how portable is end iterator decrement?
Solution 1:
I think this is the relevant clause:
ISO/IEC 14882:2003 C++ Standard 23.1.1/12 – Sequences
Table 68 lists sequence operations that are provided for some types of sequential containers but not others. An implementation shall provide these operations for all container types shown in the "container" column, and shall implement them so as to take amortized constant time.
+----------------------------------------------------------------------------+ | Table 68 | +--------------+-----------------+---------------------+---------------------+ | expression | return type | operational | container | | | | semantics | | +--------------+-----------------+---------------------+---------------------+ | a.front() | reference; | *a.begin() | vector, list, deque | | | const_reference | | | | | for constant a | | | +--------------+-----------------+---------------------+---------------------+ | a.back() | reference; | *--a.end() | vector, list, deque | | | const_reference | | | | | for constant a | | | .............................................................................. . . . . . . . . . . .............................................................................. | a.pop_back() | void | a.erase(--a.end()) | vector, list, deque | .............................................................................. . . . . . . . . . .
So for the containers listed, not only should the iterator returned from end()
be decrementable, the decremented iterator should also be dereferencable. (Unless the container is empty, of course. That invokes undefined behavior.)
In fact, vector
, list
and deque
implementations that came with the Visual C++ compiler does it exactly like the table. Of course, that's not to imply that every compiler does it like this:
// From VC++'s <list> implementation
reference back()
{ // return last element of mutable sequence
return (*(--end()));
}
const_reference back() const
{ // return last element of nonmutable sequence
return (*(--end()));
}
Note about the code in the table:
ISO/IEC 14882:2003 C++ Standard 17.3.1.2/6 – Requirements
In some cases the semantic requirements are presented as C + + code. Such code is intended as a specification of equivalence of a construct to another construct, not necessarily as the way the construct must be implemented.
So while it's true that an implementation may not implement those expressions in terms of begin()
and end()
, the C++ standard specifies that the two expressions are equivalent. In other words, a.back()
and *--a.end()
are equivalent constructs according to the above clause. It seems to me that it means that you should be able to replace every instance of a.back()
with *--a.end()
and vice-versa and have the code still work.
According to Bo Persson, the revision of the C++ standard that I have on hand has a defect with respect to Table 68.
Proposed resolution:
Change the specification in table 68 "Optional Sequence Operations" in 23.1.1/12 for "a.back()" from
*--a.end()
to
{ iterator tmp = a.end(); --tmp; return *tmp; }
and the specification for "a.pop_back()" from
a.erase(--a.end())
to
{ iterator tmp = a.end(); --tmp; a.erase(tmp); }
It appears that you can still decrement the iterator returned from end()
and dereference the decremented iterator, as long as it's not a temporary.
Solution 2:
from the documentation for std::prev
Although the expression --c.end() often compiles, it is not guaranteed to do so: c.end() is an rvalue expression, and there is no iterator requirement that specifies that decrement of an rvalue is guaranteed to work. In particular, when iterators are implemented as pointers, --c.end() does not compile, while std::prev(c.end()) does.
which means the implementation for the prefix decrement operation may not be the inside class form iterator iterator::operator--(int)
but overloaded outside class form iterator operator--(iterator&, int)
.
so you should either prefer std::prev
or do the following:
{ auto end = a.end(); --end; };