The (incorrect) idea that Python's "lists" are not contiguous memory arrays
seems to be somewhat widespread. I'm not sure where it comes from – maybe
they used have a different implementation, or maybe its just the
unfortunate insistence on calling them lists rather than arrays.


On Tue, Aug 5, 2014 at 11:57 PM, John Myles White <[email protected]>
wrote:

> For future reference, Jason’s interpretation of Python lists was
> completely accurate:
> https://docs.python.org/2/faq/design.html#how-are-lists-implemented
>
>  — John
>
> On Aug 5, 2014, at 4:08 AM, Neal Becker <[email protected]> wrote:
>
> > Thanks.  I don't think that's really equivalent though.  I believe
> python's list
> > is not contiguous, and can efficiently handle insertions and deletions,
> both at
> > the ends and in the middle.  Julia's array is a contiguous dense array,
> correct?
> >
> > Jason Merrill wrote:
> >
> >>  julia> {}
> >>  0-element Array{Any,1}
> >>
> >> is like python's arrays in that it can hold values of heterogenous
> types,
> >> and you can grow it with push!, e.g.
> >>
> >>  julia> a = {}
> >>  0-element Array{Any,1}
> >>
> >>  julia> push!(a, "grapes")
> >>  1-element Array{Any,1}:
> >>   "grapes"
> >>
> >>  julia> push!(a, 3)
> >>  2-element Array{Any,1}:
> >>    "grapes"
> >>    3
> >>
> >> If you know the type of all the things you'll be putting in your
> container,
> >> you will get better performance/memory efficiency with a typed array.
> >>
> >> julia> b = Float64[]
> >> 0-element Array{Float64,1}
> >>
> >> julia> push!(b, 1.0)
> >> 1-element Array{Float64,1}:
> >> 1.0
> >>
> >> julia> push!(b, NaN)
> >> 2-element Array{Float64,1}:
> >>   1.0
> >> NaN
> >>
> >> I think in either case, Julia is smart about growing the array by the
> right
> >> amount at the right time to get O(1) ammortized push!(), but if you
> have a
> >> guess about the eventual size of your array, you can help Julia out with
> >> `sizehint`.
> >>
> >> julia> sizehint(b, 100000)
> >>
> >> You don't have to be exactly right with sizehint--the worst that will
> >> happen is that you will use a little more memory than you needed to, or
> the
> >> array will have to be dynamically grown more often than it had to be.
> For
> >> this, I am most grateful. In Matlab, your choices are, AFAICT, 1) know
> the
> >> exact size of your container correctly ahead of time, 2) have terrible
> >> performance, or 3) allocate something that's definitely bigger than you
> >> need, and then keep track of how much of it you've used yourself and
> throw
> >> away the excess at the end.
> >>
> >>
> >> On Monday, August 4, 2014 4:16:27 PM UTC-7, Neal Becker wrote:
> >>>
> >>> What would be similar to python list type?  A container that has an
> >>> efficient
> >>> append.  A common use is construct an empty list, then grow it by
> calling
> >>> append.
> >>>
> >>>
> >
> >
>
>

Reply via email to