| Doc. no. | N1515 = 03-0098 |
| Date: | 20 Sep 2003 |
| Project: | Programming Language C++ |
| Reply to: | Matt Austern <austern@apple.com> |
Reference ISO/IEC IS 14882:1998(E)
Also see:
The purpose of this document is to record the status of issues which have come before the Library Working Group (LWG) of the ANSI (J16) and ISO (WG21) C++ Standards Committee. Issues represent potential defects in the ISO/IEC IS 14882:1998(E) document. Issues are not to be used to request new features or other extensions.
This document contains only library issues which are actively being considered by the Library Working Group. That is, issues which have a status of New, Open, Ready, and Review. See Library Defect Reports List for issues considered defects and Library Closed Issues List for issues considered closed.
The issues in these lists are not necessarily formal ISO Defect Reports (DR's). While some issues will eventually be elevated to official Defect Report status, other issues will be disposed of in other ways. See Issue Status.
This document is in an experimental format designed for both viewing via a world-wide web browser and hard-copy printing. It is available as an HTML file for browsing or PDF file for printing.
Prior to Revision 14, library issues lists existed in two slightly different versions; a Committee Version and a Public Version. Beginning with Revision 14 the two versions were combined into a single version.
This document includes [bracketed italicized notes] as a reminder to the LWG of current progress on issues. Such notes are strictly unofficial and should be read with caution as they may be incomplete or incorrect. Be aware that LWG support for a particular resolution can quickly change if new viewpoints or killer examples are presented in subsequent discussions.
For the most current official version of this document see http://www.dkuug.dk/jtc1/sc22/wg21. Requests for further information about this document should include the document number above, reference ISO/IEC 14882:1998(E), and be submitted to Information Technology Industry Council (ITI), 1250 Eye Street NW, Washington, DC 20005.
Public information as to how to obtain a copy of the C++ Standard, join the standards committee, submit an issue, or comment on an issue can be found in the comp.std.c++ FAQ. Public discussion of C++ Standard related issues occurs on news:comp.std.c++.
For committee members, files available on the committee's private web site include the HTML version of the Standard itself. HTML hyperlinks from this issues list to those files will only work for committee members who have downloaded them into the same disk directory as the issues list files.
New - The issue has not yet been reviewed by the LWG. Any Proposed Resolution is purely a suggestion from the issue submitter, and should not be construed as the view of LWG.
Open - The LWG has discussed the issue but is not yet ready to move the issue forward. There are several possible reasons for open status:
A Proposed Resolution for an open issue is still not be construed as the view of LWG. Comments on the current state of discussions are often given at the end of open issues in an italic font. Such comments are for information only and should not be given undue importance.
Dup - The LWG has reached consensus that the issue is a duplicate of another issue, and will not be further dealt with. A Rationale identifies the duplicated issue's issue number.
NAD - The LWG has reached consensus that the issue is not a defect in the Standard, and the issue is ready to forward to the full committee as a proposed record of response. A Rationale discusses the LWG's reasoning.
Review - Exact wording of a Proposed Resolution is now available for review on an issue for which the LWG previously reached informal consensus.
Ready - The LWG has reached consensus that the issue is a defect in the Standard, the Proposed Resolution is correct, and the issue is ready to forward to the full committee for further action as a Defect Report (DR).
DR - (Defect Report) - The full J16 committee has voted to forward the issue to the Project Editor to be processed as a Potential Defect Report. The Project Editor reviews the issue, and then forwards it to the WG21 Convenor, who returns it to the full committee for final disposition. This issues list accords the status of DR to all these Defect Reports regardless of where they are in that process.
TC - (Technical Corrigenda) - The full WG21 committee has voted to accept the Defect Report's Proposed Resolution as a Technical Corrigenda. Action on this issue is thus complete and no further action is possible under ISO rules.
WP - (Working Paper) - The proposed resolution has not been accepted as a Technical Corrigendum, but the full WG21 committee has voted to apply the Defect Report's Proposed Resolution to the working paper.
RR - (Record of Response) - The full WG21 committee has determined that this issue is not a defect in the Standard. Action on this issue is thus complete and no further action is possible under ISO rules.
Future - In addition to the regular status, the LWG believes that this issue should be revisited at the next revision of the standard. It is usually paired with NAD.
Issues are always given the status of New when they first appear on the issues list. They may progress to Open or Review while the LWG is actively working on them. When the LWG has reached consensus on the disposition of an issue, the status will then change to Dup, NAD, or Ready as appropriate. Once the full J16 committee votes to forward Ready issues to the Project Editor, they are given the status of Defect Report ( DR). These in turn may become the basis for Technical Corrigenda (TC), or are closed without action other than a Record of Response (RR ). The intent of this LWG process is that only issues which are truly defects in the Standard move to the formal ISO DR status.
Section: 22.2.2.1.2 [lib.facet.num.get.virtuals] Status: Open Submitter: Nathan Myers Date: 6 Aug 1998
The current description of numeric input does not account for the possibility of overflow. This is an implicit result of changing the description to rely on the definition of scanf() (which fails to report overflow), and conflicts with the documented behavior of traditional and current implementations.
Users expect, when reading a character sequence that results in a value unrepresentable in the specified type, to have an error reported. The standard as written does not permit this.
Further comments from Dietmar:
I don't feel comfortable with the proposed resolution to issue 23: It kind of simplifies the issue to much. Here is what is going on:
Currently, the behavior of numeric overflow is rather counter intuitive and hard to trace, so I will describe it briefly:
Further discussion from Redmond:
The basic problem is that we've defined our behavior, including our error-reporting behavior, in terms of C90. However, C90's method of reporting overflow in scanf is not technically an "input error". The strto_* functions are more precise.
There was general consensus that failbit should be set upon overflow. We considered three options based on this:
Straw poll: (1) 5; (2) 0; (3) 8.
Further discussion from Santa Cruz:
There was some discussion of what the intent of our error reporting mechanism was. There was general agreement on the following principles:
The crux of the disagreement was that some people, but not all, believed that the design was also based on a fourth principle: whenever converstion fails and failbit is set, nothing is to be extracted and the value of the variable being extracted into is guaranteed to be unchanged.
Some people believe that upon overflow, an implementation should "extract" a special value that allows the user to tell that it was overflow instead of some other kind of error. Straw poll: 1 person believed the standard should require that, 2 thought it should forbid it, and 6 thought the standard should allow but not require it.
Proposed resolution:
typo: 22.2.2.2.2 [lib.facet.num.put.virtuals], para 2, bullet 3. Strike "in." from the end.
Change 22.2.2.2 [lib.locale.nm.put], para 11, bullet 2 from:
The sequence of chars accumulated in stage 2 would have caused scanf to report an input failure. ios_base::failbit is assigned to err.
to:
The sequence of chars accumulated in stage 2 would have caused scanf to report an input failure or to store a value outside the range representable by val. ios_base::failbit is assigned to err.
[PJP provided wording. this treats overflow or underflow the same as an ill-formed field. It's not exactly the consensus from Santa Cruz, but he thinks it's the simplest and most robust rule and that it corresponds to widespread common practice.]
Section: 27 [lib.input.output] Status: Review Submitter: Nathan Myers Date: 6 Aug 1998
Many of the specifications for iostreams specify that character values or their int_type equivalents are compared using operators == or !=, though in other places traits::eq() or traits::eq_int_type is specified to be used throughout. This is an inconsistency; we should change uses of == and != to use the traits members instead.
Proposed resolution:
[Pre-Kona: Dietmar supplied wording]
List of changes to clause 27:
fillch == fill()to
traits::eq(fillch, fill())
c == delim for the next available input character cto
traits::eq(c, delim) for the next available input character c
c == delim for the next available input character cto
traits::eq(c, delim) for the next available input character c
c == delim for the next available input character cto
traits::eq(c, delim) for the next available input character c
c == delim for the next available input character cto
traits::eq_int_type(c, delim) for the next available input character c
The last condition will never occur if delim == traits::eof()to
The last condition will never occur if traits::eq_int_type(delim, traits::eof()).
while ((c = is.rdbuf()->snextc()) != traits::eof()) {to
while (!traits::eq_int_type(c = is.rdbuf()->snextc(), traits::eof())) {
List of changes to Chapter 21:
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
at(xpos+I) == str.at(I) for all elements ...to
traits::eq(at(xpos+I), str.at(I)) for all elements ...
c == delim for the next available input character cto
traits::eq(c, delim) for the next available input character c
Notes:
Section: 25 [lib.algorithms] Status: Review Submitter: Nico Josuttis Date: 29 Sep 1998
The standard does not state, how often a function object is copied, called, or the order of calls inside an algorithm. This may lead to surprising/buggy behavior. Consider the following example:
class Nth { // function object that returns true for the nth element
private:
int nth; // element to return true for
int count; // element counter
public:
Nth (int n) : nth(n), count(0) {
}
bool operator() (int) {
return ++count == nth;
}
};
....
// remove third element
list<int>::iterator pos;
pos = remove_if(coll.begin(),coll.end(), // range
Nth(3)), // remove criterion
coll.erase(pos,coll.end());
This call, in fact removes the 3rd AND the 6th element. This happens because the usual implementation of the algorithm copies the function object internally:
template <class ForwIter, class Predicate>
ForwIter std::remove_if(ForwIter beg, ForwIter end, Predicate op)
{
beg = find_if(beg, end, op);
if (beg == end) {
return beg;
}
else {
ForwIter next = beg;
return remove_copy_if(++next, end, beg, op);
}
}
The algorithm uses find_if() to find the first element that should be removed. However, it then uses a copy of the passed function object to process the resulting elements (if any). Here, Nth is used again and removes also the sixth element. This behavior compromises the advantage of function objects being able to have a state. Without any cost it could be avoided (just implement it directly instead of calling find_if()).
Proposed resolution:
Add a new paragraph following 25 [lib.algorithms] paragraph 8:
[Note: Unless otherwise specified, algorithms that take function objects as arguments are permitted to copy those function objects freely. Programmers for whom object identity is important should consider using a wrapper class that points to a noncopied implementation object, or some equivalent solution.]
[Dublin: Pete Becker felt that this may not be a defect, but rather something that programmers need to be educated about. There was discussion of adding wording to the effect that the number and order of calls to function objects, including predicates, not affect the behavior of the function object.]
[Pre-Kona: Nico comments: It seems the problem is that we don't have a clear statement of "predicate" in the standard. People including me seemed to think "a function returning a Boolean value and being able to be called by an STL algorithm or be used as sorting criterion or ... is a predicate". But a predicate has more requirements: It should never change its behavior due to a call or being copied. IMHO we have to state this in the standard. If you like, see section 8.1.4 of my library book for a detailed discussion.]
[Kona: Nico will provide wording to the effect that "unless otherwise specified, the number of copies of and calls to function objects by algorithms is unspecified". Consider placing in 25 [lib.algorithms] after paragraph 9.]
[Santa Cruz: The standard doesn't currently guarantee that functions object won't be copied, and what isn't forbidden is allowed. It is believed (especially since implementations that were written in concert with the standard do make copies of function objects) that this was intentional. Thus, no normative change is needed. What we should put in is a non-normative note suggesting to programmers that if they want to guarantee the lack of copying they should use something like the ref wrapper.]
[Oxford: Matt provided wording.]
Section: 23.2.5 [lib.vector.bool] Status: Open Submitter: AFNOR Date: 7 Oct 1998
vector<bool> is not a container as its reference and pointer types are not references and pointers.
Also it forces everyone to have a space optimization instead of a speed one.
See also: 99-0008 == N1185 Vector<bool> is Nonconforming, Forces Optimization Choice.
Proposed resolution:
[In Santa Cruz the LWG felt that this was Not A Defect.]
[In Dublin many present felt that failure to meet Container requirements was a defect. There was disagreement as to whether or not the optimization requirements constituted a defect.]
[The LWG looked at the following resolutions in some detail:
* Not A Defect.
* Add a note explaining that vector<bool> does not meet
Container requirements.
* Remove vector<bool>.
* Add a new category of container requirements which
vector<bool> would meet.
* Rename vector<bool>.
No alternative had strong, wide-spread, support and every alternative
had at least one "over my dead body" response.
There was also mention of a transition scheme something like (1) add
vector_bool and deprecate vector<bool> in the next standard. (2)
Remove vector<bool> in the following standard.]
[Modifying container requirements to permit returning proxies (thus allowing container requirements conforming vector<bool>) was also discussed.]
[It was also noted that there is a partial but ugly workaround in that vector<bool> may be further specialized with a customer allocator.]
[Kona: Herb Sutter presented his paper J16/99-0035==WG21/N1211, vector<bool>: More Problems, Better Solutions. Much discussion of a two step approach: a) deprecate, b) provide replacement under a new name. LWG straw vote on that: 1-favor, 11-could live with, 2-over my dead body. This resolution was mentioned in the LWG report to the full committee, where several additional committee members indicated over-my-dead-body positions.]
[Tokyo: Not discussed by the full LWG; no one claimed new insights and so time was more productively spent on other issues. In private discussions it was asserted that requirements for any solution include 1) Increasing the full committee's understanding of the problem, and 2) providing compiler vendors, authors, teachers, and of course users with specific suggestions as to how to apply the eventual solution.]
Section: 24.1.1 [lib.input.iterators] Status: Review Submitter: AFNOR Date: 7 Oct 1998
Table 72 in 24.1.1 [lib.input.iterators] specifies semantics for *r++ of:
{ T tmp = *r; ++r; return tmp; }
There are two problems with this. First, the return type is specified to be "T", as opposed to something like "convertible to T". This is too specific: we want to allow *r++ to return an lvalue.
Second, writing the semantics in terms of code misleadingly suggests that the effects *r++ should precisely replicate the behavior of this code, including side effects. (Does this mean that *r++ should invoke the copy constructor exactly as many times as the sample code above would?) See issue 334 for a similar problem.
Proposed resolution:
In Table 72 in 24.1.1 [lib.input.iterators], change the return type for *r++ from T to "convertible to T".Rationale:
This issue has two parts: the return type, and the number of times the copy constructor is invoked.
The LWG believes the the first part is a real issue. It's inappropriate for the return type to be specified so much more precisely for *r++ than it is for *r. In particular, if r is of (say) type int*, then *r++ isn't int, but int&.
The LWG does not believe that the number of times the copy constructor is invoked is a real issue. This can vary in any case, because of language rules on copy constructor elision. That's too much to read into these semantics clauses.
Section: 17.4.3.1 [lib.reserved.names] Status: Review Submitter: Judy Ward Date: 15 Dec 1998
The original issue asked whether a library implementor could specialize standard library templates for built-in types. (This was an issue because users are permitted to explicitly instantiate standard library templates.)
Specializations are no longer a problem, because of the resolution to core issue 259. Under the proposed resolution, it will be legal for a translation unit to contain both a specialization and an explicit instantiation of the same template, provided that the specialization comes first. In such a case, the explicit instantiation will be ignored. Further discussion of library issue 120 assumes that the core 259 resolution will be adopted.
However, as noted in lib-7047, one piece of this issue still remains: what happens if a standard library implementor explicitly instantiates a standard library templates? It's illegal for a program to contain two different explicit instantiations of the same template for the same type in two different translation units (ODR violation), and the core working group doesn't believe it is practical to relax that restriction.
The issue, then, is: are users allowed to explicitly instantiate standard library templates for non-user defined types? The status quo answer is 'yes'. Changing it to 'no' would give library implementors more freedom.
This is an issue because, for performance reasons, library implementors often need to explicitly instantiate standard library templates. (for example, std::basic_string<char>) Does giving users freedom to explicitly instantiate standard library templates for non-user defined types make it impossible or painfully difficult for library implementors to do this?
John Spicer suggests, in lib-8957, that library implementors have a mechanism they can use for explicit instantiations that doesn't prevent users from performing their own explicit instantiations: put each explicit instantiation in its own object file. (Different solutions might be necessary for Unix DSOs or MS-Windows DLLs.) On some platforms, library implementors might not need to do anything special: the "undefined behavior" that results from having two different explicit instantiations might be harmless.
Proposed resolution:
Append to 17.4.3.1 [lib.reserved.names] paragraph 1:
A program may explicitly instantiate any templates in the standard library only if the declaration depends on a user-defined name of external linkage and the instantiation meets the standard library requirements for the original template.
Rationale:
The LWG considered another possible resolution:
In light of the resolution to core issue 259, no normative changes in the library clauses are necessary. Add the following non-normative note to the end of 17.4.3.1 [lib.reserved.names] paragraph 1:
[Note: A program may explicitly instantiate standard library templates, even when an explicit instantiation does not depend on a user-defined name. --end note]
The LWG rejected this because it was believed that it would make it unnecessarily difficult for library implementors to write high-quality implementations. A program may not include an explicit instantiation of the same template, for the same template arguments, in two different translation units. If users are allowed to provide explicit instantiations of Standard Library templates for built-in types, then library implementors aren't, at least not without nonportable tricks.
The most serious problem is a class template that has writeable static member variables. Unfortunately, such class templates are important and, in existing Standard Library implementations, are often explicitly specialized by library implementors: locale facets, which have a writeable static member variable id. If a user's explicit instantiation collided with the implementations explicit instantiation, iostream initialization could cause locales to be constructed in an inconsistent state.
One proposed implementation technique was for Standard Library implementors to provide explicit instantiations in separate object files, so that they would not be picked up by the linker when the user also provides an explicit instantiation. However, this technique only applies for Standard Library implementations that are packaged as static archives. Most Standard Library implementations nowadays are packaged as dynamic libraries, so this technique would not apply.
The Committee is now considering standardization of dynamic linking. If there are such changes in the future, it may be appropriate to revisit this issue later.
Section: 27.6.2.5.4 [lib.ostream.inserters.character] Status: Review Submitter: Dietmar Kühl Date: 20 Jul 1999
Paragraph 4 states that the length is determined using traits::length(s). Unfortunately, this function is not defined for example if the character type is wchar_t and the type of s is char const*. Similar problems exist if the character type is char and the type of s is either signed char const* or unsigned char const*.
Proposed resolution:
Change 27.6.2.5.4 [lib.ostream.inserters.character] paragraph 4 from:
Effects: Behaves like an formatted inserter (as described in lib.ostream.formatted.reqmts) of out. After a sentry object is constructed it inserts characters. The number of characters starting at s to be inserted is traits::length(s). Padding is determined as described in lib.facet.num.put.virtuals. The traits::length(s) characters starting at s are widened using out.widen (lib.basic.ios.members). The widened characters and any required padding are inserted into out. Calls width(0).
to:
Effects: Behaves like an formatted inserter (as described in lib.ostream.formatted.reqmts) of out. After a sentry object is constructed it inserts n characters starting at s, where n is:
- traits::length(s) for the overload where the first argument is of type basic_ostream<charT, traits>& and the second is of type const charT*, and also for the overload where the first argument is of type basic_ostream<char, traits>& and the second is of type const char*.
