Created on 2020-08-27.00:00:00 last changed 12 months ago
Proposed resolution:
This wording is relative to N4885.
Edit [fs.filesystem.syn], header <filesystem> synopsis, as indicated:
[…] namespace std::filesystem { […] // [fs.dir.itr.nonmembers], range access for directory iterators directory_iterator begin(directory_iterator iter) noexcept; directory_iterator end(constdirectory_iterator&) noexcept; […] // [fs.rec.dir.itr.nonmembers], range access for recursive directory iterators recursive_directory_iterator begin(recursive_directory_iterator iter) noexcept; recursive_directory_iterator end(constrecursive_directory_iterator&) noexcept; […] } namespace std::ranges { template<> inline constexpr bool enable_borrowed_range<filesystem::directory_iterator> = true; template<> inline constexpr bool enable_borrowed_range<filesystem::recursive_directory_iterator> = true; template<> inline constexpr bool enable_view<filesystem::directory_iterator> = true; template<> inline constexpr bool enable_view<filesystem::recursive_directory_iterator> = true; }
Edit [fs.dir.itr.nonmembers] as indicated:
-1- These functions enable range access for directory_iterator.
directory_iterator begin(directory_iterator iter) noexcept;-2- Returns: iter.
directory_iterator end(constdirectory_iterator&) noexcept;-3- Returns: directory_iterator().
Edit [fs.rec.dir.itr.nonmembers] as indicated:
-1- These functions enable use of recursive_directory_iterator with range-based for statements.
recursive_directory_iterator begin(recursive_directory_iterator iter) noexcept;-2- Returns: iter.
recursive_directory_iterator end(constrecursive_directory_iterator&) noexcept;-3- Returns: recursive_directory_iterator().
[ 2021-10-14 Approved at October 2021 virtual plenary. Status changed: Voting → WP. ]
[ 2021-06-23; Reflector poll ]
Set status to Tentatively Ready after seven votes in favour during reflector poll.
[ 2021-05-17, Tim provides wording ]
Both MSVC and libstdc++'s end already take its argument by value, so the wording below just does that. Any discussion about changing or removing the poison pills is probably better suited for a paper.
[ 2021-02-22, Barry Revzin comments ]
When we do make whichever of the alternative adjustments necessary such that range<directory_iterator> is true, we should also remember to specialize enable_borrowed_range for both types to be true (since the iterator is the range, this is kind of trivially true).
[ 2020-09-06; Reflector prioritization ]
Set priority to 3 during reflector discussions.
std::filesystem::directory_iterator and std::filesystem::recursive_directory_iterator are intended to be ranges, but both fail to satisfy the concept std::ranges::range.
They both opt in to being a range the same way, via non-member functions:directory_iterator begin(directory_iterator iter) noexcept; directory_iterator end(const directory_iterator&) noexcept; recursive_directory_iterator begin(recursive_directory_iterator iter) noexcept; recursive_directory_iterator end(const recursive_directory_iterator&) noexcept;
This is good enough for a range-based for statement, but for the range concept, non-member end is looked up in a context that includes ([range.access.end]/2.6) the declarations:
void end(auto&) = delete; void end(const auto&) = delete;
Which means that non-const directory_iterator and non-const recursive_directory_iterator, the void end(auto&) overload ends up being a better match and thus the CPO ranges::end doesn't find a candidate. Which means that {recursive_,}directory_iterator is not a range, even though const {recursive_,}directory_iterator is a range.
This could be fixed by having the non-member end for both of these types just take by value (as libstdc++ currently does anyway) or by adding member functions begin() const and end() const. A broader direction would be to consider removing the poison pill overloads. Their motivation from P0970 was to support what are now called borrowed ranges — but that design now is based on specializing a variable template instead of providing a non-member begin that takes an rvalue, so the initial motivation simply no longer exists. And, in this particular case, causes harm.History | |||
---|---|---|---|
Date | User | Action | Args |
2023-11-22 15:47:43 | admin | set | status: wp -> c++23 |
2021-10-14 09:56:08 | admin | set | messages: + msg12120 |
2021-10-14 09:56:08 | admin | set | status: voting -> wp |
2021-09-29 12:57:28 | admin | set | status: ready -> voting |
2021-06-23 14:16:45 | admin | set | messages: + msg11957 |
2021-06-23 14:16:45 | admin | set | status: new -> ready |
2021-05-18 09:45:04 | admin | set | messages: + msg11815 |
2021-05-18 09:45:04 | admin | set | messages: + msg11814 |
2021-02-22 17:56:51 | admin | set | messages: + msg11694 |
2020-09-06 13:07:36 | admin | set | messages: + msg11473 |
2020-08-27 00:00:00 | admin | create |