Created on 2014-02-19.00:00:00 last changed 20 months ago
Proposed resolution:
This wording is relative to N4944.
Change [istreambuf.iterator.general] as indicated:
constexpr istreambuf_iterator() noexcept; constexpr istreambuf_iterator(default_sentinel_t) noexcept; istreambuf_iterator(const istreambuf_iterator&) noexcept = default; ~istreambuf_iterator() = default; istreambuf_iterator(istream_type& s) noexcept
;: istreambuf_iterator(s.rdbuf()) { } istreambuf_iterator(streambuf_type* s) noexcept; istreambuf_iterator(const proxy& p) noexcept; … private: streambuf_type* sbuf_; // exposition only int_type c_{}; // exposition only }; }
Change [istreambuf.iterator.cons] as indicated:
For each istreambuf_iterator constructor in this section, an end-of-stream iterator is constructed if
and only ifthe exposition-only member sbuf_ is initialized with a null pointer value or if sbuf_->sgetc() returns traits_type::eof().constexpr istreambuf_iterator() noexcept; constexpr istreambuf_iterator(default_sentinel_t) noexcept;-1- Effects: Initializes sbuf_ with nullptr.
istreambuf_iterator(istream_type& s) noexcept;
-2- Effects: Initializes sbuf_ with s.rdbuf().istreambuf_iterator(streambuf_type* s) noexcept;[Drafting note: sgetc() can throw, but this function is noexcept. Should it swallow exceptions and create an end-of-stream iterator, to avoid weakening the exception spec of an existing function?]
-3- Effects: Initializes sbuf_ with s. If s is not null, initializes c_ with s->sgetc(). Sets sbuf_ to null if sgetc exits via an exception, or if traits_type::eq_int_type(c_, traits_type::eof()) is true.
istreambuf_iterator(const proxy& p) noexcept;-4- Effects: Initializes sbuf_ with p.sbuf_. If p.sbuf_ is not null, initializes c_ with p.keep_.
Change [istreambuf.iterator.ops] as indicated:
charT operator*() const;-?- Preconditions: sbuf_ is not null.
-1- Returns:
The character obtained via the streambuf member sbuf_->sgetc().traits_type::to_char_type(c_).-?- Throws: Nothing.
istreambuf_iterator& operator++();-?- Preconditions: sbuf_ is not null.
-2- Effects:
As if by sbuf_->sbumpc().Performs c_ = sbuf_->snextc(), then sets sbuf_ to null if traits_type::eq_int_type(c_, traits_type::eof()) is true.-3- Returns: *this.
proxy operator++(int);-4-
Returns: proxy(sbuf_->sbumpc(), sbuf_).
Effects: Equivalent to:proxy p(**this, sbuf_); ++*this; return p;
bool equal(const istreambuf_iterator& b) const;-5- Returns: bool(sbuf_) == bool(b.sbuf_).
[Note: This is true if and only if both iterators are at end-of-stream, or neither is at end-of-stream, regardless of what streambuf object they use. — end note]
template<class charT, class traits> bool operator==(const istreambuf_iterator<charT, traits>& a, const istreambuf_iterator<charT, traits>& b);-6- Returns: a.equal(b).
bool equal(const istreambuf_iterator& i, default_sentinel_t s) const;-7- Returns:
i.equal(s)i.sbuf_ == nullptr.
[ 2023-04-12; Jonathan adds wording ]
[ 2023-03-31; Jonathan Wakely comments ]
I agree that they should compare equal, but that's in conflict with the
resolution of LWG 2544, which says that begin
must not be at end-of-stream because &buf
is not null.
Given the following code,
#include <sstream> std::stringbuf buf; std::istreambuf_iterator<char> begin(&buf); std::istreambuf_iterator<char> end;
it is not clear from the wording of the Standard whether begin.equal(end) must be true. In at least one implementation it is not (CC: Sun C++ 5.10 SunOS_sparc Patch 128228-25 2013/02/20) and in at least one implementation it is (gcc version 4.3.2 x86_64-unknown-linux-gnu).
[istreambuf.iterator] says that end is an end-of-stream iterator since it was default constructed. It also says that an iterator becomes equal to an end-of-stream iterator when end of stream is reached by sgetc() having returned eof(). [istreambuf.iterator::equal] says that equal() returns true iff both iterators are end of stream or not end of stream. But there seems to be no requirement that equal check for end-of-stream by calling sgetc(). Jiahan Zi at BloombergLP discovered this issue through his code failing to work correctly. Dietmar Kühl has opined in a private communication that the iterators should compare equal.History | |||
---|---|---|---|
Date | User | Action | Args |
2023-04-12 15:07:41 | admin | set | messages: + msg13518 |
2023-04-12 15:07:41 | admin | set | messages: + msg13517 |
2023-03-31 15:53:58 | admin | set | messages: + msg13504 |
2014-02-19 00:00:00 | admin | create |