- std::char_traits<char>::length(s) for the overload where the first argument is of type basic_ostream<charT, traits>& and the second is of type const char*.
- traits::length(reinterpret_cast<const char*>(s)) for the other two overloads.
Padding is determined as described in lib.facet.num.put.virtuals. The n characters starting at s are widened using out.widen (lib.basic.ios.members). The widened characters and any required padding are inserted into out. Calls width(0).
[Santa Cruz: Matt supplied new wording]
Rationale:
We have five separate cases. In two of them we can use the user-supplied traits class without any fuss. In the other three we try to use something as close to that user-supplied class as possible. In two cases we've got a traits class that's appropriate for char and what we've got is a const signed char* or a const unsigned char*; that's close enough so we can just use a reinterpret cast, and continue to use the user-supplied traits class. Finally, there's one case where we just have to give up: where we've got a traits class for some arbitrary charT type, and we somehow have to deal with a const char*. There's nothing better to do but fall back to char_traits<char>
Section: 20.1.5 [lib.allocator.requirements], 23.1 [lib.container.requirements] Status: Open Submitter: Andy Sawyer Date: 21 Oct 1999
Must the value returned by max_size() be unchanged from call to call?
Must the value returned from max_size() be meaningful?
Possible meanings identified in lib-6827:
1) The largest container the implementation can support given "best
case" conditions - i.e. assume the run-time platform is "configured to
the max", and no overhead from the program itself. This may possibly
be determined at the point the library is written, but certainly no
later than compile time.
2) The largest container the program could create, given "best case"
conditions - i.e. same platform assumptions as (1), but take into
account any overhead for executing the program itself. (or, roughly
"storage=storage-sizeof(program)"). This does NOT include any resource
allocated by the program. This may (or may not) be determinable at
compile time.
3) The largest container the current execution of the program could
create, given knowledge of the actual run-time platform, but again,
not taking into account any currently allocated resource. This is
probably best determined at program start-up.
4) The largest container the current execution program could create at
the point max_size() is called (or more correctly at the point
max_size() returns :-), given it's current environment (i.e. taking
into account the actual currently available resources). This,
obviously, has to be determined dynamically each time max_size() is
called.
Proposed resolution:
Change 20.1.5 [lib.allocator.requirements] table 32 max_size() wording from:
the largest value that can meaningfully be
passed to X::allocate
to:
the value of the largest constant expression
(5.19 [expr.const]) that could ever meaningfully be passed to X::allocate
Change 23.1 [lib.container.requirements] table 65 max_size() wording from:
size() of the largest possible container.
to:
the value of the largest constant expression
(5.19 [expr.const]) that could ever meaningfully be returned by X::size().
[Kona: The LWG informally discussed this and asked Andy Sawyer to submit an issue.]
[Tokyo: The LWG believes (1) above is the intended meaning.]
[Post-Tokyo: Beman Dawes supplied the above resolution at the request of the LWG. 21.3.3 [lib.string.capacity] was not changed because it references max_size() in 23.1. The term "compile-time" was avoided because it is not defined anywhere in the standard (even though it is used several places in the library clauses).]
[Copenhagen: Exactly what max_size means is still unclear. It may have a different meaning as a container member function than as an allocator member function. For the latter, it is probably best thought of as an architectural limit. Nathan will provide new wording.]
Section: 18.2.1 [lib.limits] Status: Open Submitter: Stephen Cleary Date: 21 Dec 1999
In some places in this section, the terms "fundamental types" and "scalar types" are used when the term "arithmetic types" is intended. The current usage is incorrect because void is a fundamental type and pointers are scalar types, neither of which should have specializations of numeric_limits.
Proposed resolution:
Change 18.2 [lib.support.limits] para 1 from:
The headers <limits>, <climits>, and <cfloat> supply characteristics of implementation-dependent fundamental types (3.9.1).
to:
The headers <limits>, <climits>, and <cfloat> supply characteristics of implementation-dependent arithmetic types (3.9.1).
Change 18.2.1 [lib.limits] para 1 from:
The numeric_limits component provides a C++ program with information about various properties of the implementation's representation of the fundamental types.
to:
The numeric_limits component provides a C++ program with information about various properties of the implementation's representation of the arithmetic types.
Change 18.2.1 [lib.limits] para 2 from:
Specializations shall be provided for each fundamental type. . .
to:
Specializations shall be provided for each arithmetic type. . .
Change 18.2.1 [lib.limits] para 4 from:
Non-fundamental standard types. . .
to:
Non-arithmetic standard types. . .
Change 18.2.1.1 [lib.numeric.limits] para 1 from:
The member is_specialized makes it possible to distinguish between fundamental types, which have specializations, and non-scalar types, which do not.
to:
The member is_specialized makes it possible to distinguish between arithmetic types, which have specializations, and non-arithmetic types, which do not.
[post-Toronto: The opinion of the LWG is that the wording in the standard, as well as the wording of the proposed resolution, is flawed. The term "arithmetic types" is well defined in C and C++, and it is not clear that the term is being used correctly. It is also not clear that the term "implementation dependent" has any useful meaning in this context. The biggest problem is that numeric_limits seems to be intended both for built-in types and for user-defined types, and the standard doesn't make it clear how numeric_limits applies to each of those cases. A wholesale review of numeric_limits is needed. A paper would be welcome.]
Section: 17.4.3.1 [lib.reserved.names] Status: Review Submitter: Dave Abrahams Date: 01 Apr 2000
The issues are:
1. How can a 3rd party library implementor (lib1) write a version of a standard algorithm which is specialized to work with his own class template?
2. How can another library implementor (lib2) write a generic algorithm which will take advantage of the specialized algorithm in lib1?
This appears to be the only viable answer under current language rules:
namespace lib1 { // arbitrary-precision numbers using T as a basic unit template <class T> class big_num { //... };// defining this in namespace std is illegal (it would be an // overload), so we hope users will rely on Koenig lookup template <class T> void swap(big_int<T>&, big_int<T>&); }#include <algorithm> namespace lib2 { template <class T> void generic_sort(T* start, T* end) { ... // using-declaration required so we can work on built-in types using std::swap; // use Koenig lookup to find specialized algorithm if available swap(*x, *y); } }
This answer has some drawbacks. First of all, it makes writing lib2 difficult and somewhat slippery. The implementor needs to remember to write the using-declaration, or generic_sort will fail to compile when T is a built-in type. The second drawback is that the use of this style in lib2 effectively "reserves" names in any namespace which defines types which may eventually be used with lib2. This may seem innocuous at first when applied to names like swap, but consider more ambiguous names like unique_copy() instead. It is easy to imagine the user wanting to define these names differently in his own namespace. A definition with semantics incompatible with the standard library could cause serious problems (see issue 225).
Why, you may ask, can't we just partially specialize std::swap()? It's because the language doesn't allow for partial specialization of function templates. If you write:
namespace std
{
template <class T>
void swap(lib1::big_int<T>&, lib1::big_int<T>&);
}
You have just overloaded std::swap, which is illegal under the current language rules. On the other hand, the following full specialization is legal:
namespace std
{
template <>
void swap(lib1::other_type&, lib1::other_type&);
}
This issue reflects concerns raised by the "Namespace issue with specialized swap" thread on comp.lang.c++.moderated. A similar set of concerns was earlier raised on the boost.org mailing list and the ACCU-general mailing list. Also see library reflector message c++std-lib-7354.
J. C. van Winkel points out (in c++std-lib-9565) another unexpected fact: it's impossible to output a container of std::pair's using copy and an ostream_iterator, as long as both pair-members are built-in or std:: types. That's because a user-defined operator<< for (for example) std::pair<const std::string, int> will not be found: lookup for operator<< will be performed only in namespace std. Opinions differed on whether or not this was a defect, and, if so, whether the defect is that something is wrong with user-defined functionality and std, or whether it's that the standard library does not provide an operator<< for std::pair<>.
Proposed resolution:
Adopt the wording proposed in Howard Hinnant's paper N1439=03-0021, "Proposed Resolution To LWG issues 225, 226, 229".
[Tokyo: Summary, "There is no conforming way to extend std::swap for user defined templates." The LWG agrees that there is a problem. Would like more information before proceeding. This may be a core issue. Core issue 229 has been opened to discuss the core aspects of this problem. It was also noted that submissions regarding this issue have been received from several sources, but too late to be integrated into the issues list. ]
[Post-Tokyo: A paper with several proposed resolutions, J16/00-0029==WG21/N1252, "Shades of namespace std functions " by Alan Griffiths, is in the Post-Tokyo mailing. It should be considered a part of this issue.]
[Toronto: Dave Abrahams and Peter Dimov have proposed a resolution that involves core changes: it would add partial specialization of function template. The Core Working Group is reluctant to add partial specialization of function templates. It is viewed as a large change, CWG believes that proposal presented leaves some syntactic issues unanswered; if the CWG does add partial specialization of function templates, it wishes to develop its own proposal. The LWG continues to believe that there is a serious problem: there is no good way for users to force the library to use user specializations of generic standard library functions, and in certain cases (e.g. transcendental functions called by valarray and complex) this is important. Koenig lookup isn't adequate, since names within the library must be qualified with std (see issue 225), specialization doesn't work (we don't have partial specialization of function templates), and users aren't permitted to add overloads within namespace std. ]
[Copenhagen: Discussed at length, with no consensus. Relevant papers in the pre-Copenhagen mailing: N1289, N1295, N1296. Discussion focused on four options. (1) Relax restrictions on overloads within namespace std. (2) Mandate that the standard library use unqualified calls for swap and possibly other functions. (3) Introduce helper class templates for swap and possibly other functions. (4) Introduce partial specialization of function templates. Every option had both support and opposition. Straw poll (first number is support, second is strongly opposed): (1) 6, 4; (2) 6, 7; (3) 3, 8; (4) 4, 4.]
[Redmond: Discussed, again no consensus. Herb presented an argument that a user who is defining a type T with an associated swap should not be expected to put that swap in namespace std, either by overloading or by partial specialization. The argument is that swap is part of T's interface, and thus should to in the same namespace as T and only in that namespace. If we accept this argument, the consequence is that standard library functions should use unqualified call of swap. (And which other functions? Any?) A small group (Nathan, Howard, Jeremy, Dave, Matt, Walter, Marc) will try to put together a proposal before the next meeting.]
[Curaçao: An LWG-subgroup spent an afternoon working on issues 225, 226, and 229. Their conclusion was that the issues should be separated into an LWG portion (Howard's paper, N1387=02-0045), and a EWG portion (Dave will write a proposal). The LWG and EWG had (separate) discussions of this plan the next day. The proposed resolution is the one proposed by Howard.]
[Santa Cruz: the LWG agreed with the general direction of Howard's paper, N1387. (Roughly: Koenig lookup is disabled unless we say otherwise; this issue is about when we do say otherwise.) However, there were concerns about wording. Howard will provide new wording. Bill and Jeremy will review it.]
[Oxford: Howard proposed the new wording.]
Rationale:
Informally: introduce a Swappable concept, and specify that the value types of the iterators passed to certain standard algorithms (such as iter_swap, swap_ranges, reverse, rotate, and sort) conform to that concept. The Swappable concept will make it clear that these algorithms use unqualified lookup for the calls to swap. Also, in 26.3.3.3 [lib.valarray.transcend] paragraph 1, state that the valarray transcendentals use unqualified lookup.
Section: 23.1.2 [lib.associative.reqmts] Status: Open Submitter: Andrew Koenig Date: 30 Apr 2000
If mm is a multimap and p is an iterator into the multimap, then mm.insert(p, x) inserts x into mm with p as a hint as to where it should go. Table 69 claims that the execution time is amortized constant if the insert winds up taking place adjacent to p, but does not say when, if ever, this is guaranteed to happen. All it says it that p is a hint as to where to insert.
The question is whether there is any guarantee about the relationship between p and the insertion point, and, if so, what it is.
I believe the present state is that there is no guarantee: The user can supply p, and the implementation is allowed to disregard it entirely.
Additional comments from Nathan:
The vote [in Redmond] was on whether to elaborately specify the use of
the hint, or to require behavior only if the value could be inserted
adjacent to the hint. I would like to ensure that we have a chance to
vote for a deterministic treatment: "before, if possible, otherwise
after, otherwise anywhere appropriate", as an alternative to the
proposed "before or after, if possible, otherwise [...]".
Proposed resolution:
In table 69 "Associative Container Requirements" in 23.1.2 [lib.associative.reqmts], in the row for a.insert(p, t), change
iterator p is a hint pointing to where the insert should start to search.
to
insertion adjacent to iterator p is preferred if more than one insertion point is valid.
and change
logarithmic in general, but amortized constant if t is inserted right after p.
to
logarithmic in general, but amortized constant if t is inserted adjacent to iterator p.
[Toronto: there was general agreement that this is a real defect: when inserting an element x into a multiset that already contains several copies of x, there is no way to know whether the hint will be used. The proposed resolution was that the new element should always be inserted as close to the hint as possible. So, for example, if there is a subsequence of equivalent values, then providing a.begin() as the hint means that the new element should be inserted before the subsequence even if a.begin() is far away. JC van Winkel supplied precise wording for this proposed resolution, and also for an alternative resolution in which hints are only used when they are adjacent to the insertion point.]
[Copenhagen: the LWG agreed to the original proposed resolution, in which an insertion hint would be used even when it is far from the insertion point. This was contingent on seeing a reference implementation showing that it is possible to implement this requirement without loss of efficiency. John Potter provided such a reference implementation.]
[Redmond: The LWG was reluctant to adopt the proposal that emerged from Copenhagen: it seemed excessively complicated, and went beyond fixing the defect that we identified in Toronto. PJP provided the new wording described in this issue. Nathan agrees that we shouldn't adopt the more detailed semantics, and notes: "we know that you can do it efficiently enough with a red-black tree, but there are other (perhaps better) balanced tree techniques that might differ enough to make the detailed semantics hard to satisfy."]
[Curaçao: Nathan should give us the alternative wording he suggests so the LWG can decide between the two options.]
Section: 23.2.4.3 [lib.vector.modifiers] Status: Open Submitter: Lisa Lippincott Date: 06 June 2000
Paragraph 2 of 23.2.4.3 [lib.vector.modifiers] describes the complexity of vector::insert:
Complexity: If first and last are forward iterators, bidirectional iterators, or random access iterators, the complexity is linear in the number of elements in the range [first, last) plus the distance to the end of the vector. If they are input iterators, the complexity is proportional to the number of elements in the range [first, last) times the distance to the end of the vector.
First, this fails to address the non-iterator forms of insert.
Second, the complexity for input iterators misses an edge case -- it requires that an arbitrary number of elements can be added at the end of a vector in constant time.
At the risk of strengthening the requirement, I suggest simply
Complexity: The complexity is linear in the number of elements inserted plus the distance to the end of the vector.
For input iterators, one may achieve this complexity by first inserting at the end of the vector, and then using rotate.
I looked to see if deque had a similar problem, and was surprised to find that deque places no requirement on the complexity of inserting multiple elements (23.2.1.3 [lib.deque.modifiers], paragraph 3):
Complexity: In the worst case, inserting a single element into a deque takes time linear in the minimum of the distance from the insertion point to the beginning of the deque and the distance from the insertion point to the end of the deque. Inserting a single element either at the beginning or end of a deque always takes constant time and causes a single call to the copy constructor of T.
I suggest:
Complexity: The complexity is linear in the number of elements inserted plus the shorter of the distances to the beginning and end of the deque. Inserting a single element at either the beginning or the end of a deque causes a single call to the copy constructor of T.
Proposed resolution:
[Toronto: It's agreed that there is a defect in complexity of multi-element insert for vector and deque. For vector, the complexity should probably be something along the lines of c1 * N + c2 * distance(i, end()). However, there is some concern about whether it is reasonable to amortize away the copies that we get from a reallocation whenever we exceed the vector's capacity. For deque, the situation is somewhat less clear. Deque is notoriously complicated, and we may not want to impose complexity requirements that would imply any implementation technique more complicated than a while loop whose body is a single-element insert.]
Section: 26.3.2.1 [lib.valarray.cons], 26.3.2.2 [lib.valarray.assign] Status: Review Submitter: Robert Klarer Date: 31 Jul 2000
This discussion is adapted from message c++std-lib-7056 posted November 11, 1999. I don't think that anyone can reasonably claim that the problem described below is NAD.
These valarray constructors can never be called:
template <class T>
valarray<T>::valarray(const slice_array<T> &);
template <class T>
valarray<T>::valarray(const gslice_array<T> &);
template <class T>
valarray<T>::valarray(const mask_array<T> &);
template <class T>
valarray<T>::valarray(const indirect_array<T> &);
Similarly, these valarray assignment operators cannot be called:
template <class T>
valarray<T> valarray<T>::operator=(const slice_array<T> &);
template <class T>
valarray<T> valarray<T>::operator=(const gslice_array<T> &);
template <class T>
valarray<T> valarray<T>::operator=(const mask_array<T> &);
template <class T>
valarray<T> valarray<T>::operator=(const indirect_array<T> &);
Please consider the following example:
#include <valarray>
using namespace std;
int main()
{
valarray<double> va1(12);
valarray<double> va2(va1[slice(1,4,3)]); // line 1
}
Since the valarray va1 is non-const, the result of the sub-expression va1[slice(1,4,3)] at line 1 is an rvalue of type const std::slice_array<double>. This slice_array rvalue is then used to construct va2. The constructor that is used to construct va2 is declared like this:
template <class T>
valarray<T>::valarray(const slice_array<T> &);
Notice the constructor's const reference parameter. When the constructor is called, a slice_array must be bound to this reference. The rules for binding an rvalue to a const reference are in 8.5.3, paragraph 5 (see also 13.3.3.1.4). Specifically, paragraph 5 indicates that a second slice_array rvalue is constructed (in this case copy-constructed) from the first one; it is this second rvalue that is bound to the reference parameter. Paragraph 5 also requires that the constructor that is used for this purpose be callable, regardless of whether the second rvalue is elided. The copy-constructor in this case is not callable, however, because it is private. Therefore, the compiler should report an error.
Since slice_arrays are always rvalues, the valarray constructor that has a parameter of type const slice_array<T> & can never be called. The same reasoning applies to the three other constructors and the four assignment operators that are listed at the beginning of this post. Furthermore, since these functions cannot be called, the valarray helper classes are almost entirely useless.
Proposed resolution:
slice_array:
gslice_array:
mask_array:
indirect_array:
[Proposed resolution was modified in Santa Cruz: explicitly make copy constructor and copy assignment operators public, instead of removing them.]
