
> doSomethingElseInstead()

This style is especially recommended when the two clauses are equal in
importance.

> versus this block:
> #----------------------------------
> if statement_true:
> doSomething()
> return
> doSomethingElseInstead()

This style is especially recommended when the doSomethingElseInstead()
block is the "normal" procedure, and the doSomething() block is a special
case. Not necessarily rare, but nevertheless special in some sense.

Of course, the decision as to which is the "special" case and which is
the "normal" case is often entirely arbitrary.



--
Steven

[Jan Riechers]

On 21.07.2012 12:06, Steven D'Aprano wrote:
>
> But in general, you're worrying too much about trivia. One way or the
> other, any speed difference will be trivial. Write whatever style reads
> and writes most naturally, and only worry about what's faster where it
> actually counts.
>

>

> Notice that I try to make each function do the same amount of work, so
> that we're seeing only the difference between "else" vs "no else".
>
> Now let's test the speed difference with Python 2.7. Because this is
> timing small code snippets, we should use the timeit module to time the
> code:
>
> from timeit import Timer
> setup = "from __main__ import with_else, without_else"
> t1 = Timer("for i in (0, 1): result = with_else(i)", setup)
> t2 = Timer("for i in (0, 1): result = without_else(i)", setup)
>
> Each snippet calls the function twice, once to take the if branch, then
> to take the else branch.
>
> Now we time how long it takes to run each code snippet 1000000 times. We
> do that six times each, and print the best (lowest) speed:
>
> py> min(t1.repeat(repeat=6))
> 0.9761919975280762
> py> min(t2.repeat(repeat=6))
> 0.9494419097900391
>
> So there is approximately 0.03 second difference per TWO MILLION
> if...else blocks, or about 15 nanoseconds each. This is highly unlikely
> to be the bottleneck in your code. Assuming the difference is real, and
> not just measurement error, the difference is insignificant.
>
> So, don't worry about which is faster. Write whichever is more natural,
> easier to read and write.
>
>

Hello Steven,

very nice example and thank you very much for also for the Timeit test!
Actually it confirms my assumption in some way:

[SNIP myself]

So if there is some overhead in some fashion in case we don't offer the
else, assuming the interpreter has to exit the evaluation of the
"if"-statement clause and return to a "normal parsing code"-state
outside the if statement itself.

[SNAP]

Without having looked at Andrew's bytecode excecution hint, using the
dis module, to see how the interpreter handles the task on lower level.

But fare enough for me :)

But I agree, the return in my example is misleading and it would be
illegal outside of a function call. I just added it to make clear that
the fellow code below the return should not be executed in comparison to
the 2nd example.

Thank you very much
Jan

[Thomas 'PointedEars' Lahn]

Jan Riechers wrote:

> I have one very basic question about speed,memory friendly coding, and
> coding style of the following easy "if"-statement in Python 2.7, but Im
> sure its also the same in Python 3.x
>
> Block
> #----------------------------------
> if statemente_true:
> doSomething()
> else:
> doSomethingElseInstead()
>
> #----------------------------------
>
> versus this block:
> #----------------------------------
> if statement_true:
> doSomething()
> return
>
> doSomethingElseInstead()
>
> #----------------------------------
>
>
> I understand the first pattern that I tell the interpreter to do:

A common misconception. As a writer of Python source code, (usually) you
never tell the (CPython) interpreter anything (but to start working on the
source code). Python source code is automatically *compiled* into bytecode
by the (CPython) interpreter, and that bytecode is executed by a virtual
machine.¹ So at most, you are telling that virtual machine to do something,
through the bytecode created from your source code.

> Check if the conditional is true, run "doSomething()" else go inside the
> else block and "doSomethingElseInstead()".
>
> while the 2nd does only checks:
> doSomething() if statement_true, if not, just go directly to
> "doSomethingElseInstead()
>
>
> Now, very briefly, what is the better way to proceed in terms of
> execution speed, readability, coding style?

