C++ Logo

STD-PROPOSALS

Advanced search

Subject: Re: [std-proposals] count-based for loop
From: Sebastian Büttner (sebastian.buettner_at_[hidden])
Date: 2019-08-31 21:36:35


Hi Sophia,

the proposal got rejected in Jacksonville 2016.

There were lots of concerns regarding breaking brace-balancing with this
proposal for the half open range syntax: [min..max). This seemed to have
been a concern regarding tools, compilers and in general for programmers
since they are not used to such syntax. Even though this syntax is known
in mathematical context it was understood that this might not be so true
for programmers.
A library based solution has the additional benefit of adding a name to
the range generator allowing for a more clear expression of the
programmers intent even if he/she or anyone who has to read the code is
unfamiliar with the mathematical syntax. One possibility that came up was:

     for (auto i : index_range(0, 5))

Additionally ranges were on their way and many commitee members argued
against a language solution and in favor of a library solution since it
could be done already easily with ranges.

     Sebastian

On 29.08.19 23:23, Sophia Poirier via Std-Proposals wrote:
> Interesting, thanks Andrew.  Do you recall and would mind sharing the
> reasons for its rejection?
>
> thanks,
> Sophia
>
>
>> On Aug 29, 2019, at 10:21 AM, Andrew Tomazos <andrewtomazos_at_[hidden]
>> <mailto:andrewtomazos_at_[hidden]>> wrote:
>>
>> I plan to target C++23 with it.
>>
>> I proposed a similar core language feature in 2016 in P0286:
>>
>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0268r0.pdf
>>
>> but the proposal was rejected.
>>
>>
>> On Fri, Aug 30, 2019 at 2:46 AM Sophia Poirier <spoirier_at_[hidden]
>> <mailto:spoirier_at_[hidden]>> wrote:
>>
>> Thanks Andrew, that's a great collection of utilities.  Is that a
>> proposal you are working towards submitting?  (I note it has no
>> paper number.)
>>
>> I do prefer to see it language level as I think it is intuitive
>> as I have suggested, plus it is such a basic and fundamental
>> logical operation (and allows serving those who cannot use the
>> STL), however I do recognize there are fewer hurdles to landing
>> library features.
>>
>> - Sophia
>>
>>
>>> On Aug 28, 2019, at 5:23 AM, Andrew Tomazos
>>> <andrewtomazos_at_[hidden] <mailto:andrewtomazos_at_[hidden]>> wrote:
>>>
>>> See:
>>>
>>> https://docs.google.com/document/d/1gBdBualdIU1bpgW_El4GT-p0he9Yr7D52xmLmHes5Qo/edit?usp=sharing
>>>
>>> On Wed, Aug 28, 2019 at 3:42 AM Sophia Poirier via Std-Proposals
>>> <std-proposals_at_[hidden]
>>> <mailto:std-proposals_at_[hidden]>> wrote:
>>>
>>>> On Aug 27, 2019, at 10:26 AM, Barry Revzin
>>>> <barry.revzin_at_[hidden] <mailto:barry.revzin_at_[hidden]>> wrote:
>>>>
>>>> On Tue, Aug 27, 2019, 12:02 PM Sophia Poirier via
>>>> Std-Proposals <std-proposals_at_[hidden]
>>>> <mailto:std-proposals_at_[hidden]>> wrote:
>>>>
>>>> Thanks, yes I have that as an alternate example in my
>>>> longer notes.  My understanding is that it still
>>>> suffers from the type-matching problem as traditional
>>>> for loops. Your example of:
>>>>
>>>> for (const auto i : std::views::iota(0, count))
>>>>
>>>> only works when count is an int. Otherwise, if for
>>>> example count is uint32_t, it would need to be:
>>>>
>>>> for (const auto i : std::views::iota(uint32_t{0}, count)
>>>> or:
>>>> for (const auto i : std::views::iota(0u, count))
>>>> or:
>>>> for (const auto i : std::views::iota<uint32_t>(0, count))
>>>> or:
>>>> for (const auto i :
>>>> std::views::iota<decltype(count)>(0, count))
>>>>
>>>> or something along those lines, or you will get
>>>> template instantiation failure compiler error.  I think
>>>> that if std::views::iota had a constructor overload
>>>> that was simply the second argument (count) with
>>>> implicit zero start, then it would be a good option. 
>>>> However I believe there is interest to reserve such an
>>>> overload perhaps for infinite ranges?
>>>>
>>>> thanks,
>>>> Sophia
>>>>
>>>>
>>>> This is true. But we can write a helper function to get the
>>>> correct type of 0 so we don't need the ugliness at point of
>>>> use:
>>>>
>>>> template <std::integral T>
>>>> auto upto(T n) {
>>>> return views::iota(T{0}, n);
>>>> }
>>>>
>>>> We end up with:
>>>>
>>>> for (const auto i : upto(count))
>>>>
>>>> Barry
>>>>
>>>
>>> This could be the basis of an alternate library proposal, true.
>>>
>>> - Sophia
>>>
>



STD-PROPOSALS list run by herb.sutter at gmail.com

Standard Proposals Archives on Google Groups