Rationale:
Keeping the valarray constructors private is untenable. Merely making valarray a friend of the helper classes isn't good enough, because access to the copy constructor is checked in the user's environment.
Making the assignment operator public is not strictly necessary to solve this problem. A majority of the LWG (straw poll: 13-4) believed we should make the assignment operators public, in addition to the copy constructors, for reasons of symmetry and user expectation.
Section: 20.1.5 [lib.allocator.requirements] Status: Open Submitter: Matt Austern Date: 22 Aug 2000
From lib-7752:
I've been assuming (and probably everyone else has been assuming) that allocator instances have a particular property, and I don't think that property can be deduced from anything in Table 32.
I think we have to assume that allocator type conversion is a homomorphism. That is, if x1 and x2 are of type X, where X::value_type is T, and if type Y is X::template rebind<U>::other, then Y(x1) == Y(x2) if and only if x1 == x2.
Further discussion: Howard Hinnant writes, in lib-7757:
I think I can prove that this is not provable by Table 32. And I agree it needs to be true except for the "and only if". If x1 != x2, I see no reason why it can't be true that Y(x1) == Y(x2). Admittedly I can't think of a practical instance where this would happen, or be valuable. But I also don't see a need to add that extra restriction. I think we only need:
if (x1 == x2) then Y(x1) == Y(x2)
If we decide that == on allocators is transitive, then I think I can prove the above. But I don't think == is necessarily transitive on allocators. That is:
Given x1 == x2 and x2 == x3, this does not mean x1 == x3.
Example:
x1 can deallocate pointers from: x1, x2, x3
x2 can deallocate pointers from: x1, x2, x4
x3 can deallocate pointers from: x1, x3
x4 can deallocate pointers from: x2, x4x1 == x2, and x2 == x4, but x1 != x4
Proposed resolution:
[Toronto: LWG members offered multiple opinions. One opinion is that it should not be required that x1 == x2 implies Y(x1) == Y(x2), and that it should not even be required that X(x1) == x1. Another opinion is that the second line from the bottom in table 32 already implies the desired property. This issue should be considered in light of other issues related to allocator instances.]
Section: 24.4.1 [lib.reverse.iterators] Status: Open Submitter: Steve Cleary Date: 27 Nov 2000
This came from an email from Steve Cleary to Fergus in reference to issue 179. The library working group briefly discussed this in Toronto and believed it should be a separate issue. There was also some reservations about whether this was a worthwhile problem to fix.
Steve said: "Fixing reverse_iterator. std::reverse_iterator can (and should) be changed to preserve these additional requirements." He also said in email that it can be done without breaking user's code: "If you take a look at my suggested solution, reverse_iterator doesn't have to take two parameters; there is no danger of breaking existing code, except someone taking the address of one of the reverse_iterator global operator functions, and I have to doubt if anyone has ever done that. . . But, just in case they have, you can leave the old global functions in as well -- they won't interfere with the two-template-argument functions. With that, I don't see how any user code could break."
Proposed resolution:
Section: 24.4.1.1 [lib.reverse.iterator] add/change the following declarations:
A) Add a templated assignment operator, after the same manner
as the templated copy constructor, i.e.:
template < class U >
reverse_iterator < Iterator >& operator=(const reverse_iterator< U >& u);
B) Make all global functions (except the operator+) have
two template parameters instead of one, that is, for
operator ==, !=, <, >, <=, >=, - replace:
template < class Iterator >
typename reverse_iterator< Iterator >::difference_type operator-(
const reverse_iterator< Iterator >& x,
const reverse_iterator< Iterator >& y);
with:
template < class Iterator1, class Iterator2 >
typename reverse_iterator < Iterator1 >::difference_type operator-(
const reverse_iterator < Iterator1 > & x,
const reverse_iterator < Iterator2 > & y);
Also make the addition/changes for these signatures in 24.4.1.3 [lib.reverse.iter.ops].
[ Copenhagen: The LWG is concerned that the proposed resolution introduces new overloads. Experience shows that introducing overloads is always risky, and that it would be inappropriate to make this change without implementation experience. It may be desirable to provide this feature in a different way. ]
Section: 25.2.4 [lib.alg.replace] Status: Review Submitter: Martin Sebor Date: 15 Dec 2000
(revision of the further discussion) There are a number of problems with the requires clauses for the algorithms in 25.1 and 25.2. The requires clause of each algorithm should describe the necessary and sufficient requirements on the inputs to the algorithm such that the algorithm compiles and runs properly. Many of the requires clauses fail to do this. Here is a summary of the kinds of mistakes:
Here is the list of algorithms that contain mistakes:
Also, in the requirements for EqualityComparable, the requirement that the operator be defined for const objects is lacking.
Proposed resolution:
20.1.1 Change p1 from
In Table 28, T is a type to be supplied by a C++ program instantiating a template, a, b, and c are values of type T.
to
In Table 28, T is a type to be supplied by a C++ program instantiating a template, a, b, and c are values of type const T.
25 Between p8 and p9
Add the following sentence:
When the description of an algorithm gives an expression such as *first == value for a condition, it is required that the expression evaluate to either true or false in boolean contexts.
25.1.2 Change p1 by deleting the requires clause.
25.1.6 Change p1 by deleting the requires clause.
25.1.9
Change p4 from
-4- Requires: Type T is EqualityComparable (20.1.1), type Size is convertible to integral type (4.7.12.3).
to
-4- Requires: The type Size is convertible to integral type (4.7.12.3).
25.2.4 Change p1 from
-1- Requires: Type T is Assignable (23.1 ) (and, for replace(), EqualityComparable (20.1.1 )).
to
-1- Requires: The expression *first = new_value must be valid.
and change p4 from
-4- Requires: Type T is Assignable (23.1) (and, for replace_copy(), EqualityComparable (20.1.1)). The ranges [first, last) and [result, result + (last - first)) shall not overlap.
to
-4- Requires: The results of the expressions *first and new_value must be writable to the result output iterator. The ranges [first, last) and [result, result + (last - first)) shall not overlap.
25.2.5 Change p1 from
-1- Requires: Type T is Assignable (23.1). The type Size is convertible to an integral type (4.7.12.3).
to
-1- Requires: The expression value must be is writable to the output iterator. The type Size is convertible to an integral type (4.7.12.3).
25.2.7 Change p1 from
-1- Requires: Type T is EqualityComparable (20.1.1).
to
-1- Requires: The value type of the iterator must be Assignable (23.1).
Rationale:
The general idea of the proposed solution is to remove the faulty requires clauses and let the returns and effects clauses speak for themselves. That is, the returns clauses contain expressions that must be valid, and therefore already imply the correct requirements. In addition, a sentence is added at the beginning of chapter 25 saying that expressions given as conditions must evaluate to true or false in a boolean context. An alternative would be to say that the type of these condition expressions must be literally bool, but that would be imposing a greater restriction that what the standard currently says (which is convertible to bool).
Section: 25.1.1 [lib.alg.foreach] Status: Open Submitter: Angelika Langer Date: 03 Jan 2001
The specification of the for_each algorithm does not have a "Requires" section, which means that there are no restrictions imposed on the function object whatsoever. In essence it means that I can provide any function object with arbitrary side effects and I can still expect a predictable result. In particular I can expect that the function object is applied exactly last - first times, which is promised in the "Complexity" section.
I don't see how any implementation can give such a guarantee without imposing requirements on the function object.
Just as an example: consider a function object that removes elements from the input sequence. In that case, what does the complexity guarantee (applies f exactly last - first times) mean?
One can argue that this is obviously a nonsensical application and a theoretical case, which unfortunately it isn't. I have seen programmers shooting themselves in the foot this way, and they did not understand that there are restrictions even if the description of the algorithm does not say so.
Proposed resolution:
Add a "Requires" section to section 25.1.1 similar to those proposed for transform and the numeric algorithms (see issue 242):
-2- Requires: In the range [first, last], f shall not invalidate iterators or subranges.
[Copenhagen: The LWG agrees that a function object passed to an algorithm should not invalidate iterators in the range that the algorithm is operating on. The LWG believes that this should be a blanket statement in Clause 25, not just a special requirement for for_each. ]
Section: 25.3.5 [lib.alg.set.operations] Status: Review Submitter: Matt Austern Date: 03 Jan 2001
The standard library contains four algorithms that compute set operations on sorted ranges: set_union, set_intersection, set_difference, and set_symmetric_difference. Each of these algorithms takes two sorted ranges as inputs, and writes the output of the appropriate set operation to an output range. The elements in the output range are sorted.
The ordinary mathematical definitions are generalized so that they apply to ranges containing multiple copies of a given element. Two elements are considered to be "the same" if, according to an ordering relation provided by the user, neither one is less than the other. So, for example, if one input range contains five copies of an element and another contains three, the output range of set_union will contain five copies, the output range of set_intersection will contain three, the output range of set_difference will contain two, and the output range of set_symmetric_difference will contain two.
Because two elements can be "the same" for the purposes of these set algorithms, without being identical in other respects (consider, for example, strings under case-insensitive comparison), this raises a number of unanswered questions:
The standard should either answer these questions, or explicitly say that the answers are unspecified. I prefer the former option, since, as far as I know, all existing implementations behave the same way.
Proposed resolution:
Add the following to the end of 25.3.5.2 [lib.set.union] paragraph 5:
If [first1, last1) contains m elements that are equivalent to each other and [first2, last2) contains n elements that are equivalent to them, then max(m, n) of these elements will be copied to the output range: all m of these elements from [first1, last1), and the last max(n-m, 0) of them from [first2, last2), in that order.
Add the following to the end of 25.3.5.3 [lib.set.intersection] paragraph 5:
If [first1, last1) contains m elements that are equivalent to each other and [first2, last2) contains n elements that are equivalent to them, the first min(m, n) of those elements from [first1, last1) are copied to the output range.
Add a new paragraph, Notes, after 25.3.5.4 [lib.set.difference] paragraph 4:
If [first1, last1) contains m elements that are equivalent to each other and [first2, last2) contains n elements that are equivalent to them, the last max(m-n, 0) elements from [first1, last1) are copied to the output range.
Add a new paragraph, Notes, after 25.3.5.5 [lib.set.symmetric.difference] paragraph 4:
If [first1, last1) contains m elements that are equivalent to each other and [first2, last2) contains n elements that are equivalent to them, then |m - n| of those elements will be copied to the output range: the last m - n of these elements from [first1, last1) if m > n, and the last n - m of these elements from [first2, last2) if m < n.
[Santa Cruz: it's believed that this language is clearer than what's in the Standard. However, it's also believed that the Standard may already make these guarantees (although not quite in these words). Bill and Howard will check and see whether they think that some or all of these changes may be redundant. If so, we may close this issue as NAD.]
Rationale:
For simple cases, these descriptions are equivalent to what's already in the Standard. For more complicated cases, they describe the behavior of existing implementations.
Section: 17.4.3.1.1 [lib.macro.names] Status: Open Submitter: James Kanze Date: 11 Jan 2001
Paragraph 2 of 17.4.3.1.1 [lib.macro.names] reads: "A translation unit that includes a header shall not contain any macros that define names declared in that header." As I read this, it would mean that the following program is legal:
#define npos 3.14 #include <sstream>
since npos is not defined in <sstream>. It is, however, defined in <string>, and it is hard to imagine an implementation in which <sstream> didn't include <string>.
I think that this phrase was probably formulated before it was decided that a standard header may freely include other standard headers. The phrase would be perfectly appropriate for C, for example. In light of 17.4.4.1 [lib.res.on.headers] paragraph 1, however, it isn't stringent enough.
Proposed resolution:
In paragraph 2 of 17.4.3.1.1 [lib.macro.names], change "A translation unit that includes a header shall not contain any macros that define names declared in that header." to "A translation unit that includes a header shall not contain any macros that define names declared in any standard header."
[Copenhagen: the general idea is clearly correct, but there is concern about making sure that the two paragraphs in 17.4.3.1.1 [lib.macro.names] remain consistent. Nathan will provide new wording.]
Section: 24.1.4 [lib.bidirectional.iterators], 24.1.5 [lib.random.access.iterators] Status: Open Submitter: John Potter Date: 22 Jan 2001
In section 24.1.4 [lib.bidirectional.iterators], Table 75 gives the return type of *r-- as convertible to T. This is not consistent with Table 74 which gives the return type of *r++ as T&. *r++ = t is valid while *r-- = t is invalid.
In section 24.1.5 [lib.random.access.iterators], Table 76 gives the return type of a[n] as convertible to T. This is not consistent with the semantics of *(a + n) which returns T& by Table 74. *(a + n) = t is valid while a[n] = t is invalid.
Discussion from the Copenhagen meeting: the first part is uncontroversial. The second part, operator[] for Random Access Iterators, requires more thought. There are reasonable arguments on both sides. Return by value from operator[] enables some potentially useful iterators, e.g. a random access "iota iterator" (a.k.a "counting iterator" or "int iterator"). There isn't any obvious way to do this with return-by-reference, since the reference would be to a temporary. On the other hand, reverse_iterator takes an arbitrary Random Access Iterator as template argument, and its operator[] returns by reference. If we decided that the return type in Table 76 was correct, we would have to change reverse_iterator. This change would probably affect user code.
History: the contradiction between reverse_iterator and the Random Access Iterator requirements has been present from an early stage. In both the STL proposal adopted by the committee (N0527==94-0140) and the STL technical report (HPL-95-11 (R.1), by Stepanov and Lee), the Random Access Iterator requirements say that operator[]'s return value is "convertible to T". In N0527 reverse_iterator's operator[] returns by value, but in HPL-95-11 (R.1), and in the STL implementation that HP released to the public, reverse_iterator's operator[] returns by reference. In 1995, the standard was amended to reflect the contents of HPL-95-11 (R.1). The original intent for operator[] is unclear.
In the long term it may be desirable to add more fine-grained iterator requirements, so that access method and traversal strategy can be decoupled. (See "Improved Iterator Categories and Requirements", N1297 = 01-0011, by Jeremy Siek.) Any decisions about issue 299 should keep this possibility in mind.
Further discussion: I propose a compromise between John Potter's resolution, which requires T& as the return type of a[n], and the current wording, which requires convertible to T. The compromise is to keep the convertible to T for the return type of the expression a[n], but to also add a[n] = t as a valid expression. This compromise "saves" the common case uses of random access iterators, while at the same time allowing iterators such as counting iterator and caching file iterators to remain random access iterators (iterators where the lifetime of the object returned by operator*() is tied to the lifetime of the iterator).
Note that the compromise resolution necessitates a change to reverse_iterator. It would need to use a proxy to support a[n] = t.
Note also there is one kind of mutable random access iterator that will no longer meet the new requirements. Currently, iterators that return an r-value from operator[] meet the requirements for a mutable random access iterartor, even though the expression a[n] = t will only modify a temporary that goes away. With this proposed resolution, a[n] = t will be required to have the same operational semantics as *(a + n) = t.
Proposed resolution:
In section 24.1.4 [lib.bidirectdional.iterators], change the return type in table 75 from "convertible to T" to T&.
In section 24.1.5 [lib.random.access.iterators], change the operational semantics for a[n] to " the r-value of a[n] is equivalent to the r-value of *(a + n)". Add a new row in the table for the expression a[n] = t with a return type of convertible to T and operational semantics of *(a + n) = t.
Section: 27.6 [lib.iostream.format] Status: Open Submitter: Martin Sebor Date: 19 Mar 2001
The descriptions of the constructors of basic_istream<>::sentry (27.6.1.1.2 [lib.istream::sentry]) and basic_ostream<>::sentry (27.6.2.3 [lib.ostream::sentry]) do not explain what the functions do in case an exception is thrown while they execute. Some current implementations allow all exceptions to propagate, others catch them and set ios_base::badbit instead, still others catch some but let others propagate.
The text also mentions that the functions may call setstate(failbit) (without actually saying on what object, but presumably the stream argument is meant). That may have been fine for basic_istream<>::sentry prior to issue 195, since the function performs an input operation which may fail. However, issue 195 amends 27.6.1.1.2 [lib.istream::sentry], p2 to clarify that the function should actually call setstate(failbit | eofbit), so the sentence in p3 is redundant or even somewhat contradictory.
The same sentence that appears in 27.6.2.3 [lib.ostream::sentry], p3 doesn't seem to be very meaningful for basic_istream<>::sentry which performs no input. It is actually rather misleading since it would appear to guide library implementers to calling setstate(failbit) when os.tie()->flush(), the only called function, throws an exception (typically, it's badbit that's set in response to such an event).
Additional comments from Martin, who isn't comfortable with the current proposed resolution (see c++std-lib-11530)
The istream::sentry ctor says nothing about how the function deals with exemptions (27.6.1.1.2, p1 says that the class is responsible for doing "exception safe"(*) prefix and suffix operations but it doesn't explain what level of exception safety the class promises to provide). The mockup example of a "typical implementation of the sentry ctor" given in 27.6.1.1.2, p6, removed in ISO/IEC 14882:2003, doesn't show exception handling, either. Since the ctor is not classified as a formatted or unformatted input function, the text in 27.6.1.1, p1 through p4 does not apply. All this would seem to suggest that the sentry ctor should not catch or in any way handle exceptions thrown from any functions it may call. Thus, the typical implementation of an istream extractor may look something like [1].
The problem with [1] is that while it correctly sets ios::badbit if an exception is thrown from one of the functions called from the sentry ctor, if the sentry ctor reaches EOF while extracting whitespace from a stream that has eofbit or failbit set in exceptions(), it will cause an ios::failure to be thrown, which will in turn cause the extractor to set ios::badbit.
The only straightforward way to prevent this behavior is to move the definition of the sentry object in the extractor above the try block (as suggested by the example in 22.2.8, p9 and also indirectly supported by 27.6.1.3, p1). See [2]. But such an implementation will allow exceptions thrown from functions called from the ctor to freely propagate to the caller regardless of the setting of ios::badbit in the stream object's exceptions().
So since neither [1] nor [2] behaves as expected, the only possible solution is to have the sentry ctor catch exceptions thrown from called functions, set badbit, and propagate those exceptions if badbit is also set in exceptions(). (Another solution exists that deals with both kinds of sentries, but the code is non-obvious and cumbersome -- see [3].)
Please note that, as the issue points out, current libraries do not behave consistently, suggesting that implementors are not quite clear on the exception handling in istream::sentry, despite the fact that some LWG members might feel otherwise. (As documented by the parenthetical comment here: http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/papers/2003/n1480.html#309)
Also please note that those LWG members who in Copenhagen felt that "a sentry's constructor should not catch exceptions, because sentries should only be used within (un)formatted input functions and that exception handling is the responsibility of those functions, not of the sentries," as noted here http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/papers/2001/n1310.html#309 would in effect be either arguing for the behavior described in [1] or for extractors implemented along the lines of [3].