Since this is comp.lang.python, you just need to check against the Zen of
Python to know what you should do ;-)

<http://www.python.org/dev/peps/pep-0020/>

For me, this boils down in this case to the common recommendation "return
early, return often" as "explicit is better than implicit" and "readability
counts". If there is nothing else than the `else' block in the function,
there is no use for you to continue in the function, so you should return
explicitly at this point.

On the other hand, if you can *avoid repeating code* in each branch by _not_
returning in the first branch, you should do that instead ("practicality
beats purity").

HTH

_____
¹ This is not unlike in other so-called "scripting languages"; although for
reasons that escape me, the software that compiles the source code – the
compiler – is called the (C)Python *interpreter*, even in
<http://docs.python.org/faq/general.html>.
--
PointedEars

Please do not Cc: me. / Bitte keine Kopien per E-Mail.

[Devin Jeanpierre]

On Sat, Jul 21, 2012 at 5:06 AM, Steven D'Aprano
<steve+comp....@pearwood.info> wrote:
> So there is approximately 0.03 second difference per TWO MILLION
> if...else blocks, or about 15 nanoseconds each. This is highly unlikely
> to be the bottleneck in your code. Assuming the difference is real, and
> not just measurement error, the difference is insignificant.

It's probably real. For if-else, the true case needs to make a jump
before it returns, but for if-return, there's no jump and the return
is inlined.

-- Devin

> So, don't worry about which is faster. Write whichever is more natural,
> easier to read and write.

The most important advice. Even when it's a larger difference! :)

-- Devin

[88888 Dihedral]

Jan Riechers於 2012年7月21日星期六UTC+8下午3時33分27秒寫道：

> Hello Pythonlist,
>
> I have one very basic question about speed,memory friendly coding, and

> coding style of the following easy &quot;if&quot;-statement in Python 2.7, but Im

> sure its also the same in Python 3.x
>
> Block
> #----------------------------------
> if statemente_true:

if an evaluated expression result is non-zero, then


> doSomething()

> else:
# execute this block if the expression evaluated as zero

> doSomethingElseInstead()
>
> #----------------------------------
>
> versus this block:
> #----------------------------------
> if statement_true:
> doSomething()
> return
>
> doSomethingElseInstead()
>
> #----------------------------------
>
>
> I understand the first pattern that I tell the interpreter to do:

> Check if the conditional is true, run &quot;doSomething()&quot; else go inside the
> else block and &quot;doSomethingElseInstead()&quot;.

>
> while the 2nd does only checks:
> doSomething() if statement_true, if not, just go directly to

> &quot;doSomethingElseInstead()

>
>
> Now, very briefly, what is the better way to proceed in terms of
> execution speed, readability, coding style?
> Letting out the fact that, in order to prevent

> &quot;doSomethingElseInstead&quot;-Block to execute, a return has to provided.

>
> Thank you for reading and hope someone brings light into that.
>
> Your fellow python programmer
> Jan

Well, the C-style branching is inherited in python.

Expressions and statements are different.

[88888 Dihedral]

Jan Riechers於 2012年7月21日星期六UTC+8下午3時33分27秒寫道：

> Hello Pythonlist,
>
> I have one very basic question about speed,memory friendly coding, and

> coding style of the following easy &quot;if&quot;-statement in Python 2.7, but Im

> sure its also the same in Python 3.x
>
> Block
> #----------------------------------
> if statemente_true:

if an evaluated expression result is non-zero, then


> doSomething()

> else:
# execute this block if the expression evaluated as zero

> doSomethingElseInstead()
>
> #----------------------------------
>
> versus this block:
> #----------------------------------
> if statement_true:
> doSomething()
> return
>
> doSomethingElseInstead()
>
> #----------------------------------
>
>
> I understand the first pattern that I tell the interpreter to do:

> Check if the conditional is true, run &quot;doSomething()&quot; else go inside the
> else block and &quot;doSomethingElseInstead()&quot;.

>
> while the 2nd does only checks:
> doSomething() if statement_true, if not, just go directly to

> &quot;doSomethingElseInstead()

>
>
> Now, very briefly, what is the better way to proceed in terms of
> execution speed, readability, coding style?
> Letting out the fact that, in order to prevent

> &quot;doSomethingElseInstead&quot;-Block to execute, a return has to provided.

