Re: the Result Type
On 12/08/2013 12:24 AM, seany wrote:
std.algorithm.splitter seems to return all its return values as a type
Result, without quotes, and i dont not seem to be able to cast it to
string[] or int[] with cast(string[]) ( or even cast (string) - i tried
that too).
I tried to use a function
void function(T, R)(T arr, out R output)
{
foreach(elem; arr)
{
output ~= elemM
}
}
i have an occasion where R is int[], and one where it is string[]; and
but it says, that int can not be appended to string[], and string can
not be appended to int[], and compilation fails.
Nonetheless removing either of the occasions, solves it.
how to deal with this Result type?
Many other algorithms return Result, which are independent from each
other. :) (Lookup Voldemort types.)
Such types are just lazy ranges. When you eagerly need an actual array
of the elements, call std.array.array on Result:
import std.array;
import std.algorithm;
void main()
{
auto input = hello world;
auto splittedWords = input.splitter(' ').array;
static assert(is (typeof(splittedWords) == string[]));
}
Ali
Re: the Result Type
O_O with that knowledge, would also be possible to define new types (not aliases, but new encapsulated types) representing things such as Graph, Ring, Topology, surreal number, etc? I dont find this in your book, would you consider either adding this Volodemrot types, or in case they already are there, give me a link (both the introductions in d wiki and dr. doobs blog was very short)
Re: the Result Type
Am Sun, 08 Dec 2013 09:24:53 +0100
schrieb seany se...@uni-bonn.de:
std.algorithm.splitter seems to return all its return values as a
type Result, without quotes, and i dont not seem to be able to
cast it to string[] or int[] with cast(string[]) ( or even cast
(string) - i tried that too).
I tried to use a function
void function(T, R)(T arr, out R output)
{
foreach(elem; arr)
{
output ~= elemM
}
}
i have an occasion where R is int[], and one where it is
string[]; and
but it says, that int can not be appended to string[], and string
can not be appended to int[], and compilation fails.
Nonetheless removing either of the occasions, solves it.
how to deal with this Result type?
That confused me as a beginner as well until I understood that
these Result types are actually just structs. They are
generated at compile-time and share commonly known methods
like .front or .popFront() that identify them as ranges.
Arrays are a sub-set of ranges. The algorithms in D typically
take a range as input and a range as output. Where possible
they try not to go over the whole input at once, but only as
needed. This is called lazy evaluation and is one reason you
don't get returned a complete array. In some cases the input
might even be infinite: Try returning an array for splitter
over a random number generator as input range!
So what do you do with a Range struct? There are 3 options:
* If you really want an array you can call the function
array() on the Result:
http://dlang.org/phobos/std_array.html#.array
* If you want to further process your Result with other
algorithms, just chain them together. E.g.:
arr.splitter(abc).sort.uniq();
* If you want to iterate over the Result, foreach works on
those Result ranges:
foreach (elem; arr.splitter(abc)) {...}
The range concept makes it easy to apply any kind of algorithm
on stuff that can be represented as a consecutive items of
the same type. And it allows those algorithms to only do as
much work as needed to get the result. A good example is
http://dlang.org/phobos/std_range.html#take , which sets a
limit to the number of items to be used from the input.
So you can have an infinite random number generator, but only
take the first 50 numbers like so: randomGen.take(50);
By the way File(myfile).byLine() gives you a range over the
lines of text of myfile. That's quite handy if you have an
array of strings like a dictionary in a text file.
--
Marco
Re: the Result Type
Am Sun, 08 Dec 2013 09:59:55 +0100
schrieb seany se...@uni-bonn.de:
O_O
with that knowledge, would also be possible to define new types
(not aliases, but new encapsulated types) representing things
such as Graph, Ring, Topology, surreal number, etc?
All these Result types are simply structs. Structs with
template parameters. Your question is a bit vague. You will
not be able to use D algorithms on anything but consecutive
items of the same type as I worte in my other post, if that
is the question.
Other than that the Result types are just:
struct Result(T)
{
...
}
You can call them templated struct or generic type. Many
languages offer this in some way including C++, Java, Delphi
and C#. If you have used any of those programming languages
before you are probably already familiar with the concept.
In the most simple form you can use them to create type
specialized containers, graphs or this:
struct Matrix(T, size_t width, size_t height)
if (isNumeric!T)
{
...
}
(A matrix with a fixed width and height that can be used if T
is a built-in integer or floating point type.)
--
Marco
Re: the Result Type
On 12/08/2013 12:59 AM, seany wrote: I dont find this in your book, would you consider either adding this Volodemrot types, I think at least a short mention is in order. :) There are two reasons why they don't appear in the book (yet): 1) They are not a proper language feature, rather a happy discovery. (I remember Andrei's original newsgroup post about this discovery but I cannot find it at this time.) 2) They were discovered after I started writing the book. Ali
Re: the Result Type
Ali Çehreli:
When you eagerly need an actual array of the elements, call
std.array.array on Result:
import std.array;
import std.algorithm;
void main()
{
auto input = hello world;
auto splittedWords = input.splitter(' ').array;
Or just use the eager split() function.
Bye,
bearophile
Re: the Result Type
On Sunday, 8 December 2013 at 11:02:40 UTC, Ali Çehreli wrote: (I remember Andrei's original newsgroup post about this discovery but I cannot find it at this time.) http://www.digitalmars.com/d/archives/digitalmars/D/announce/Voldemort_Types_in_D_23511.html this?
Re: the Result Type
On 12/08/2013 03:41 AM, seany wrote: On Sunday, 8 December 2013 at 11:02:40 UTC, Ali Çehreli wrote: (I remember Andrei's original newsgroup post about this discovery but I cannot find it at this time.) http://www.digitalmars.com/d/archives/digitalmars/D/announce/Voldemort_Types_in_D_23511.html this? There was an earlier newsgroup post like look what I've just discovered. Ali
Re: the Result Type
On Sun, Dec 8, 2013 at 9:59 AM, seany se...@uni-bonn.de wrote: O_O with that knowledge, would also be possible to define new types (not aliases, but new encapsulated types) representing things such as Graph, Ring, Topology, surreal number, etc? Other posters already answered your questions concerning ranges. In short, you can define internal structs (or classes, or functions) inside a function, another struct, a class, and return them. They can be parameterized on the host type template parameters, if any. All the different types in std.algorithm and std.range have *nothing* in common (no common base class, nor are they part of a common datatype), *except* they all respect the range 'concept'. As long as something (class or struct) has the three basic range methods (empty / front / popFront), it can be used with other functions in std.algorithm and std.range. Returning to you question, now. You can do exactly with other types what was done with ranges in the standard library. As long as you define the basic properties of a Graph / Ring / Whatever, and associate it with a guard template (see std.range.isInputRange), any type can be used as a Graph. Maybe you could give us more details on what you want to realize exactly?