The original proposed resolution (Revision 25 of the issues list) clarifies the role of the sentry ctor WRT exception handling by making it clear that extractors (both library or user-defined) should be implemented along the lines of [2] (as opposed to [1]) and that no exception thrown from the callees should propagate out of either function unless badbit is also set in exceptions().
[1] Extractor that catches exceptions thrown from sentry:
struct S { long i; };
istream& operator>> (istream &strm, S &s)
{
ios::iostate err = ios::goodbit;
try {
const istream::sentry guard (strm, false);
if (guard) {
use_facet<num_get<char> >(strm.getloc ())
.get (istreambuf_iterator<char>(strm),
istreambuf_iterator<char>(),
strm, err, s.i);
}
}
catch (...) {
bool rethrow;
try {
strm.setstate (ios::badbit);
rethrow = false;
}
catch (...) {
rethrow = true;
}
if (rethrow)
throw;
}
if (err)
strm.setstate (err);
return strm;
}
[2] Extractor that propagates exceptions thrown from sentry:
istream& operator>> (istream &strm, S &s)
{
istream::sentry guard (strm, false);
if (guard) {
ios::iostate err = ios::goodbit;
try {
use_facet<num_get<char> >(strm.getloc ())
.get (istreambuf_iterator<char>(strm),
istreambuf_iterator<char>(),
strm, err, s.i);
}
catch (...) {
bool rethrow;
try {
strm.setstate (ios::badbit);
rethrow = false;
}
catch (...) {
rethrow = true;
}
if (rethrow)
throw;
}
if (err)
strm.setstate (err);
}
return strm;
}
[3] Extractor that catches exceptions thrown from sentry but doesn't set badbit if the exception was thrown as a result of a call to strm.clear().
istream& operator>> (istream &strm, S &s)
{
const ios::iostate state = strm.rdstate ();
const ios::iostate except = strm.exceptions ();
ios::iostate err = std::ios::goodbit;
bool thrown = true;
try {
const istream::sentry guard (strm, false);
thrown = false;
if (guard) {
use_facet<num_get<char> >(strm.getloc ())
.get (istreambuf_iterator<char>(strm),
istreambuf_iterator<char>(),
strm, err, s.i);
}
}
catch (...) {
if (thrown && state & except)
throw;
try {
strm.setstate (ios::badbit);
thrown = false;
}
catch (...) {
thrown = true;
}
if (thrown)
throw;
}
if (err)
strm.setstate (err);
return strm;
}
Proposed resolution:
Remove the last sentence of 27.6.1.1.2 [lib.istream::sentry] p5 (but not the footnote, which should be moved to the preceding sentence).
Remove the last sentence of 27.6.2.3 [lib.ostream::sentry] p3 (but not the footnote, which should be moved to the preceding sentence).
Rationale:
The LWG feels that no clarification of EH policy is necessary: the standard is precise about which operations sentry's constructor performs, and about which of those operations can throw. However, the sentence at the end should be removed because it's redundant.
Section: 27.6.1.3 [lib.istream.unformatted] Status: Open Submitter: Howard Hinnant Date: 09 Oct 201
I think we have a defect.
According to lwg issue 60 which is now a dr, the description of seekg in 27.6.1.3 [lib.istream.unformatted] paragraph 38 now looks like:
Behaves as an unformatted input function (as described in 27.6.1.3, paragraph 1), except that it does not count the number of characters extracted and does not affect the value returned by subsequent calls to gcount(). After constructing a sentry object, if fail() != true, executes rdbuf()>pubseekpos( pos).
And according to lwg issue 243 which is also now a dr, 27.6.1.3, paragraph 1 looks like:
Each unformatted input function begins execution by constructing an object of class sentry with the default argument noskipws (second) argument true. If the sentry object returns true, when converted to a value of type bool, the function endeavors to obtain the requested input. Otherwise, if the sentry constructor exits by throwing an exception or if the sentry object returns false, when converted to a value of type bool, the function returns without attempting to obtain any input. In either case the number of extracted characters is set to 0; unformatted input functions taking a character array of non-zero size as an argument shall also store a null character (using charT()) in the first location of the array. If an exception is thrown during input then ios::badbit is turned on in *this'ss error state. If (exception()&badbit)!= 0 then the exception is rethrown. It also counts the number of characters extracted. If no exception has been thrown it ends by storing the count in a member object and returning the value specified. In any event the sentry object is destroyed before leaving the unformatted input function.
And finally 27.6.1.1.2/5 says this about sentry:
If, after any preparation is completed, is.good() is true, ok_ != false otherwise, ok_ == false.
So although the seekg paragraph says that the operation proceeds if !fail(), the behavior of unformatted functions says the operation proceeds only if good(). The two statements are contradictory when only eofbit is set. I don't think the current text is clear which condition should be respected.
Further discussion from Redmond:
PJP: It doesn't seem quite right to say that seekg is "unformatted". That makes specific claims about sentry that aren't quite appropriate for seeking, which has less fragile failure modes than actual input. If we do really mean that it's unformatted input, it should behave the same way as other unformatted input. On the other hand, "principle of least surprise" is that seeking from EOF ought to be OK.
Dietmar: nothing should depend on eofbit. Eofbit should only be examined by the user to determine why something failed.
[Taken from c++std-lib-8873, c++std-lib-8874, c++std-lib-8876]
Proposed resolution:
[Santa Cruz: On the one hand, it would clearly be silly to seek to a non-EOF position without resetting eofbit. On the other hand, having seek clear eofbit explicitly would set a major precedent: there is currently no place where any of the flags are reset without the user explicitly asking for them to be. This is the tip of a general problem, that the various flags are stickier than many users might expect. Bill, Gaby, and Howard will discuss this issue and propose a resolution.]
Section: 22.1.1.1.1 [lib.locale.category] Status: Review Submitter: P.J. Plauger, Nathan Myers Date: 23 Oct 2001
In 22.1.1.1.1 [lib.locale.category] paragraph 1, the category members are described as bitmask elements. In fact, the bitmask requirements in 17.3.2.1.2 [lib.bitmask.types] don't seem quite right: none and all are bitmask constants, not bitmask elements.
In particular, the requirements for none interact poorly with the requirement that the LC_* constants from the C library must be recognizable as C++ locale category constants. LC_* values should not be mixed with these values to make category values.
We have two options for the proposed resolution. Informally: option 1 removes the requirement that LC_* values be recognized as category arguments. Option 2 changes the category type so that this requirement is implementable, by allowing none to be some value such as 0x1000 instead of 0.
Nathan writes: "I believe my proposed resolution [Option 2] merely re-expresses the status quo more clearly, without introducing any changes beyond resolving the DR.
Proposed resolution:
Replace the first two paragraphs of 22.1.1.1 [lib.locale.types] with:
typedef int category;Valid category values include the locale member bitmask elements collate, ctype, monetary, numeric, time, and messages, each of which represents a single locale category. In addition, locale member bitmask constant none is defined as zero and represents no category. And locale member bitmask constant all is defined such that the expression
(collate | ctype | monetary | numeric | time | messages | all) == allis true, and represents the union of all categories. Further the expression (X | Y), where X and Y each represent a single category, represents the union of the two categories.
locale member functions expecting a category argument require one of the category values defined above, or the union of two or more such values. Such a category argument identifies a set of locale categories. Each locale category, in turn, identifies a set of locale facets, including at least those shown in Table 51:
[Curaçao: need input from locale experts.]
Rationale:
The LWG considered, and rejected, an alternate proposal (described as "Option 2" in the discussion). The main reason for rejecting it was that library implementors were concerened about implementation difficult, given that getting a C++ library to work smoothly with a separately written C library is already a delicate business. Some library implementers were also concerned about the issue of adding extra locale categories.
Option 2:
Replace the first paragraph of 22.1.1.1 [lib.locale.types] with:Valid category values include the enumerated values. In addition, the result of applying commutative operators | and & to any two valid values is valid, and results in the setwise union and intersection, respectively, of the argument categories. The values all and none are defined such that for any valid value cat, the expressions (cat | all == all), (cat & all == cat), (cat | none == cat) and (cat & none == none) are true. For non-equal values cat1 and cat2 of the remaining enumerated values, (cat1 & cat2 == none) is true. For any valid categories cat1 and cat2, the result of (cat1 & ~cat2) is valid, and equals the setwise union of those categories found in cat1 but not found in cat2. [Footnote: it is not required that all equal the setwise union of the other enumerated values; implementations may add extra categories.]
Section: 21.1.2 [lib.char.traits.typedefs] Status: Review Submitter: Martin Sebor Date: 2 Dec 2001
(1) There are no requirements on the stateT template parameter of fpos listed in 27.4.3. The interface appears to require that the type be at least Assignable and CopyConstructible (27.4.3.1, p1), and I think also DefaultConstructible (to implement the operations in Table 88).
21.1.2, p3, however, only requires that char_traits<charT>::state_type meet the requirements of CopyConstructible types.
(2) Additionally, the stateT template argument has no corresponding typedef in fpos which might make it difficult to use in generic code.
Proposed resolution:
Modify 21.1.2, p4 from
Requires: state_type shall meet the requirements of CopyConstructible types (20.1.3).
Requires: state_type shall meet the requirements of Assignable (23.1, p4), CopyConstructible (20.1.3), and DefaultConstructible (20.1.4) types.
Rationale:
The LWG feels this is two issues, as indicated above. The first is a defect---std::basic_fstream is unimplementable without these additional requirements---and the proposed resolution fixes it. The second is questionable; who would use that typedef? The class template fpos is used only in a very few places, all of which know the state type already. Unless motivation is provided, the second should be considered NAD.
Section: 23.1.1 [lib.sequence.reqmts] Status: Review Submitter: Yaroslav Mironov Date: 23 Jan 2002
Table 68 "Optional Sequence Operations" in 23.1.1/12 specifies operational semantics for "a.back()" as "*--a.end()", which may be ill-formed [because calling operator-- on a temporary (the return) of a built-in type is ill-formed], provided a.end() returns a simple pointer rvalue (this is almost always the case for std::vector::end(), for example). Thus, the specification is not only incorrect, it demonstrates a dangerous construct: "--a.end()" may successfully compile and run as intended, but after changing the type of the container or the mode of compilation it may produce compile-time error.
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; *tmp; }
and the specification for "a.pop_back()" from
a.erase(--a.end())
to
{ iterator tmp = a.end(); --tmp; a.erase(tmp); }
[Curaçao: LWG changed PR from "{ X::iterator tmp = a.end(); return *--tmp; }" to "*a.rbegin()", and from "{ X::iterator tmp = a.end(); a.erase(--tmp); }" to "a.erase(rbegin())".]
[There is a second possible defect; table 68 "Optional Sequence Operations" in the "Operational Semantics" column uses operations present only in the "Reversible Container" requirements, yet there is no stated dependency between these separate requirements tables. Ask in Santa Cruz if the LWG would like a new issue opened.]
[Santa Cruz: the proposed resolution is even worse than what's in the current standard: erase is undefined for reverse iterator. If we're going to make the change, we need to define a temporary and use operator--. Additionally, we don't know how prevalent this is: do we need to make this change in more than one place? Martin has volunteered to review the standard and see if this problem occurs elsewhere.]
[Oxford: Matt provided new wording to address the concerns raised in Santa Cruz. It does not appear that this problem appears anywhere else in clauses 23 or 24.]
Section: 22.2.1 [lib.category.ctype] Status: Open Submitter: Matt Austern Date: 23 Jan 2002
What should the following program print?
#include <locale>
#include <iostream>
class my_ctype : public std::ctype<char>
{
typedef std::ctype<char> base;
public:
my_ctype(std::size_t refs = 0) : base(my_table, false, refs)
{
std::copy(base::classic_table(), base::classic_table() + base::table_size,
my_table);
my_table[(unsigned char) '_'] = (base::mask) (base::print | base::space);
}
private:
mask my_table[base::table_size];
};
int main()
{
my_ctype ct;
std::cout << "isspace: " << ct.is(std::ctype_base::space, '_') << " "
<< "isalpha: " << ct.is(std::ctype_base::alpha, '_') << std::endl;
}
The goal is to create a facet where '_' is treated as whitespace.
On gcc 3.0, this program prints "isspace: 1 isalpha: 0". On Microsoft C++ it prints "isspace: 1 isalpha: 1".
I believe that both implementations are legal, and the standard does not give enough guidance for users to be able to use std::ctype's protected interface portably.
The above program assumes that ctype_base::mask enumerators like space and print are disjoint, and that the way to say that a character is both a space and a printing character is to or those two enumerators together. This is suggested by the "exposition only" values in 22.2.1 [lib.category.ctype], but it is nowhere specified in normative text. An alternative interpretation is that the more specific categories subsume the less specific. The above program gives the results it does on the Microsoft compiler because, on that compiler, print has all the bits set for each specific printing character class.
From the point of view of std::ctype's public interface, there's no important difference between these two techniques. From the point of view of the protected interface, there is. If I'm defining a facet that inherits from std::ctype<char>, I'm the one who defines the value that table()['a'] returns. I need to know what combination of mask values I should use. This isn't so very esoteric: it's exactly why std::ctype has a protected interface. If we care about users being able to write their own ctype facets, we have to give them a portable way to do it.
Related reflector messages: lib-9224, lib-9226, lib-9229, lib-9270, lib-9272, lib-9273, lib-9274, lib-9277, lib-9279.
Issue 339 is related, but not identical. The proposed resolution if issue 339 says that ctype_base::mask must be a bitmask type. It does not say that the ctype_base::mask elements are bitmask elements, so it doesn't directly affect this issue.
More comments from Benjamin Kosnik, who believes that that C99 compatibility essentially requires what we're calling option 1 below.
I think the C99 standard is clear, that isspace -> !isalpha.
--------
#include <locale>
#include <iostream>
class my_ctype : public std::ctype<char>
{
private:
typedef std::ctype<char> base;
mask my_table[base::table_size];
public:
my_ctype(std::size_t refs = 0) : base(my_table, false, refs)
{
std::copy(base::classic_table(), base::classic_table() + base::table_size,
my_table);
mask both = base::print | base::space;
my_table[static_cast<mask>('_')] = both;
}
};
int main()
{
using namespace std;
my_ctype ct;
cout << "isspace: " << ct.is(ctype_base::space, '_') << endl;
cout << "isprint: " << ct.is(ctype_base::print, '_') << endl;
// ISO C99, isalpha iff upper | lower set, and !space.
// 7.5, p 193
// -> looks like g++ behavior is correct.
// 356 -> bitmask elements are required for ctype_base
// 339 -> bitmask type required for mask
cout << "isalpha: " << ct.is(ctype_base::alpha, '_') << endl;
}
Proposed resolution:
Informally, we have three choices:
Either of the first two options is just as good from the standpoint of portability. Either one will require some implementations to change.
[ More discussion is needed. Nobody likes option 3. Options 1 and 2 are both controversial, 2 perhaps less so. Benjamin thinks that option 1 is required for C99 compatibility. ]
Section: 22.2.2.2.1 [lib.facet.num.put.members] Status: Review Submitter: Martin Sebor Date: 12 Mar 2002
22.2.2.2.1, p1:
iter_type put (iter_type out, ios_base& str, char_type fill,
bool val) const;
...
1 Returns: do_put (out, str, fill, val).
AFAICS, the behavior of do_put (..., bool) is not documented anywhere, however, 22.2.2.2.2, p23:
iter_type put (iter_type out, ios_base& str, char_type fill, bool val) const;Effects: If (str.flags() & ios_base::boolalpha) == 0 then do out = do_put(out, str, fill, (int)val) Otherwise dostring_type s = val ? use_facet<ctype<charT> >(loc).truename() : use_facet<ctype<charT> >(loc).falsename();and then insert the characters of s into out. out.
This means that the bool overload of do_put() will never be called, which contradicts the first paragraph. Perhaps the declaration should read do_put(), and not put()?
Note also that there is no Returns clause for this function, which should probably be corrected, just as should the second occurrence of "out." in the text.
I think the least invasive change to fix it would be something like the following:
Proposed resolution:
In 22.2.2.2.2 [lib.facet.num.put.virtuals], just above paragraph 1, remove the bool overload.
In 22.2.2.2.2 [lib.facet.num.put.virtuals], p23, make the following changes
Replace put() with do_put() in the declaration of the member function.
Change the Effects clause to a Returns clause (to avoid the requirement to call do_put(..., int) from do_put (..., bool)) like so:
23 Returns: If (str.flags() &
ios_base::boolalpha) == 0 then
do_put (out, str, fill, (long)val)
Otherwise the function obtains a string s as if by
string_type s =
val ? use_facet<ctype<charT> >(loc).truename()
: use_facet<ctype<charT> >(loc).falsename();
and then inserts each character c of s into out via
*out++ = c
and returns out.
Rationale:
This fixes a couple of obvious typos, and also fixes what appears to be a requirement of gratuitous inefficiency.
Section: 20.3.6.2 [lib.bind.1st] Status: Open Submitter: Andrew Demkin Date: 26 Apr 2002
The definition of bind1st() (20.3.6.2 [lib.bind.1st]) can result in the construction of an unsafe binding between incompatible pointer types. For example, given a function whose first parameter type is 'pointer to T', it's possible without error to bind an argument of type 'pointer to U' when U does not derive from T:
foo(T*, int);
struct T {};
struct U {};
U u;
int* p;
int* q;
for_each(p, q, bind1st(ptr_fun(foo), &u)); // unsafe binding
The definition of bind1st() includes a functional-style conversion to map its argument to the expected argument type of the bound function (see below):
typename Operation::first_argument_type(x)
A functional-style conversion (5.2.3 [expr.type.conv]) is defined to be semantically equivalent to an explicit cast expression (5.4 [expr.cast]), which may (according to 5.4, paragraph 5) be interpreted as a reinterpret_cast, thus masking the error.
The problem and proposed change also apply to 20.3.6.4 [lib.bind.2nd].
Proposed resolution:
The simplest and most localized change to prevent such errors is to require bind1st() use a static_cast expression rather than the functional-style conversion; that is, have bind1st() return:
binder1st<Operation>( op,
static_cast<typename Operation::first_argument_type>(x)).
A more agressive solution is to change the semantics of functional-style conversions to not permit a reinterpret_cast. For contexts that require the semantics of reinterpret_cast, the language may want to require the use of an explicit cast expression such as '(T) x' or 'reinterpret_cast<T>(x)' and limit the behavior of the functional notation to match statically-checked and standard conversions (as defined by 5.2.9 and 4.10, etc.). Although changing the semantics of functional-style conversions may seem drastic and does have language-wide ramifications, it has the benefit of better unifying the conversion rules for user defined types and built-in types, which can be especially important for generic template programming.