>
> Thank you for reading and hope someone brings light into that.
>
> Your fellow python programmer
> Jan

[88888 Dihedral]

Chris Angelico於 2012年7月21日星期六UTC+8下午5時04分12秒寫道：
> On Sat, Jul 21, 2012 at 5:33 PM, Jan Riechers &lt;janp...@freenet.de&gt; wrote:
> &gt; Block
> &gt; #----------------------------------
> &gt; if statemente_true:
> &gt; doSomething()
> &gt; else:
> &gt; doSomethingElseInstead()
> &gt;
> &gt; #----------------------------------

>
> This means, to me, that the two options are peers - you do this or you do that.
>

> &gt; versus this block:
> &gt; #----------------------------------
> &gt; if statement_true:
> &gt; doSomething()
> &gt; return
> &gt;
> &gt; doSomethingElseInstead()
> &gt;
> &gt; #----------------------------------
>
> This would be for an early abort. Don&#39;t bother doing most of this
> function&#39;s work, just doSomething. Might be an error condition, or

> perhaps an optimized path.
>
> Definitely for error conditions, I would use the second option. The

> &quot;fail and bail&quot; notation keeps the entire error handling in one place:
>
> def func(x,y,z):
> if x&lt;0:
> y+=5
> return
> if y&lt;0:
> raise PEBKAC(&quot;There&#39;s an idiot here somewhere&quot;)

> # ... do the rest of the work
>

This is the caller responsible style when passing parameters to
functions.


Checking types of parameters both in the caller and the callee
does slow down a little bit.

> Note the similarity between the control structures. Raising an
> exception immediately terminates processing, without polluting the
> rest of the function with an unnecessary indentation level. Early
> aborting through normal function return can do the same thing.
>

> But this is purely a matter of style. I don&#39;t think there&#39;s any

> significance in terms of processing time or memory usage, and even if
> there is, it would be dwarfed by considerations of readability. Make

> your code look like what it&#39;s doing, and let the execution take care
> of itself.
>
> ChrisA

[88888 Dihedral]

Chris Angelico於 2012年7月21日星期六UTC+8下午5時04分12秒寫道：
> On Sat, Jul 21, 2012 at 5:33 PM, Jan Riechers &lt;janp...@freenet.de&gt; wrote:
> &gt; Block
> &gt; #----------------------------------
> &gt; if statemente_true:
> &gt; doSomething()
> &gt; else:
> &gt; doSomethingElseInstead()
> &gt;

> &gt; #----------------------------------

>
> This means, to me, that the two options are peers - you do this or you do that.
>

> &gt; versus this block:
> &gt; #----------------------------------
> &gt; if statement_true:
> &gt; doSomething()
> &gt; return
> &gt;
> &gt; doSomethingElseInstead()
> &gt;

> &gt; #----------------------------------
>
> This would be for an early abort. Don&#39;t bother doing most of this
> function&#39;s work, just doSomething. Might be an error condition, or

> perhaps an optimized path.
>
> Definitely for error conditions, I would use the second option. The

> &quot;fail and bail&quot; notation keeps the entire error handling in one place:
>
> def func(x,y,z):

> if x&lt;0:
> y+=5
> return
> if y&lt;0:

> raise PEBKAC(&quot;There&#39;s an idiot here somewhere&quot;)

> # ... do the rest of the work
>

This is the caller responsible style when passing parameters to
functions.


Checking types of parameters both in the caller and the callee
does slow down a little bit.

> Note the similarity between the control structures. Raising an
> exception immediately terminates processing, without polluting the
> rest of the function with an unnecessary indentation level. Early
> aborting through normal function return can do the same thing.
>

> But this is purely a matter of style. I don&#39;t think there&#39;s any

> significance in terms of processing time or memory usage, and even if
> there is, it would be dwarfed by considerations of readability. Make

> your code look like what it&#39;s doing, and let the execution take care
> of itself.
>
> ChrisA