May be worth discussing in here: https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei

On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
> May be worth discussing in here: https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei

I don't understand the use case. PR #5153 seems to be about getting the last element, not about some lookahead. When I think about lookahead, I always think of parsers. If the lexer provides an input range of tokens, my recursive descent parser might need 3 tokens lookahead, for example. For the lexer input a "push back a character into my input" is more convenient.

Should it give some performance guarantees? I guess, we don't want to refresh the buffers at every popFront, so it would probably be as lazy as possible.

On 3/2/17 7:51 AM, qznc wrote:
> On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
>> May be worth discussing in here:
>> https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei
>
> I don't understand the use case. PR #5153 seems to be about getting the
> last element, not about some lookahead. When I think about lookahead, I
> always think of parsers. If the lexer provides an input range of tokens,
> my recursive descent parser might need 3 tokens lookahead, for example.
> For the lexer input a "push back a character into my input" is more
> convenient.

Yah, parsing would be an obvious application. Others such as adjacentFind might also benefit.

> Should it give some performance guarantees? I guess, we don't want to
> refresh the buffers at every popFront, so it would probably be as lazy
> as possible.

Yah, a circular buffer should help here.


Andrei

On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
> May be worth discussing in here: https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei

//My understanding of the concept, is this correct?
assert(iota(5).lookAhead(1, 2).array ==
[   [0, 1],
    [0, 1, 2],
    [0, 1, 2, 3],
    [1, 2, 3, 4],
    [2, 3, 4],
    [3, 4],
    [4]
]);

I would prefer if the second argument meant length, not distance to look back. So the call equivalent to the one above would be lookAhead(-2, 4).

On Friday, 3 March 2017 at 11:22:37 UTC, Dukc wrote:
> [snip]

correction:
> assert(iota(5).lookAhead(1, 2).array ==
> [   [0, 1],
>     [0, 1, 2],
>     [0, 1, 2, 3],
>     [1, 2, 3, 4],
>     [2, 3, 4]
> ]);

On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
> May be worth discussing in here: https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei

Almost similar to this: https://github.com/dlang/phobos/pull/4027

Perhaps we should use that as base?

On Friday, 3 March 2017 at 11:56:26 UTC, Dukc wrote:
> On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
>> May be worth discussing in here: https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei
>
> Almost similar to this: https://github.com/dlang/phobos/pull/4027
>
> Perhaps we should use that as base?

My PR even included such a circular buffer until Andrei suggested to move it to a separate PR.
However, I think we are all misunderstanding  Andrei. He isn't referring to a buffer here - have a look at the issue.

On 3/3/17 6:22 AM, Dukc wrote:
> iota(5).lookAhead(1, 2)

should be iota(5).lookaround!(1, 2)

On 3/3/17 6:56 AM, Dukc wrote:
> On Thursday, 2 March 2017 at 01:44:19 UTC, Andrei Alexandrescu wrote:
>> May be worth discussing in here:
>> https://issues.dlang.org/show_bug.cgi?id=17238 -- Andrei
>
> Almost similar to this: https://github.com/dlang/phobos/pull/4027
>
> Perhaps we should use that as base?

Yes! cc @wilzbach