[Santa Cruz: it's clear that a function-style cast is wrong. Maybe a static cast would be better, or maybe no cast at all. Jeremy will check with the original author of this part of the Standard and will see what the original intent was.]
Section: 27 [lib.input.output] Status: Review Submitter: Walter Brown, Marc Paterno Date: 10 May 2002
Some stream and streambuf member functions are declared non-const, even thought they appear only to report information rather than to change an object's logical state. They should be declared const. See document N1360 for details and rationale.
The list of member functions under discussion: in_avail, showmanyc, tellg, tellp, is_open.
Related issue: 73
Proposed resolution:
In 27.8.1.5, 27.8.1.7, 27.8.1.8, 27.8.1.10, 27.8.1.11, and 27.8.1.13
Replace
bool is_open();
with
bool is_open() const;
Rationale:
Of the changes proposed in N1360, the only one that is safe is changing the filestreams' is_open to const. The LWG believed that this was NAD the first time it considered this issue (issue 73), but now thinks otherwise. The corresponding streambuf member function, after all,is already const.
The other proposed changes are less safe, because some streambuf functions that appear merely to report a value do actually perform mutating operations. It's not even clear that they should be considered "logically const", because streambuf has two interfaces, a public one and a protected one. These functions may, and often do, change the state as exposed by the protected interface, even if the state exposed by the public interface is unchanged.
Note that implementers can make this change in a binary compatible way by providing both overloads; this would be a conforming extension.
Section: 27 [lib.input.output] Status: Open Submitter: Walter Brown, Marc Paterno Date: 10 May 2002
The following member functions are declared const, yet return non-const pointers. We believe they are should be changed, because they allow code that may surprise the user. See document N1360 for details and rationale.
[Santa Cruz: the real issue is that we've got const member functions that return pointers to non-const, and N1360 proposes replacing them by overloaded pairs. There isn't a consensus about whether this is a real issue, since we've never said what our constness policy is for iostreams. N1360 relies on a distinction between physical constness and logical constness; that distinction, or those terms, does not appear in the standard.]
Proposed resolution:
In 27.4.4 and 27.4.4.2
Replace
basic_ostream<charT,traits>* tie() const;
with
basic_ostream<charT,traits>* tie(); const basic_ostream<charT,traits>* tie() const;
and replace
basic_streambuf<charT,traits>* rdbuf() const;
with
basic_streambuf<charT,traits>* rdbuf(); const basic_streambuf<charT,traits>* rdbuf() const;
In 27.5.2 and 27.5.2.3.1
Replace
char_type* eback() const;
with
char_type* eback(); const char_type* eback() const;
Replace
char_type gptr() const;
with
char_type* gptr(); const char_type* gptr() const;
Replace
char_type* egptr() const;
with
char_type* egptr(); const char_type* egptr() const;
In 27.5.2 and 27.5.2.3.2
Replace
char_type* pbase() const;
with
char_type* pbase(); const char_type* pbase() const;
Replace
char_type* pptr() const;
with
char_type* pptr(); const char_type* pptr() const;
Replace
char_type* epptr() const;
with
char_type* epptr(); const char_type* epptr() const;
In 27.7.2, 27.7.2.2, 27.7.3 27.7.3.2, 27.7.4, and 27.7.6
Replace
basic_stringbuf<charT,traits,Allocator>* rdbuf() const;
with
basic_stringbuf<charT,traits,Allocator>* rdbuf(); const basic_stringbuf<charT,traits,Allocator>* rdbuf() const;
In 27.8.1.5, 27.8.1.7, 27.8.1.8, 27.8.1.10, 27.8.1.11, and 27.8.1.13
Replace
basic_filebuf<charT,traits>* rdbuf() const;
with
basic_filebuf<charT,traits>* rdbuf(); const basic_filebuf<charT,traits>* rdbuf() const;
Section: 21.3.5.6 [lib.string::replace] Status: Open Submitter: Beman Dawes Date: 3 Jun 2002
21.3.5.6 [lib.string::replace] basic_string::replace, second signature, given in paragraph 1, has two "Throws" paragraphs (3 and 5).
In addition, the second "Throws" paragraph (5) includes specification (beginning with "Otherwise, the function replaces ...") that should be part of the "Effects" paragraph.
Proposed resolution:
[This is a typo that escalated. It's clear that what's in the Standard is wrong. It's less clear what the fix ought to be. Someone who understands string replace well needs to work on this.]
Section: 27.3 [lib.iostream.objects] Status: Open Submitter: Ruslan Abdikeev Date: 8 Jul 2002
Is it safe to use standard iostream objects from constructors of static objects? Are standard iostream objects constructed and are their associations established at that time?
Surpisingly enough, Standard does NOT require that.
27.3/2 [lib.iostream.objects] guarantees that standard iostream objects are constructed and their associations are established before the body of main() begins execution. It also refers to ios_base::Init class as the panacea for constructors of static objects.
However, there's nothing in 27.3 [lib.iostream.objects], in 27.4.2 [lib.ios.base], and in 27.4.2.1.6 [lib.ios::Init], that would require implementations to allow access to standard iostream objects from constructors of static objects.
Details:
Core text refers to some magic object ios_base::Init, which will be discussed below:
"The [standard iostream] objects are constructed, and their associations are established at some time prior to or during first time an object of class basic_ios<charT,traits>::Init is constructed, and in any case before the body of main begins execution." (27.3/2 [lib.iostream.objects])
The first non-normative footnote encourages implementations to initialize standard iostream objects earlier than required.
However, the second non-normative footnote makes an explicit and unsupported claim:
"Constructors and destructors for static objects can access these [standard iostream] objects to read input from stdin or write output to stdout or stderr." (27.3/2 footnote 265 [lib.iostream.objects])
The only bit of magic is related to that ios_base::Init class. AFAIK, the rationale behind ios_base::Init was to bring an instance of this class to each translation unit which #included <iostream> or related header. Such an inclusion would support the claim of footnote quoted above, because in order to use some standard iostream object it is necessary to #include <iostream>.
However, while Standard explicitly describes ios_base::Init as an appropriate class for doing the trick, I failed to found a mention of an _instance_ of ios_base::Init in Standard.
Proposed resolution:
At the end of header <iostream> synopsis in 27.3 [lib.iostream.objects]
namespace std
{
... extern istream cin; ...
add the following lines
namespace
{
ios_base::Init <some_implementation_defined_name>;
}
}
[Santa Cruz: The LWG is leaning toward NAD. There isn't any normative wording saying that the Init scheme will be used, but that is probably intentional. Implementers use dirty tricks for iostream initialization, and doing it portably is somewhere between difficult and impossible. Too much constraint in this area is dangerous, and if we are to make any changes it would probably be more appropriate forthem to be nonnormative. Martin will try to come up with clearer wording that expreses this intent.]
Section: 23.1 [lib.container.requirements] Status: Open Submitter: Frank Compagner Date: 20 Jul 2002
The requirements for multiset and multimap containers (23.1 [lib.containers.requirements], 23.1.2 [lib.associative.reqmnts], 23.3.2 [lib.multimap] and 23.3.4 [lib.multiset]) make no mention of the stability of the required (mutating) member functions. It appears the standard allows these functions to reorder equivalent elements of the container at will, yet the pervasive red-black tree implementation appears to provide stable behaviour.
This is of most concern when considering the behaviour of erase(). A stability requirement would guarantee the correct working of the following 'idiom' that removes elements based on a certain predicate function.
multimap<int, int> m;
multimap<int, int>::iterator i = m.begin();
while (i != m.end()) {
if (pred(i))
m.erase (i++);
else
++i;
}
Although clause 23.1.2/8 guarantees that i remains a valid iterator througout this loop, absence of the stability requirement could potentially result in elements being skipped. This would make this code incorrect, and, furthermore, means that there is no way of erasing these elements without iterating first over the entire container, and second over the elements to be erased. This would be unfortunate, and have a negative impact on both performance and code simplicity.
If the stability requirement is intended, it should be made explicit (probably through an extra paragraph in clause 23.1.2).
If it turns out stability cannot be guaranteed, i'd argue that a remark or footnote is called for (also somewhere in clause 23.1.2) to warn against relying on stable behaviour (as demonstrated by the code above). If most implementations will display stable behaviour, any problems emerging on an implementation without stable behaviour will be hard to track down by users. This would also make the need for an erase_if() member function that much greater.
This issue is somewhat related to LWG issue 130.
[Santa Cruz: More people need to look at this. Much user code may assume stability. On the other hand, it seems drastic to add a new requirement now.]
Proposed resolution:
Section: 27.7.1.3 [lib.stringbuf.virtuals] Status: Open Submitter: Ray Lischner Date: 14 Aug 2002
In Section 27.7.1.3 [lib.stringbuf.virtuals], Table 90, the implication is that the four conditions should be mutually exclusive, but they are not. The first two cases, as written, are subcases of the third. I think it would be clearer if the conditions were rewritten as follows:
(which & (ios_base::in|ios_base::out)) == ios_base::in
(which & (ios_base::in|ios_base::out)) == ios_base::out
(which & (ios_base::in|ios_base::out)) == (ios_base::in|ios_base::out) and way == either ios_base::beg or ios_base::end
Otherwise
As written, it is unclear what should be the result if cases 1 & 2 are true, but case 3 is false, e.g.,
seekoff(0, ios_base::cur, ios_base::in | ios_base::out)
[Santa Cruz: The ambiguity seems real. We need to do a survey of implementations before we decide on a solution.]
Proposed resolution:
Section: 22.1.1 [lib.locale] Status: Open Submitter: Martin Sebor Date: 6 Sep 2002
I think there is a problem with 22.1.1, p6 which says that
-6- An instance of locale is immutable; once a facet reference
is obtained from it, that reference remains usable as long
as the locale value itself exists.
and 22.1.1.2, p4:
const locale& operator=(const locale& other) throw();
-4- Effects: Creates a copy of other, replacing the current value.
How can a reference to a facet obtained from a locale object remain valid after an assignment that clearly must replace all the facets in the locale object? Imagine a program such as this
std::locale loc ("de_DE");
const std::ctype<char> &r0 = std::use_facet<std::ctype<char> >(loc);
loc = std::locale ("en_US");
const std::ctype<char> &r1 = std::use_facet<std::ctype<char> >(loc);
Is r0 really supposed to be preserved and destroyed only when loc goes out of scope?
Proposed resolution:
Suggest to replace 22.1.1 [lib.locale], p6 with
-6- Unless assigned a new value, locale objects are immutable;
once a facet reference is obtained from it, that reference
remains usable as long as the locale object itself exists
or until the locale object is assigned the value of another,
distinct locale object.
[Santa Cruz: Dietmar agrees with this general direction, but is uncomfortable about the proposed wording. He and Martin will try to come up with better wording.]
Section: 22.2.1.1.2 [lib.locale.ctype.virtuals] Status: Review Submitter: Martin Sebor Date: 6 Sep 2002
The last sentence in 22.2.1.1.2, p11 below doesn't seem to make sense.
charT do_widen (char c) const;
-11- Effects: Applies the simplest reasonable transformation from
a char value or sequence of char values to the corresponding
charT value or values. The only characters for which unique
transformations are required are those in the basic source
character set (2.2). For any named ctype category with a
ctype<charT> facet ctw and valid ctype_base::mask value
M (is(M, c) || !ctw.is(M, do_widen(c))) is true.
Shouldn't the last sentence instead read
For any named ctype category with a ctype<char> facet ctc
and valid ctype_base::mask value M
(ctc.is(M, c) || !is(M, do_widen(c))) is true.
I.e., if the narrow character c is not a member of a class of characters then neither is the widened form of c. (To paraphrase footnote 224.)
Proposed resolution:
Replace the last sentence of 22.2.1.1.2 [lib.locale.ctype.virtuals], p11 with the following text:
For any named ctype category with a ctype<char> facet ctc
and valid ctype_base::mask value M
(ctc.is(M, c) || !is(M, do_widen(c))) is true.
Rationale:
The LWG believes this is just a typo, and that this is the correct fix.
Section: 22.2.1.5 [lib.locale.codecvt] Status: Review Submitter: Martin Sebor Date: 30 Aug 2002
It seems that the descriptions of codecvt do_in() and do_out() leave sufficient room for interpretation so that two implementations of codecvt may not work correctly with the same filebuf. Specifically, the following seems less than adequately specified:
Finally, the conditions described at the end of 22.2.1.5.2 [lib.locale.codecvt.virtuals], p4 don't seem to be possible:
"A return value of partial, if (from_next == from_end), indicates that either the destination sequence has not absorbed all the available destination elements, or that additional source elements are needed before another destination element can be produced."
If the value is partial, it's not clear to me that (from_next ==from_end) could ever hold if there isn't enough room in the destination buffer. In order for (from_next==from_end) to hold, all characters in that range must have been successfully converted (according to 22.2.1.5.2 [lib.locale.codecvt.virtuals], p2) and since there are no further source characters to convert, no more room in the destination buffer can be needed.
It's also not clear to me that (from_next==from_end) could ever hold if additional source elements are needed to produce another destination character (not element as incorrectly stated in the text). partial is returned if "not all source characters have been converted" according to Table 53, which also implies that (from_next==from) does NOT hold.
Could it be that the intended qualifying condition was actually (from_next != from_end), i.e., that the sentence was supposed to read
"A return value of partial, if (from_next != from_end),..."
which would make perfect sense, since, as far as I understand it, partial can only occur if (from_next != from_end)?
Proposed resolution:
To address these issues, I propose that paragraphs 2, 3, and 4 be rewritten as follows. The proposal incorporates the accepted resolution of lwg issue 19.
-2- Effects: Converts characters in the range of source elements
[from, from_end), placing the results in sequential positions
starting at destination to. Converts no more than (from_end from)
source elements, and stores no more than (to_limit to)
destination elements.
Stops if it encounters a sequence of source elements it cannot
convert to a valid destination character. It always leaves the
from_next and to_next pointers pointing one beyond the last
element successfully converted.
[Note: If returns noconv, internT and externT are the same type
and the converted sequence is identical to the input sequence
[from, from_next). to_next is set equal to to, the value of
state is unchanged, and there are no changes to the values in
[to, to_limit). --end note]
-3- Notes: Its operations on state are unspecified.
[Note: This argument can be used, for example, to maintain shift
state, to specify conversion options (such as count only), or to
identify a cache of seek offsets. --end note]
-4- Returns: An enumeration value, as summarized in Table 53:
Table 53 -- do_in/do_out result values
Value Meaning
+---------+----------------------------------------------------+
| ok | successfully completed the conversion of all |
| | complete characters in the source range |
+---------+----------------------------------------------------+
| partial | the characters in the source range would, after |
| | conversion, require space greater than that |
| | available in the destination range |
+---------+----------------------------------------------------+
| error | encountered either a sequence of elements in the |
| | source range forming a valid source character that |
| | could not be converted to a destination character, |
| | or a sequence of elements in the source range that |
| | could not possibly form a valid source character |
+---------+----------------------------------------------------+
| noconv | internT and externT are the same type, and input |
| | sequence is identical to converted sequence |
+---------+----------------------------------------------------+
A return value of partial, i.e., if (from_next != from_end),
indicates that either the destination sequence has not absorbed
all the available destination elements, or that additional
source elements are needed before another destination character
can be produced.
[Santa Cruz: The LWG agrees that this is an important issue and that this general direction is probably correct. Dietmar, Howard, PJP, and Matt will review this wording.]
Section: 25.3.3.3 [lib.equal.range] Status: Open Submitter: Hans Bos Date: 18 Oct 2002
Section 25.3.3.3 [lib.equal.range] states that at most 2 * log(last - first) + 1 comparisons are allowed for equal_range.
It is not possible to implement equal_range with these constraints.
In a range of one element as in:
int x = 1;
equal_range(&x, &x + 1, 1)
it is easy to see that at least 2 comparison operations are needed.
For this case at most 2 * log(1) + 1 = 1 comparison is allowed.
I have checked a few libraries and they all use the same (nonconforming) algorithm for equal_range that has a complexity of
2* log(distance(first, last)) + 2.
I guess this is the algorithm that the standard assumes for equal_range.
It is easy to see that 2 * log(distance) + 2 comparisons are enough since equal range can be implemented with lower_bound and upper_bound (both log(distance) + 1).
I think it is better to require something like 2log(distance) + O(1) (or even logarithmic as multiset::equal_range). Then an implementation has more room to optimize for certain cases (e.g. have log(distance) characteristics when at most match is found in the range but 2log(distance) + 4 for the worst case).
[Santa Cruz: The issue is real, but of greater scope than just equal_range: it affects all of the binary search algorithms. What is the complexity supposed to be for ranges of 0 or 1 elements? What base are we using for the logarithm? Are these bounds supposed to be exact, or asymptotic? (If the latter, of course, then none of the other questions matter.)]
Proposed resolution:
Section: 17 [lib.library] Status: Open Submitter: Matt Austern Date: 23 Oct 2002
Many function templates have parameters that are passed by value; a typical example is find_if's pred parameter in 25.1.2 [lib.alg.find]. Are the corresponding template parameters (Predicate in this case) implicitly required to be CopyConstructible, or does that need to be spelled out explicitly?
This isn't quite as silly a question as it might seem to be at first sight. If you call find_if in such a way that template argument deduction applies, then of course you'll get call by value and you need to provide a copy constructor. If you explicitly provide the template arguments, however, you can force call by reference by writing something like find_if<my_iterator, my_predicate&>. The question is whether implementation are required to accept this, or whether this is ill-formed because my_predicate& is not CopyConstructible.
The scope of this problem, if it is a problem, is unknown. Function object arguments to generic algorithms in clauses 25 [lib.algorithms] and 26 [lib.numerics] are obvious examples. A review of the whole library is necessary.
Proposed resolution:
[ This is really two issues. First, predicates are typically passed by value but we don't say they must be Copy Constructible. They should be. Second: is specialization allowed to transform value arguments into references? References aren't copy constructible, so this should not be allowed. ]
Section: 24.4.1.3.11 [lib.reverse.iter.opindex] Status: Open Submitter: Matt Austern Date: 23 Oct 2002
In 24.4.1.3.11 [lib.reverse.iter.opindex], reverse_iterator<>::operator[] is specified as having a return type of reverse_iterator::reference, which is the same as iterator_traits<Iterator>::reference. (Where Iterator is the underlying iterator type.)
The trouble is that Iterator's own operator[] doesn't necessarily have a return type of iterator_traits<Iterator>::reference. Its return type is merely required to be convertible to Iterator's value type. The return type specified for reverse_iterator's operator[] would thus appear to be impossible.
