[Python-Dev] timsort for jython

Tim Peters tim.one@comcast.net
Fri, 09 Aug 2002 16:33:55 -0400

> OTOH, I do expect that once code relies on stability, we'll
> have about as much chance of taking that away as getting rid of
> list.append().

[Patrick K. O'Brien]
> There you go again! Your flip comment has got me thinking about the "one
> best idiom" for list appending.

It was a serious enough comment, judged against the universe of all comments
I make <wink>.  I expect that if a hypothetical 3x-faster non-stable sort
algorithm got discovered for 2.6, we wouldn't be able to call it list.sort()
then.  It's very hard to take any perceived goodness away, ever.

> So I'll ask the question. Is there a reason to want to get rid of
> list.append()?

I believe-- and sincerely hope --that I'm the only one who ever suggested

> How does one decide between list.append() and augmented assignment (+=),
> such as:
> >>> l = []
> >>> l.append('something')
> >>> l.append('else')
> >>> l += ['to']
> >>> l += ['consider']
> >>> l
> ['something', 'else', 'to', 'consider']
> >>>

Clarity:  l.append() is obvious; I'd never append a single item via +=.
More, I'd probably do

    push = L.append

outside a loop and call push('something') inside the loop.  "+=" as a
synonym for list.extend() is only interesting if you're writing polymorphic
code that wants to exploit the ability to define __iadd__.  For sane people,
that's approximately never <wink>.

> And wasn't someone documenting current idioms in light of recent Python
> features? Did that ever get posted anywhere?

Rings a bell, but beats me.