On Mon, 17 May 2010 17:01:22 -0400, Andrei Alexandrescu
<seewebsiteforem...@erdani.org> wrote:
On 05/17/2010 03:16 PM, Steven Schveighoffer wrote:
Comparing splitByOneOf(str, "; ")) to splitter(str, set(';', ' ')), I
see one major difference here -- "; " is a literal, set(';', ' ') is
not.
I would expect that 'set' as a generic set type would implement it's
guts as some sort of tree/hash, which means a lot of overhead for a
simple argument. With the literal version, the notation is in the
function, not the type. While it seems rather neat, the overhead should
be considered.
A compromise:
foreach(x; splitter(str, either("; ")))
Which can be implemented without heap activity.
These are good points. They have gone through my mind as well, but
lately I've started to take a somewhat more liberal view of containers.
For example, I'm thinking that Set!T (which would be the type returned
by set()) could automatically use arrays and linear search for small
sets. Other special cases come to mind, such as the small array
optimization. In other words Set!T would not have a guaranteed
implementation, but instead exploit magnitude to choose among a spectrum
of implementation alternatives.
These are good ideas.
Of course using a different name such as either() is even better for the
implementation because it can make additional implementation dictated by
the restricted use of either(). For example either() could return a
FixedSet!T that does not accept adding new members and is optimized
accordingly.
FixedSet!T would be easily implemented as a sorted array for any number of
members (unsorted for under some threshold number of members).
My point was simply that all input parameters besides the string literal
require heap activity to maintain the set. Much less if you allocate one
array, but for something like "; ", I would like to see zero heap activity
:)
-Steve