Related issue: 299. Jeremy will work on this.
Proposed resolution:
[ Comments from Dave Abrahams: IMO we should resolve 386 by just saying that the return type of reverse_iterator's operator[] is unspecified, allowing the random access iterator requirements to impose an appropriate return type. If we accept 299's proposed resolution (and I think we should), the return type will be readable and writable, which is about as good as we can do. ]
Section: 26.2 [lib.complex.numbers] Status: Review Submitter: Gabriel Dos Reis Date: 8 Nov 2002
The absence of explicit description of std::complex<T> layout makes it imposible to reuse existing software developed in traditional languages like Fortran or C with unambigous and commonly accepted layout assumptions. There ought to be a way for practitioners to predict with confidence the layout of std::complex<T> whenever T is a numerical datatype. The absence of ways to access individual parts of a std::complex<T> object as lvalues unduly promotes severe pessimizations. For example, the only way to change, independently, the real and imaginary parts is to write something like
complex<T> z; // ... // set the real part to r z = complex<T>(r, z.imag()); // ... // set the imaginary part to i z = complex<T>(z.real(), i);
At this point, it seems appropriate to recall that a complex number is, in effect, just a pair of numbers with no particular invariant to maintain. Existing practice in numerical computations has it that a complex number datatype is usually represented by Cartesian coordinates. Therefore the over-encapsulation put in the specification of std::complex<> is not justified.
Proposed resolution:
Add the following requirements to 26.2 [lib.complex.numbers] as 26.2/4:
If z is an lvalue expression of type cv std::complex<T> then
- the expression reinterpret_cast<cv T(&)[2]>(z) is well-formed; and
- reinterpret_cast<cvT(&)[2]>(z)[0]designates the real part of z; and
- reinterpret_cast<cvT(&)[2]>(z)[1]designates the imaginary part of z.
Moreover, if a is an expression of pointer type cv complex<T>* and the expression a[i] is well-defined for an integer expression i then:
- reinterpret_cast<cvT*>(a)[2+i] designates the real part of a[i]; and
- reinterpret_cast<cv T*>(a)[2+i+1] designates the imaginary part of a[i].
In the header synopsis in 26.2.1 [lib.complex.synopsis], replace
template<class T> T real(const complex<T>&); template<class T> T imag(const complex<T>&);
with
template<class T> const T& real(const complex<T>&); template<class T> T& real( complex<T>&); template<class T> const T& imag(const complex<T>&); template<class T> T& imag( complex<T>&);
In 26.2.7 [lib.complex.value.ops] paragraph 1, change
template<class T> T real(const complex<T>&);
to
template<class T> const T& real(const complex<T>&); template<class T> T& real( complex<T>&);
and change the Returns clause to "Returns: The real part of x
.In 26.2.7 [lib.complex.value.ops] paragraph 2, change
template<class T> T imag(const complex<T>&);
to
template<class T> const T& imag(const complex<T>&); template<class T> T& imag( complex<T>&);
and change the Returns clause to "Returns: The imaginary part of x
.Rationale:
The LWG believes that C99 compatibility would be enough justification for this change even without other considerations. All existing implementations already have the layout proposed here.
Section: 26.3.2 [lib.template.valarray] Status: Review Submitter: Gabriel Dos Reis Date: 8 Nov 2002
Consider the following program:
#include <iostream>
#include <ostream>
#include <vector>
#include <valarray>
#include <algorithm>
#include <iterator>
template<typename Array>
void print(const Array& a)
{
using namespace std;
typedef typename Array::value_type T;
copy(&a[0], &a[0] + a.size(),
ostream_iterator<T>(std::cout, " "));
}
template<typename T, unsigned N>
unsigned size(T(&)[N]) { return N; }
int main()
{
double array[] = { 0.89, 9.3, 7, 6.23 };
std::vector<double> v(array, array + size(array));
std::valarray<double> w(array, size(array));
print(v); // #1
std::cout << std::endl;
print(w); // #2
std::cout << std::endl;
}
While the call numbered #1 succeeds, the call numbered #2 fails because the const version of the member function valarray<T>::operator[](size_t) returns a value instead of a const-reference. That seems to be so for no apparent reason, no benefit. Not only does that defeats users' expectation but it also does hinder existing software (written either in C or Fortran) integration within programs written in C++. There is no reason why subscripting an expression of type valarray<T> that is const-qualified should not return a const T&.
Proposed resolution:
In the class synopsis in 26.3.2 [lib.template.valarray], and in 26.3.2.3 [lib.valarray.access] just above paragraph 1, change
T operator[](size_t const;)
to
const T& operator[](size_t const;)
Rationale:
Return by value seems to serve no purpose. Valaray was explicitly designed to have a specified layout so that it could easily be integrated with libraries in other languages, and return by value defeats that purpose. It is believed that this change will have no impact on allowable optimizations.
Section: 22.1.3.2 [lib.conversions] Status: Review Submitter: James Kanze Date: 10 Dec 2002
The specifications of toupper and tolower both specify the functions as const, althought they are not member functions, and are not specified as const in the header file synopsis in section 22.1 [lib.locales].
Proposed resolution:
In 22.1.3.2 [lib.conversions], remove const from the function declarations of std::toupper and std::tolower
Rationale:
Fixes an obvious typo
Section: 27.6.2.5 [lib.ostream.formatted] Status: Review Submitter: Martin Sebor Date: 27 Dec 2002
There is a contradiction in Formatted output about what bit is supposed to be set if the formatting fails. On sentence says it's badbit and another that it's failbit.
27.6.2.5.1, p1 says in the Common Requirements on Formatted output functions:
... If the generation fails, then the formatted output function
does setstate(ios::failbit), which might throw an exception.
27.6.2.5.2, p1 goes on to say this about Arithmetic Inserters:
... The formatting conversion occurs as if it performed the following code fragment:
bool failed =
use_facet<num_put<charT,ostreambuf_iterator<charT,traits>
> >
(getloc()).put(*this, *this, fill(), val). failed();
... If failed is true then does setstate(badbit) ...
The original intent of the text, according to Jerry Schwarz (see c++std-lib-10500), is captured in the following paragraph:
In general "badbit" should mean that the stream is unusable because of some underlying failure, such as disk full or socket closure; "failbit" should mean that the requested formatting wasn't possible because of some inconsistency such as negative widths. So typically if you clear badbit and try to output something else you'll fail again, but if you clear failbit and try to output something else you'll succeed.
In the case of the arithmetic inserters, since num_put cannot report failure by any means other than exceptions (in response to which the stream must set badbit, which prevents the kind of recoverable error reporting mentioned above), the only other detectable failure is if the iterator returned from num_put returns true from failed().
Since that can only happen (at least with the required iostream specializations) under such conditions as the underlying failure referred to above (e.g., disk full), setting badbit would seem to be the appropriate response (indeed, it is required in 27.6.2.5.2, p1). It follows that failbit can never be directly set by the arithmetic (it can only be set by the sentry object under some unspecified conditions).
The situation is different for other formatted output functions which can fail as a result of the streambuf functions failing (they may do so by means other than exceptions), and which are then required to set failbit.
The contradiction, then, is that ostream::operator<<(int) will set badbit if the disk is full, while operator<<(ostream&, char) will set failbit under the same conditions. To make the behavior consistent, the Common requirements sections for the Formatted output functions should be changed as proposed below.
Proposed resolution:
In paragraph one of section 27.6.2.5 [lib.ostream.formatted], delete the sentence beginning with "If the generation fails".
Rationale:
There isn't any contradiction here. put returns a streambuf iterator. failed() is a member function of the streambuf iterator. If it's set then that's a streambuf error, not a conversion error.
The real problem isn't that there's a contradiction, but that the "If the generation fails" part makes little sense. "Generation" isn't clearly defined. It's not clear what it means for generation to fail, or even whether it can fail. The intention is probably that generaion meant formatting, as opposed to character insertion, and that this sentence was intended as analogous to character parsing.
A more precise definition would be that we set failbit if the facet reports failure. However, the mechanism for the facet reporting failure is that it sets failbit! Saying that we set failbit if the facet sets failbit would be silly, so the best thing to say is nothing.
Section: 26.5 [lib.c.math] Status: Review Submitter: James Kanze Date: 3 Jan 2003
In 26.5 [lib.c.math], the C++ standard refers to the C standard for the definition of rand(); in the C standard, it is written that "The implementation shall behave as if no library function calls the rand function."
In 25.2.11 [lib.alg.random.shuffle], there is no specification as to how the two parameter version of the function generates its random value. I believe that all current implementations in fact call rand() (in contradiction with the requirement avove); if an implementation does not call rand(), there is the question of how whatever random generator it does use is seeded. Something is missing.
Proposed resolution:
In [lib.c.math], add a paragraph specifying that the C definition of rand shal be modified to say that "Unless otherwise specified, the implementation shall behave as if no library function calls the rand function."
In [lib.alg.random.shuffle], add a sentence to the effect that "In the two argument form of the function, the underlying source of random numbers is implementation defined. [Note: in particular, an implementation is permitted to use rand.]
Rationale:
The original proposed resolution proposed requiring the two-argument from of random_shuffle to use rand. We don't want to do that, because some existing implementations already use something else: gcc uses lrand48, for example. Using rand presents a problem if the number of elements in the sequence is greater than RAND_MAX.
Section: 23.3.5.1 [lib.bitset.cons] Status: Open Submitter: Martin Sebor Date: 5 Jan 2003
23.3.5.1, p6 [lib.bitset.cons] talks about a generic character having the value of 0 or 1 but there is no definition of what that means for charT other than char and wchar_t. And even for those two types, the values 0 and 1 are not actually what is intended -- the values '0' and '1' are. This, along with the converse problem in the description of to_string() in 23.3.5.2, p33, looks like a defect remotely related to DR 303.
http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/lwg-defects.html#303
23.3.5.1:
-6- An element of the constructed string has value zero if the
corresponding character in str, beginning at position pos,
is 0. Otherwise, the element has the value one.
23.3.5.2:
-33- Effects: Constructs a string object of the appropriate
type and initializes it to a string of length N characters.
Each character is determined by the value of its
corresponding bit position in *this. Character position N
?- 1 corresponds to bit position zero. Subsequent decreasing
character positions correspond to increasing bit positions.
Bit value zero becomes the character 0, bit value one becomes
the character 1.
Also note the typo in 23.3.5.1, p6: the object under construction is a bitset, not a string.
Proposed resolution:
Change the constructor's function declaration immediately before 23.3.5.1 [lib.bitset.cons] p3 to:
template <class charT, class traits, class Allocator>
explicit
bitset(const basic_string<charT, traits, Allocator>& str,
typename basic_string<charT, traits, Allocator>::size_type pos = 0,
typename basic_string<charT, traits, Allocator>::size_type n =
basic_string<charT, traits, Allocator>::npos,
charT zero = charT('0'));
Change the first two sentences of 23.3.5.1 [lib.bitset.cons] p6 to: "An element of the constructed string has value 0 if the corresponding character in str, beginning at position pos, is zero. Otherwise, the element has the value 1.
Change the declaration of the to_string member function immediately before 23.3.5.2 [lib.bitset.members] p33 to:
template <class charT, class traits, class Allocator>
basic_string<charT, traits, Allocator>
to_string(charT zero = charT('0'), charT one = charT('1')) const;
Change the last sentence of 23.3.5.2 [lib.bitset.members] p33 to: "Bit value 0 becomes the character zero, bit value 1 becomes the character one.
Change 23.3.5.3 [lib.bitset.operators] p8 to:
Returns:
os << x.template to_string<charT,traits,allocator<charT> >(
use_facet<ctype<charT> >(os.getloc()).widen('0'),
use_facet<ctype<charT> >(os.getloc()).widen('1'));
Rationale:
There is a real problem here: we need the character values of '0' and '1', and we have no way to get them since strings don't have imbued locales. In principle the "right" solution would be to provide an extra object, either a ctype facet or a full locale, which would be used to widen '0' and '1'. However, there was some discomfort about using such a heavyweight mechanism. The proposed resolution allows those users who care about this issue to get it right.
We fix the inserter to use the new arguments. Note that we already fixed the analogous problem with the extractor in issue 303.
Section: 27.6.2.3 [lib.ostream::sentry] Status: Open Submitter: Martin Sebor Date: 5 Jan 2003
17.4.4.8, p3 prohibits library dtors from throwing exceptions.
27.6.2.3, p4 says this about the ostream::sentry dtor:
-4- If ((os.flags() & ios_base::unitbuf) && !uncaught_exception())
is true, calls os.flush().
27.6.2.6, p7 that describes ostream::flush() says:
-7- If rdbuf() is not a null pointer, calls rdbuf()->pubsync().
If that function returns ?-1 calls setstate(badbit) (which
may throw ios_base::failure (27.4.4.3)).
That seems like a defect, since both pubsync() and setstate() can throw an exception.
Proposed resolution:
[ The contradiction is real. Clause 17 says destructors may never throw exceptions, and clause 27 specifies a destructor that does throw. In principle we might change either one. We're leaning toward changing clause 17: putting in an "unless otherwise specified" clause, and then putting in a footnote saying the sentry destructor is the only one that can throw. PJP suggests specifying that sentry::~sentry() should internally catch any exceptions it might cause. ]
Section: 27.6.2.3 [lib.ostream::sentry] Status: Open Submitter: Martin Sebor Date: 5 Jan 2003
While reviewing unformatted input member functions of istream for their behavior when they encounter end-of-file during input I found that the requirements vary, sometimes unexpectedly, and in more than one case even contradict established practice (GNU libstdc++ 3.2, IBM VAC++ 6.0, STLPort 4.5, SunPro 5.3, HP aCC 5.38, Rogue Wave libstd 3.1, and Classic Iostreams).
The following unformatted input member functions set eofbit if they encounter an end-of-file (this is the expected behavior, and also the behavior of all major implementations):
basic_istream<charT, traits>&
get (char_type*, streamsize, char_type);
Also sets failbit if it fails to extract any characters.
basic_istream<charT, traits>&
get (char_type*, streamsize);
Also sets failbit if it fails to extract any characters.
basic_istream<charT, traits>&
getline (char_type*, streamsize, char_type);
Also sets failbit if it fails to extract any characters.
basic_istream<charT, traits>&
getline (char_type*, streamsize);
Also sets failbit if it fails to extract any characters.
basic_istream<charT, traits>&
ignore (int, int_type);
basic_istream<charT, traits>&
read (char_type*, streamsize);
Also sets failbit if it encounters end-of-file.
streamsize readsome (char_type*, streamsize);
The following unformated input member functions set failbit but not eofbit if they encounter an end-of-file (I find this odd since the functions make it impossible to distinguish a general failure from a failure due to end-of-file; the requirement is also in conflict with all major implementation which set both eofbit and failbit):
int_type get();
basic_istream<charT, traits>&
get (char_type&);
These functions only set failbit of they extract no characters, otherwise they don't set any bits, even on failure (I find this inconsistency quite unexpected; the requirement is also in conflict with all major implementations which set eofbit whenever they encounter end-of-file):
basic_istream<charT, traits>&
get (basic_streambuf<charT, traits>&, char_type);
basic_istream<charT, traits>&
get (basic_streambuf<charT, traits>&);
This function sets no bits (all implementations except for STLport and Classic Iostreams set eofbit when they encounter end-of-file):
int_type peek ();
Proposed resolution:
Informally, what we want is a global statement of intent saying that eofbit gets set if we trip across EOF, and then we can take away the specific wording for individual functions. A full review is necessary. The wording currently in the standard is a mishmash, and changing it on an individual basis wouldn't make things better. Dietmar will do this work.
Section: 20.4.1.1 [lib.allocator.members] Status: Ready Submitter: Markus Mauhart Date: 27 Feb 2003
20.4.1.1 [lib.allocator.members] allocator members, contains the following 3 lines:
12 Returns: new((void *) p) T( val)
void destroy(pointer p);
13 Returns: ((T*) p)->~T()
The type cast "(T*) p" in the last line is redundant cause we know that std::allocator<T>::pointer is a typedef for T*.
Proposed resolution:
Replace "((T*) p)" with "p".
Rationale:
Just a typo, this is really editorial.
Section: 20.1.5 [lib.allocator.requirements] Status: New Submitter: Markus Mauhart Date: 27 Feb 2003
I think that in par2 of 20.1.5 [lib.allocator.requirements] the last two lines of table 32 contain two incorrect type casts. The lines are ...
a.construct(p,t) Effect: new((void*)p) T(t) a.destroy(p) Effect: ((T*)p)?->~T()
.... with the prerequisits coming from the preceding two paragraphs, especially from table 31:
alloc<T> a ;// an allocator for T
alloc<T>::pointer p ;// random access iterator
// (may be different from T*)
alloc<T>::reference r = *p;// T&
T const& t ;
For that two type casts ("(void*)p" and "(T*)p") to be well-formed this would require then conversions to T* and void* for all alloc<T>::pointer, so it would implicitely introduce extra requirements for alloc<T>::pointer, additionally to the only current requirement (being a random access iterator).
Proposed resolution:
"(void*)p" should be replaced with "(void*)&*p" and that "((T*)p)?->" should be replaced with "(*p)." or with "(&*p)->".
Note: Actually I would prefer to replace "((T*)p)?->dtor_name" with "p?->dtor_name", but AFAICS this is not possible cause of an omission in 13.5.6 [over.ref] (for which I have filed another DR on 29.11.2002).
Section: 20.1.5 [lib.allocator.requirements], 20.4.1.1 [lib.allocator.members], Status: New Submitter: Markus Mauhart Date: 27 Feb 2003
This applies to the new expression that is contained in both par12 of 20.4.1.1 [lib.allocator.members] and in par2 (table 32) of 20.1.5 [lib.allocator.requirements]. I think this new expression is wrong, involving unintended side effects.
20.4.1.1 [lib.allocator.members] contains the following 3 lines:
11 Returns: the largest value N for which the call allocate(N,0) might succeed.
void construct(pointer p, const_reference val);
12 Returns: new((void *) p) T( val)
20.1.5 [lib.allocator.requirements] in table 32 has the following line:
a.construct(p,t) Effect: new((void*)p) T(t)
.... with the prerequisits coming from the preceding two paragraphs, especially from table 31:
alloc<T> a ;// an allocator for T
alloc<T>::pointer p ;// random access iterator
// (may be different from T*)
alloc<T>::reference r = *p;// T&
T const& t ;
Cause of using "new" but not "::new", any existing "T::operator new" function will hide the global placement new function. When there is no "T::operator new" with adequate signature, every_alloc<T>::construct(..) is ill-formed, and most std::container<T,every_alloc<T>> use it; a workaround would be adding placement new and delete functions with adequate signature and semantic to class T, but class T might come from another party. Maybe even worse is the case when T has placement new and delete functions with adequate signature but with "unknown" semantic: I dont like to speculate about it, but whoever implements any_container<T,any_alloc> and wants to use construct(..) probably must think about it.
Proposed resolution:
Therefore I think that "new" should be replaced with "::new" in both cases.
Section: 21.3.5.8 [lib.string::swap] Status: New Submitter: Beman Dawes Date: 25 Mar 2003
std::basic_string, 21.3 [lib.basic.string] paragraph 2 says that basic_string "conforms to the requirements of a Sequence, as specified in (23.1.1)." The sequence requirements specified in (23.1.1) to not include any prohibition on swap members throwing exceptions.
Section 23.1 [lib.container.requirements] paragraph 10 does limit conditions under which exceptions may be thrown, but applies only to "all container types defined in this clause" and so excludes basic_string::swap because it is defined elsewhere.
Eric Niebler points out that 21.3 [lib.basic.string] paragraph 5 explicitly permits basic_string::swap to invalidates iterators, which is disallowed by 23.1 [lib.container.requirements] paragraph 10. Thus the standard would be contradictory if it were read or extended to read as having basic_string meet 23.1 [lib.container.requirements] paragraph 10 requirements.
Yet several LWG members have expressed the belief that the original intent was that basic_string::swap should not throw exceptions as specified by 23.1 [lib.container.requirements] paragraph 10, and that the standard is unclear on this issue. The complexity of basic_string::swap is specified as "constant time", indicating the intent was to avoid copying (which could cause a bad_alloc or other exception). An important use of swap is to ensure that exceptions are not thrown in exception-safe code.
Note: There remains long standing concern over whether or not it is possible to reasonably meet the 23.1 [lib.container.requirements] paragraph 10 swap requirements when allocators are unequal. The specification of basic_string::swap exception requirements is in no way intended to address, prejudice, or otherwise impact that concern.
Proposed resolution:
In 21.3.5.8 [lib.string::swap], add a throws clause:
Throws: Shall not throw exceptions.
Section: 17.4.3.4 [lib.replacement.functions], 18.4.1 [lib.new.delete] Status: New Submitter: Matt Austern Date: 24 Apr 2003
The eight basic dynamic memory allocation functions (single-object and array versions of ::operator new and ::operator delete, in the ordinary and nothrow forms) are replaceable. A C++ program may provide an alternative definition for any of them, which will be used in preference to the implementation's definition.
Three different parts of the standard mention requirements on replacement functions: 17.4.3.4 [lib.replacement.functions], 18.4.1.1 [lib.new.delete.single] and 18.4.1.2 [lib.new.delete.array], and 3.7.3 [basic.stc.dynamic].
None of these three places say whether a replacement function may be declared inline. 18.4.1.1 [lib.new.delete.single] paragraph 2 specifies a signature for the replacement function, but that's not enough: the inline specifier is not part of a function's signature. One might also reason from 7.1.2 [dcl.fct.spec] paragraph 2, which requires that "an inline function shall be defined in every translation unit in which it is used," but this may not be quite specific enough either. We should either explicitly allow or explicitly forbid inline replacement memory allocation functions.
Proposed resolution:
Add a new sentence to the end of 17.4.3.4 [lib.replacement.functions] paragraph 3: "The program's definitions shall not be specified as inline."
Rationale:
The fact that inline isn't mentioned appears to have been nothing more than an oversight. Existing implementations do not permit inline functions as replacement memory allocation functions. Providing this functionality would be difficult in some cases, and is believed to be of limited value.
Section: 25.4 [lib.alg.c.library] Status: New Submitter: Ray Lischner Date: 08 Apr 2003
Section 25.4 [lib.alg.c.library] describes bsearch and qsort, from the C standard library. Paragraph 4 does not list any restrictions on qsort, but it should limit the base parameter to point to POD. Presumably, qsort sorts the array by copying bytes, which requires POD.
Proposed resolution:
Section: 23.2.4.2 [lib.vector.capacity] Status: New Submitter: Dave Abrahams Date: 27 Apr 2003
There is a possible defect in the standard: the standard text was never intended to prevent arbitrary ForwardIterators, whose operations may throw exceptions, from being passed, and it also wasn't intended to require a temporary buffer in the case where ForwardIterators were passed (and I think most implementations don't use one). As is, the standard appears to impose requirements that aren't met by any existing implementation.
Proposed resolution:
Replace 23.2.4.2 [lib.vector.capacity] paragraph 1 with:
1 Notes: Causes reallocation if the new size is greater than the old capacity. If no reallocation happens, all the iterators and references before the insertion point remain valid. If an exception is thrown other than by the copy constructor or assignment operator of T (or, if first and last satisfy the forward iterator requirements, an operation on first or last) there are no effects.
Section: 24.1 [lib.iterator.requirements] Status: New Submitter: Nathan Myers Date: 3 June 2003
Clause 24.1 [lib.iterator.requirements], paragraph 5, says that the only expression that is defined for a singular iterator is "an assignment of a non-singular value to an iterator that holds a singular value". This means that destroying a singular iterator (e.g. letting an automatic variable go out of scope) is technically undefined behavior. This seems overly strict, and probably unintentional.
Proposed resolution:
Change the sentence in question to "... the only exceptions are destroying an iterator that holds a singular value, or the assignment of a non-singular value to an iterator that holds a singular value."
Section: 24.1 [lib.iterator.requirements] Status: New Submitter: Nathan Myers Date: 3 June 2003
I've been discussing iterator semantics with Dave Abrahams, and a surprise has popped up. I don't think this has been discussed before.
24.1 [lib.iterator.requirements] says that the only operation that can be performed on "singular" iterator values is to assign a non-singular value to them. (It doesn't say they can be destroyed, and that's probably a defect.) Some implementations have taken this to imply that there is no need to initialize the data member of a reverse_iterator<> in the default constructor. As a result, code like
std::vector<std::reverse_iterator<char*> > v(7); v.reserve(1000);
invokes undefined behavior, because it must default-initialize the vector elements, and then copy them to other storage. Of course many other vector operations on these adapters are also left undefined, and which those are is not reliably deducible from the standard.
I don't think that 24.1 was meant to make standard-library iterator types unsafe. Rather, it was meant to restrict what operations may be performed by functions which take general user- and standard iterators as arguments, so that raw pointers would qualify as iterators. However, this is not clear in the text, others have come to the opposite conclusion.
One question is whether the standard iterator adaptors have defined copy semantics. Another is whether they have defined destructor semantics: is
{ std::vector<std::reverse_iterator<char*> > v(7); }
undefined too?
Note this is not a question of whether algorithms are allowed to rely on copy semantics for arbitrary iterators, just whether the types we actually supply support those operations. I believe the resolution must be expressed in terms of the semantics of the adapter's argument type. It should make clear that, e.g., the reverse_iterator<T> constructor is actually required to execute T(), and so copying is defined if the result of T() is copyable.
Issue 235, which defines reverse_iterator's default constructor more precisely, has some relevance to this issue. However, it is not the whole story.
The issue was whether
reverse_iterator() { }
is allowed, vs.
reverse_iterator() : current() { }
The difference is when T is char*, where the first leaves the member uninitialized, and possibly equal to an existing pointer value, or (on some targets) may result in a hardware trap when copied.
8.5 paragraph 5 seems to make clear that the second is required to satisfy DR 235, at least for non-class Iterator argument types.
But that only takes care of reverse_iterator, and doesn't establish a policy for all iterators. (The reverse iterator adapter was just an example.) In particular, does my function
template <typename Iterator> void f() { std::vector<Iterator> v(7); }
evoke undefined behavior for some conforming iterator definitions? I think it does, now, because vector<> will destroy those singular iterator values, and that's explicitly disallowed.
24.1 shouldn't give blanket permission to copy all singular iterators, because then pointers wouldn't qualify as iterators. However, it should allow copying of that subset of singular iterator values that are default-initialized, and it should explicitly allow destroying any iterator value, singular or not, default-initialized or not.
Related issue: 407
Proposed resolution:
Section: 27.8.1.7 [lib.ifstream.members], 27.8.1.10 [lib.ofstream.members] Status: New Submitter: Nathan Myers Date: 3 June 2003
A strict reading of 27.8.1 [lib.fstreams] shows that opening or closing a basic_[io]fstream does not affect the error bits. This means, for example, that if you read through a file up to EOF, and then close the stream and reopen it at the beginning of the file, the EOF bit in the stream's error state is still set. This is counterintuitive.
The LWG considered this issue once before, as issue 22, and put in a footnote to clarify that the strict reading was indeed correct. We did that because we believed the standard was unambiguous and consistent, and that we should not make architectural changes in a TC. Now that we're working on a new revision of the language, those considerations no longer apply.
Proposed resolution:
Section: 23.2.3.1 [lib.queue], 23.2.3.3 [lib.stack] Status: New Submitter: Hans Bos Date: 7 Jun 2003
Sections 23.2.3.1 [lib.queue] and 23.2.3.3 [lib.stack] list comparison operators (==, !=, <, <=, >, =>) for queue and stack. Only the semantics for queue::operator== (23.2.3.1 [lib.queue] par2) and queue::operator< (23.2.3.1 [lib.queue] par3) are defined.
Proposed resolution:
Section: 25.3.5 [lib.alg.set.operations] Status: New Submitter: Daniel Frey Date: 9 Jul 2003
25.3.5 [lib.alg.set.operations] paragraph 1 reads: "The semantics of the set operations are generalized to multisets in a standard way by defining union() to contain the maximum number of occurrences of every element, intersection() to contain the minimum, and so on."
This is wrong. The name of the functions are set_union() and set_intersection(), not union() and intersection().
Proposed resolution:
Change that sentence to use the correct names.
Section: 27.4.4.3 [lib.iostate.flags] Status: New Submitter: Martin Sebor Date: 10 Jul 2003
The Effects clause in 27.4.4.3 [lib.iostate.flags] paragraph 5 says that the function only throws if the respective bits are already set prior to the function call. That's obviously not the intent. The typo ought to be corrected and the text reworded as: "If (state & exceptions()) == 0, returns. ..."
Proposed resolution:
In 27.4.4.3 [lib.iostate.flags] paragraph 5, replace "If (rdstate() & exceptions()) == 0" with "If (state & exceptions()) == 0".
Section: 23.2.4.3 [lib.vector.modifiers] Status: New Submitter: Matt Austern Date: 19 Aug 2003
Consider the following code fragment:
int A[8] = { 1,3,5,7,9,8,4,2 };
std::vector<int> v(A, A+8);
std::vector<int>::iterator i1 = v.begin() + 3;
std::vector<int>::iterator i2 = v.begin() + 4;
v.erase(i1);
Which iterators are invalidated by v.erase(i1): i1, i2, both, or neither?
On all existing implementations that I know of, the status of i1 and i2 is the same: both of them will be iterators that point to some elements of the vector (albeit not the same elements they did before). You won't get a crash if you use them. Depending on exactly what you mean by "invalidate", you might say that neither one has been invalidated because they still point to something, or you might say that both have been invalidated because in both cases the elements they point to have been changed out from under the iterator.
The standard doesn't say either of those things. It says that erase invalidates all iterators and references "after the point of the erase". This doesn't include i1, since it's at the point of the erase instead of after it. I can't think of any sensible definition of invalidation by which one can say that i2 is invalidated by i1 isn't.
(This issue is important if you try to reason about iterator validity based only on the guarantees in the standard, rather than reasoning from typical implementation techniques. Strict debugging modes, which some programmers find useful, do not use typical implementation techniques.)
Proposed resolution:
In 23.2.4.3 [lib.vector.modifiers] paragraph 3, change "Invalidates all the iterators and references after the point of the erase" to "Invalidates iterators and references at or after the point of the erase".
Rationale:
I believe this was essentially a typographical error, and that it was taken for granted that erasing an element invalidates iterators that point to it. The effects clause in question treats iterators and references in parallel, and it would seem counterintuitive to say that a reference to an erased value remains valid.
Section: 27.6.1.4 [lib.istream.manip] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
According to 27.6.1.4, the ws() manipulator is not required to construct the sentry object. The manipulator is also not a member function so the text in 27.6.1, p1 through 4 that describes the exception policy for istream member functions does not apply. That seems inconsistent with the rest of extractors and all the other input functions (i.e., ws will not cause a tied stream to be flushed before extraction, it doesn't check the stream's exceptions or catch exceptions thrown during input, and it doesn't affect the stream's gcount).
Proposed resolution:
I propose that the manipulator be required to behave like an unformatted member function. It should not behave as a formatted member function since those set failbit in the sentry ctor according to DR195(*) and ws is explicitly forbidden from doing that (sure, it could clear the bit, but why bother?) (*) http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/lwg-defects.html#195
Section: 18.2.2 [lib.c.limits] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
Given two overloads of the function foo(), one taking an argument of type
int and the other taking a long, which one will the call foo(LONG_MAX)
resolve to? The expected answer should be foo(long), but whether that
is true depends on the #defintion of the LONG_MAX macro, specifically
its type. This issue is about the fact that the type of these macros
is not actually required to be the same as the the type each respective
limit.
Section 18.2.2 of the C++ Standard does not specify the exact types of
the XXX_MIN and XXX_MAX macros #defined in the <climits> and <limits.h>
headers such as INT_MAX and LONG_MAX and instead defers to the C standard.
Section 5.2.4.2.1, p1 of the C standard specifies that "The values [of
these constants] shall be replaced by constant expressions suitable for use
in #if preprocessing directives. Moreover, except for CHAR_BIT and MB_LEN_MAX,
the following shall be replaced by expressions that have the same type as
would an expression that is an object of the corresponding type converted
according to the integer promotions."
The "corresponding type converted according to the integer promotions" for
LONG_MAX is, according to 6.4.4.1, p5 of the C standard, the type of long
converted to the first of the following set of types that can represent it:
int, long int, long long int. So on an implementation where (sizeof(long)
== sizeof(int)) this type is actually int, while on an implementation where
(sizeof(long) > sizeof(int)) holds this type will be long.
This is not an issue in C since the type of the macro cannot be detected
by any conforming C program, but it presents a portability problem in C++
where the actual type is easily detectable by overload resolution.
Proposed resolution:
Specify the exact type of each XXX_MIN and XXX_MAX constant #defined in the header <climits>. Note that it is not possible to #define these macros so that they are usable in preprocessor arithmetic expressions, having any type other than char, int, unsigned int, long, and unsigned long. This means that the type of SHRT_MIN, for instance, has to be either int ot long in C++.
Section: 22.2.1.1.2 [lib.locale.ctype.virtuals] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The Effects and Returns clauses of the do_widen() member function of the ctype facet fail to specify the behavior of the function on failure. That the function may not be able to simply cast the narrow character argument to the type of the result since doing so may yield the wrong value for some wchar_t encodings. Popular implementations of ctype<wchar_t> that use mbtowc() and UTF-8 as the native encoding (e.g., GNU glibc) will fail when the argument's MSB is set. There is no way for the the rest of locale and iostream to reliably detect this failure.
Proposed resolution:
A partial solution might be to specify that ctype<wchar_t>::do_widen(char) must return WEOF to indicate a failure.
Section: 27.4.2.1.6 [lib.ios::Init] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The dtor of the ios_base::Init object is supposed to call flush() on the
6 standard iostream objects cout, cerr, clog, wcout, wcerr, and wclog.
This call may cause an exception to be thrown.
17.4.4.8, p3 prohibits all library destructors from throwing exceptions.
The question is: What should this dtor do if one or more of these calls
to flush() ends up throwing an exception? This can happen quite easily
if one of the facets installed in the locale imbued in the iostream
object throws.
Proposed resolution:
I'm not sure what the best approach is in this case -- ignore the exception and proceed with the cleanup (required by the clause that prohibits library dtors from throwing), abort right there and then, or allow the exception to propagate (and allow an unfriendly termination of the program).
Section: 27.6.1.1.2 [lib.istream::sentry] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
27.6.1.1.2, p2 says that istream::sentry ctor prepares for input if is.good() is true. p4 then goes on to say that the ctor sets the sentry::ok_ member to true if the stream state is good after any preparation. 27.6.1.2.1, p1 then says that a formatted input function endeavors to obtain the requested input if the sentry's operator bool() returns true. Given these requirements, no formatted extractor should ever set failbit if the initial stream rdstate() == eofbit. That is contrary to the behavior of all implementations I tested. The program below prints out eof = 1, fail = 0 eof = 1, fail = 1 on all of them.
#include <sstream>
#include <cstdio>
int main()
{
std::istringstream strm ("1");
int i = 0;
strm >> i;
std::printf ("eof = %d, fail = %d\n",
!!strm.eof (), !!strm.fail ());
strm >> i;
std::printf ("eof = %d, fail = %d\n",
!!strm.eof (), !!strm.fail ());
}
Comments from Jerry Schwarz (c++std-lib-11373):
Jerry Schwarz wrote:
I don't know where (if anywhere) it says it in the standard, but the
formatted extractors are supposed to set failbit if they don't extract
any characters. If they didn't then simple loops like
while (cin >> x);
would loop forever.
Further comments from Martin Sebor:
The question is which part of the extraction should prevent this from happening
by setting failbit when eofbit is already set. It could either be the sentry
object or the extractor. It seems that most implementations have chosen to
set failbit in the sentry [...] so that's the text that will need to be
corrected.
Proposed resolution:
A possible change might go in the Effects clause of the sentry ctor in
27.6.1.1.2, p1. At the end of the paragraph, i.e., after the text beginning
with
Effects: If is.good() is true, ...
add
Otherwise, if (is.rdstate() == ios_base::eofbit) is true,
the function calls is.setstate(failbit) (which may throw
ios_base::failure).
Another possible wording is
Otherwise, if is.eof() is true, the function calls
is.setstate(ios_base::failbit) (which may throw
ios_base::failure).
The difference between the two is that when failbit is set in is.exceptions(),
the first option will cause ios_base::failure to be thrown only during the
first call to the sentry ctor, while the second option will cause
ios_base::failure to be thrown each time the sentry ctor is invoked. I think
I like the first alternative better (i.e., there's no reason to call
setstate(failbit) if the bit is already known to be set).
The change does not address the possibility of just badbit being set.
Should it? I.e., should the sentry ctor set failbit if (only) badbit is set?
I haven't done a full survey of all the input functions to make sure the
change doesn't break something else. A second pair eyes would be helpful.
Survey posted in c++std-lib-11409:
Below is a survey of a number of popular implementations, including
classic iostreams. Three different sets of behavior exist (case 2
is the same as case 1 except that istream::sentry does not set
eofbit or failbit in an initially good stream object when its
attempt to extract whitespace fails).
The major difference seems to be in how badbit is treated in the
initial stream state. Most implementations still set failbit in
this case, but there are three implementations of classic
iostreams that do not. Since all surveyed implementations of
standard iostreams do set failbit in this case, I think we
should probably either require such behavior, or leave it
unspecified (in which case I would prefer to be explicit
about it to avoid any confusion).
0. Compaq C++ classic + standard, g++ 2.95.2 classic,
g++ 3.2 standard, HP aCC 3.45 standard, Rogue Wave 3.1.1,
Sunpro 5.5 classic + standard, IBM VAC++ 6.0 standard,
STLport 4.5 classic + standard
1. HP aCC 4.45 classic, SGI MIPSpro 7.3 classic, IBM VAC++ 6.0 classic
2. SGI MIPSpro 7.3 standard
0 1 2
=========+=====+=====+=====
B-- 00 | B-F B-- B-F
B-- 01 | B-F B-- B-F
B-- 10 | B-F B-- B-F
B-- 11 | B-F B-- B-F
-E- 00 | -EF -EF -EF
-E- 01 | -EF -EF -EF
-E- 10 | -EF -EF -EF
-E- 11 | -EF -EF -EF
--F 00 | --F --F --F
--F 01 | --F --F --F
--F 10 | --F --F --F
--F 11 | --F --F --F
--- 00 | --- --- ---
--- 01 | --- --- ---
--- 10 | -EF -EF ---
--- 11 | --- --- ---
BE- 00 | BEF BEF BEF
BE- 01 | BEF BEF BEF
BE- 10 | BEF BEF BEF
BE- 11 | BEF BEF BEF
B-F 00 | B-F B-F B-F
B-F 01 | B-F B-F B-F
B-F 10 | B-F B-F B-F
B-F 11 | B-F B-F B-F
BEF 00 | BEF BEF BEF
BEF 01 | BEF BEF BEF
BEF 10 | BEF BEF BEF
BEF 11 | BEF BEF BEF
-EF 00 | -EF -EF -EF
-EF 01 | -EF -EF -EF
-EF 10 | -EF -EF -EF
-EF 11 | -EF -EF -EF
=========+=====+=====+=====
^^^ ^^ ^^^ ^^^ ^^^
| || | | |
| || +-----+-----+-- final stream state (Bad, Eof, Good,
| || or '-')
| |+-------------------- noskiws argument to sentry or ipfx(int)
| +--------------------- value of (ios::skipws & rdflags()) != 0
+------------------------- initial stream state (Bad, Eof, Good,
'-')
#include <istream>
#include <streambuf>
#include <stdio.h>
struct mybuf: std::streambuf { };
int main ()
{
static const std::ios::iostate states[] = {
std::ios::badbit,
std::ios::eofbit,
std::ios::failbit,
std::ios::goodbit,
std::ios::badbit | std::ios::eofbit,
std::ios::badbit | std::ios::failbit,
std::ios::badbit | std::ios::eofbit | std::ios::failbit,
std::ios::eofbit | std::ios::failbit
};
for (int i = 0; i != sizeof states / sizeof *states; ++i) {
for (int skipws = 0; skipws != 2; ++skipws) {
for (int noskipws = 0; noskipws != 2; ++noskipws) {
mybuf sb;
std::istream strm (&sb);
strm.setstate (states [i]);
if (skipws)
strm.setf (std::ios::skipws);
else
strm.unsetf (std::ios::skipws);
// strm.ipfx (noskipws);
std::istream::sentry guard (strm, !!noskipws);
const std::ios::iostate state = strm.rdstate ();
printf ("%c%c%c %d%d -> %c%c%c\n",
states [i] & std::ios::badbit ? 'B' : '-',
states [i] & std::ios::eofbit ? 'E' : '-',
states [i] & std::ios::failbit ? 'F' : '-',
skipws, noskipws,
state & std::ios::badbit ? 'B' : '-',
state & std::ios::eofbit ? 'E' : '-',
state & std::ios::failbit ? 'F' : '-');
}
}
}
}
Section: 27.8.1 [lib.fstreams] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
7.19.1, p2, of C99 requires that the FILE type only be declared in <stdio.h>. None of the (implementation-defined) members of the struct is mentioned anywhere for obvious reasons. C++ says in 27.8.1, p2 that FILE is a type that's defined in <cstdio>. Is it really the intent that FILE be a complete type or is an implementation allowed to just declare it without providing a full definition?
Proposed resolution:
Change 27.8.1, p2 to say that FILE is declared in <cstdio> and a note saying that it's implementation-defined whether the type is complete or not.
Section: 27.5.2.1 [lib.streambuf.cons] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The reflector thread starting with c++std-lib-11346 notes that the class template basic_streambuf, along with basic_stringbuf and basic_filebuf, is copy-constructible but that the semantics of the copy constructors are not defined anywhere. Further, different implementations behave differently in this respect: some prevent copy construction of objects of these types by declaring their copy ctors and assignment operators private, others exhibit undefined behavior, while others still give these operations well-defined semantics.
Note that this problem doesn't seem to be isolated to just the three
types mentioned above. A number of other types in the library section
of the standard provide a compiler-generated copy ctor and assignment
operator yet fail to specify their semantics. A (possibly still
incomplete) list of such types along with the headers they are defined
in is provided below, courtesy of Howard Hinnant:
<limits>
numeric_limits
<stdexcept>
logic_error
domain_error
invalid_argument
length_error
out_of_range
runtime_error
range_error
overflow_error
underflow_error
<utility>
pair
<functional>
unary_function
binary_function
plus
minus
multiplies
divides
modulus
negate
equal_to
not_equal_to
greater
less
greater_equal
less_equal
logical_and
logical_or
logical_not
unary_negate
binary_negate
binder1st
binder2nd
pointer_to_unary_function
pointer_to_binary_function
mem_fun_t
mem_fun1_t
mem_fun_ref_t
mem_fun1_ref_t
const_mem_fun_t
const_mem_fun1_t
const_mem_fun_ref_t
const_mem_fun1_ref_t
<memory>
allocator<void>
allocator // operator= only, which also isn't required by Allocator Requirements
raw_storage_iterator
<string>
char_traits<char>
char_traits<wchar_t>
<locale>
ctype_base
ctype
ctype_byname
ctype<char>
ctype_byname<char>
codecvt_base
codecvt
codecvt_byname
num_get
num_put
numpunct
numpunct_byname
collate
collate_byname
time_base
time_get
time_get_byname
time_put
time_put_byname
money_get
money_put
money_base
moneypunct
moneypunct_byname
messages_base
messages
messages_byname
<queue>
queue
priority_queue
<stack>
stack
<vector>
vector<bool>::reference // copy ctor only
<map>
map::value_compare
multimap::value_compare
<bitset>
bitset::reference // copy ctor only
bitset
<iterator>
iterator_traits
iterator_traits<T*>
iterator_traits<const T*>
iterator
input_iterator_tag
output_iterator_tag
forward_iterator_tag
bidirectional_iterator_tag
random_access_iterator_tag
reverse_iterator
back_insert_iterator
front_insert_iterator
insert_iterator
istream_iterator // operator= only
ostream_iterator // operator= only
istreambuf_iterator
istreambuf_iterator::proxy
ostreambuf_iterator
<complex>
complex<float> // copy ctor only
complex<double> // copy ctor only
complex<long double> // copy ctor only
<valarray>
slice
gslice
<ios>
ios_base // addressed by TC issue 50
ios_base::failure
ios_base::Init
fpos
<streambuf>
basic_streambuf
<istream>
basic_istream
basic_iostream
<ostream>
basic_ostream
<sstream>
basic_stringbuf
basic_istringstream
basic_ostringstream
basic_stringstream
<fstream>
basic_filebuf
basic_ifstream
basic_ofstream
basic_fstream
<strstream>
strstreambuf
istrstream
ostrstream
strstream
Proposed resolution:
The standard ought to be clarified to either explicitly disallow copy ctors and assignment operators for these types, or to specify what the exact semantics of these functions are.
Section: 17.4.3.1 [lib.reserved.names] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
It has been suggested that 17.4.3.1, p1 may or may not allow programs to explicitly specialize members of standard templates on user-defined types. The answer to the question might have an impact where library requirements are given using the "as if" rule. I.e., if programs are allowed to specialize member functions they will be able to detect an implementation's strict conformance to Effects clauses that describe the behavior of the function in terms of the other member function (the one explicitly specialized by the program) by relying on the "as if" rule.
Proposed resolution:
While I think programs should be allowed to explicitly specialize member functions of standard templates, I don't find it reasonable to expect implementations to follow the "as if" rule to the letter. I propose to add a clause to chapter 17 saying that where a function is described in terms of another (non-virtual) function using the "as if" rule, an implementation may substitute another (non-virtual) function call as long as the effects on the object remain the same as in the absence of any specializations of the called function in the program.
Section: 27 [lib.input.output] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
A third party test suite tries to exercise istream::ignore(N) with a negative value of N and expects that the implementation will treat N as if it were 0. Our implementation asserts that (N >= 0) holds and aborts the test.
I can't find anything in section 27 that prohibits such values but I don't see what the effects of such calls should be, either (this applies to a number of unformatted input functions as well as some member functions of the basic_streambuf template).
Proposed resolution:
I propose that we add to each function in clause 27 that takes an argument, say N, of type streamsize a Requires clause saying that "N >= 0." The intent is to allow negative streamsize values in calls to precision() and width() but disallow it in calls to streambuf::sgetn(), istream::ignore(), or ostream::write().
Section: 17.3.1.1 [lib.structure.summary] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The text in 17.3.1.1, p1 says:
"Paragraphs labelled "Note(s):" or "Example(s):" are informative, other
paragraphs are normative."
The library section makes heavy use of paragraphs labeled "Notes(s),"
some of which are clearly intended to be normative (see list 1), while
some others are not (see list 2). There are also those where the intent
is not so clear (see list 3).
List 1 -- Examples of (presumably) normative Notes:
20.4.1.1, p3, 20.4.1.1, p10, 21.3.1, p11, 22.1.1.2, p11, 23.2.1.3, p2,
25.3.7, p3, 26.2.6, p14a, 27.5.2.4.3, p7.
List 2 -- Examples of (presumably) informative Notes:
18.4.1.3, p3, 21.3.5.6, p14, 22.2.1.5.2, p3, 25.1.1, p4, 26.2.5, p1,
27.4.2.5, p6.
List 3 -- Examples of Notes that are not clearly either normative
or informative:
22.1.1.2, p8, 22.1.1.5, p6, 27.5.2.4.5, p4.
None of these lists is meant to be exhaustive.
Proposed resolution:
One possible solution is to identify all the paragraphs marked "Note(s):"
that are intended to be normative and either remove the labeling or replace
it with some other word, such as "Effects:" to make the normative intent
clear.
Another possible solution is to change 17.3.1.1, p1 and remove the mention
of Notes being informative, and then go through all the paragraphs marked
"Note(s):" that are intended to be informative and make change preserving
their informative nature.
Section: 20.4.3 [lib.temporary.buffer] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The standard is not clear about the requirements on the value returned from a call to get_temporary_buffer(0). In particular, it fails to specify whether the call should return a distinct pointer each time it is called (like operator new), or whether the value is unspecified (as if returned by malloc). The standard also fails to mention what the required behavior is when the argument is less than 0.
Proposed resolution:
In the first case, I propose that the function behave like malloc, i.e.,
If the size of the space requested is zero, the behavior is
implementation-defined: either pair(0, 0) is returned, or the behavior is
as if the size were some nonzero value, except that the returned pointer
is not dereferenceable.
In the second case, I propose that the function fail by returning pair(0, 0).
Section: 25.1.9 [lib.alg.search], 25.2.5 [lib.alg.fill], 25.2.6 [lib.alg.generate] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The complexity requirements for these function templates are incorrect
(or don't even make sense) for negative n:
25.1.9, p7 (search_n):
Complexity: At most (last1 - first1) * count applications
of the corresponding predicate.
25.2.5, p3 (fill_n):
Complexity: Exactly last - first (or n) assignments.
25.2.6, p3 (generate_n):
Complexity: Exactly last - first (or n) assignments.
In addition, the Requirements or the Effects clauses for the latter two
templates don't say anything about the behavior when n is negative.
Proposed resolution:
My proposed fix is to:
Change 25.1.9, p7 to
Complexity: At most (last1 - first1) * count applications
of the corresponding predicate if count is non-negative,
or 0 otherwise.
Change 25.2.5, p2 from
Effects: Assigns value through all the iterators in the
range [first, last) or [first, first + n).
to
Effects: Assigns value through all the iterators in the
range [first, last), or [first, first + n) if n is non
negative, none otherwise.
Change 25.2.5, p3 to:
Complexity: Exactly last - first (or n if n is non-negative,
or 0 otherwise) assignments.
Change 25.2.6, p1 from
Effects: Invokes the function object gen and assigns the return
value of gen though all the iterators in the range [first, last)
or [first, first + n).
to (notice the correction for the misspelled "through"):
Effects: Invokes the function object genand assigns the return
value of gen through all the iterators in the range [first, last),
or [first, first + n) if n is non-negative, or [first, first)
otherwise.
Change 25.2.6, p3 to:
Complexity: Exactly last - first (or n if n is non-negative,
or 0 otherwise) assignments.
The alternative is to require that n be non-negative. I chose
the other option in order to keep the requirements in line with
those of search_n which allows negative counts.
Section: 22.2.2.1.2 [lib.facet.num.get.virtuals] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The requirements specified in Stage 2 and reiterated in the rationale
of DR 221 (and echoed again in DR 303) specify that num_get<charT>::
do_get() compares characters on the stream against the widened elements
of "012...abc...ABCX+-"
An implementation is required to allow programs to instantiate the num_get
template on any charT that satisfies the requirements on a user-defined
character type. These requirements do not include the ability of the
character type to be equality comparable (the char_traits template must
be used to perform tests for equality). Hence, the num_get template cannot
be implemented to support any arbitrary character type. The num_get template
must either make the assumption that the character type is equality-comparable
(as some popular implementations do), or it may use char_traits<charT> to do
the comparisons (some other popular implementations do that). This diversity
of approaches makes it difficult to write portable programs that attempt to
instantiate the num_get template on user-defined types.
Proposed resolution:
Specify a mechanism whereby the num_get template may compare objects of
user-defined character types. Several possible solutions, each having
some drawbacks, were discussed in a thread on c++std-lib@research.att.com
starting with c++std-lib-10623.
One possibility is to require that user-defined character types be equality
comparable, thus obviating tye need for the char_traits::eq() function. This
solution would make the requirements imposed on user-defined character types
by the locale templates inconsistent with those of basic_string.
Another option is to require that programs that instantiate num_get on
user-defined types provide an explicit specialization of the std::char_traits
template on that type and require num_get to use it. This solution would
render the basic_string Traits template parameter useless.
Yet another option is to specify that when the num_get facet is instantiated
on a user-defined character type, say charT, that defines the nested type
charT::traits_type, num_get should use it to do comparisons. Otherwise,
num_get should use std::char_traits<charT>. This solution, while more
robust than the first two, might set the unwarranted expectation that
a program that provides two traits classes for the same character type
will have meaningful semantics.
Finally, arguably the most flexible solution is to remove the requirement
from Stage 2 that num_get compare the widened set of character literals and
instead have the template compare the narrowed characters read from the
input sequence. This issue proposes to adopt this last option as the
resolution.
Section: 21.3.5.5 [lib.string::erase] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
23.1.1, p3 along with Table 67 specify as a prerequisite for a.erase(q)
that q must be a valid dereferenceable iterator into the sequence a.
However, 21.3.5.5, p5 describing string::erase(p) only requires that
p be a valid iterator.
This may be interepreted as a relaxation of the general requirement,
which is most likely not the intent.
Proposed resolution:
Either clarify 21.3.5.5, p5 to say that the iterator is required to be valid and dereferenceable or drop the requirement altogether and rely on the general container requirements outlined in Table 67.
Section: 27.4.4.3 [lib.iostate.flags] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The Effects clause in 27.4.4.3, p5 describing the effects of a call to the ios_base member function clear(iostate state) says that the function only throws if the respective bits are already set prior to the function call. That's obviously not the intent. If it was, a call to clear(badbit) on an object for which (rdstate() == goodbit && exceptions() == badbit) holds would not result in an exception being thrown.
Proposed resolution:
The text ought to be changed from
"If (rdstate() & exceptions()) == 0, returns. ..."
to
"If (state & exceptions()) == 0, returns. ..."
Section: 26.3.2.4 [lib.valarray.sub] Status: New Submitter: Martin Sebor Date: 18 Sep 2003
The standard fails to specify the behavior of valarray::operator[](slice) and other valarray subset operations when they are passed an "invalid" slice object, i.e., either a slice that doesn't make sense at all (e.g., slice (0, 1, 0) or one that doesn't specify a valid subset of the valarray object (e.g., slice (2, 1, 1) for a valarray of size 1).
Proposed resolution:
Clarify the text of the standard to define what constitutes a valid and invalid slice object, and either specify the exact behavior of these member functions when passed an invalid slice object, or explicitly specify that the behavior is undefined under these conditions for the sake of efficiency.
Section: 20.1.5 [lib.allocator.requirements], 25 [lib.algorithms] Status: New Submitter: Matt Austern Date: 20 Sep 2003
Clause 20.1.5 [lib.allocator.requirements] paragraph 4 says that implementations are permitted to supply containers that are unable to cope with allocator instances and that container implementations may assume that all instances of an allocator type compare equal. We gave implementers this latitude as a temporary hack, and eventually we want to get rid of it. What happens when we're dealing with allocators that don't compare equal?
In particular: suppose that v1 and v2 are both objects of type vector<int, my_alloc> and that v1.get_allocator() != v2.get_allocator(). What happens if we write v1.swap(v2)? Informally, three possibilities:
1. This operation is illegal. Perhaps we could say that an implementation is required to check and to throw an exception, or perhaps we could say it's undefined behavior.
2. The operation performs a slow swap (i.e. using three invocations of operator=, leaving each allocator with its original container. This would be an O(N) operation.
3. The operation swaps both the vectors' contents and their allocators. This would be an O(1) operation. That is:
my_alloc a1(...);
my_alloc a2(...);
assert(a1 != a2);
vector<int, my_alloc> v1(a1);
vector<int, my_alloc> v2(a2);
assert(a1 == v1.get_allocator());
assert(a2 == v2.get_allocator());
v1.swap(v2);
assert(a1 == v2.get_allocator());
assert(a2 == v1.get_allocator());
Proposed resolution:
----- End of document -----