Document Number:	WG21/N1108
	X3J16/97-0070
Date:	20 July 1997
Project:	Programming Language C++
Reply-To:	Sean A. Corfield
	sean@ocsltd.com

Clamage convened the meeting as chair at 9:05 GMT on Monday, 14 July 1997. Miller was vice-chair, and Corfield was the secretary.

BSI hosted the meeting.

Clamage noted that whereas the Nashua meeting was devoted to J16 resolution of US ballot comments, the London meeting would be devoted to WG21 resolution of all the NB ballot comments.

Corfield circulated an attendance list each day, which is attached as Appendix A of these minutes. Miller circulated a copy of the membership list (SD-2 = 97-0001R3) for members to make corrections.

Clamage reminded the attendees that this is a co-located meeting of WG21 and J16. (The joint membership is denoted WG21+J16 in these minutes.)

Clamage explained the voting rules:

• In straw votes, all WG21 technical experts may vote, even those who haven't attended previous WG21 meetings. A J16 attendee may vote only if he/she is the voting representative of a member organization that has met the X3's meeting attendance requirements. (The voting representative is the principal member, or an alternative if the principal is not present.) A WG21 technical expert who is also a J16 voting member still casts only one vote in a straw vote.
• In WG21 formal votes, only the head of each national delegation may vote.
• In J16 formal votes, only one representative from each J16 member organization may vote, and only then if the organization meets the attendance requirements.

Jean Stride, BSI, explained the facilities provided by the BSI.

Corfield explained the computing facilities provided by Object Consultancy Services Ltd and the Internet facilities provided by Demon Internet Services Ltd.

Deferred to 1.7.

Corfield said that he had not received any corrections for the minutes.

Motion by Dawes/Lajoie:

Motion passed J16: lots yes, 0 no, 0 abstain.

Deferred to 1.7.

Plum summarized the WG21 meeting discussions. [Note: see N1106 = 97-0068 for details.]

Plauger asked for the ISO ballot vote to be noted in the minutes:

Plum reiterated our goal of closing every issue at this meeting.

Benito gave the WG14 (C language) liaison report. ISO C is currently in the registration ballot. The ISO C committee have added inline, declarations in for loops and extended identifiers, all in a manner compatible with C++. He said they have also added a bool type compatible with C++, either as a keyword or as a macro.

Colvin introduced a revision of the paper from Abrahams on exception safety in the library.

Koenig said a few typographical bugs have been fixed since Nashua.

There were three Core Language Working Groups chaired by Lajoie, Adamczyk and Gibbons, and two Library Working Groups chaired by Dawes and Schwarz.

The committees recessed at 10:25am.

The committees reconvened at 16:00 on Wednesday afternoon.

The WG chairs summarised their progress.

Clamage said that WGs would present only controversial issues today.

Adamczyk presented the "major" issues from his group detailed in N1091 = 97-0053. Only one item generated any discussion.

Adamczyk explained the resolution of Canada comment 10, as described in N1084 = 97-0046 and noted that this passed with large majority in the combined Core WG.

Straw vote passed J16: 19 yes, 3 no, a few abstain.

Straw vote passed WG21: 3 yes, 1 no, 2 abstain. A count of W21 representatives showed Germany, Netherlands, France, Sweden, Canada, USA and UK were present. Japan was temporarily absent.

Stroustrup presented his proposal, N1043 = 97-0005, to change the handling of default arguments so that they are checked at instantiation instead of declaration.

Clamage explained the background to the proposal.

template< class T > void foo( const T& = T() );
struct C { C(int); };
C c;
foo( c );	// fails on T()

There was a library proposal considered that would introduce:

template< class U > U default_value()
	{ return U(); }
template< class T > void foo( const T& = default_value<T>() );

Gibbons said that Core III felt the language did not need to change to support this usage - it was an editorial issue over when an expression is "used". Plum established that no one actually objected to this clarification.

There was extensive discussion about how to characterize the proposal for a vote. Plauger said that Stroustrup's proposal needs a vote in its own right.

Straw vote on Stroustrup's proposal passed J16: 27 yes, 8 no.

Straw vote on Stroustrup's proposal passed WG21: 5 yes, 0 no, 2 abstain.

Straw vote on adding default_value<T>() to the library failed: 0 yes, lots no.

Straw vote on adding an unknown number of extra signatures to the library failed: 1 yes, lots no.

Dawes explained the proposed resolution to the problem of ambiguities between container member functions taking iterator ranges as parameters (and 'size_t, T' signatures). The proposed resolution is to add wording that the library should "do the right thing". There were no questions about this issue.

Dawes then discussed the proposal for introducing an exception safety policy to the library. This was not controversial.

Dawes then discussed the problem that pointers to C library functions had non-portable linkage. Plum asked for clarification. Dawes explained what the LWG decided:

• Signal handlers must be extern "C".
• There a few C library routines that take pointers to functions - these will be overloaded so that they take pointer to function arguments that are extern "C" and extern "C++" linkage.
• There will be normative encouragement that other C library functions "should" be extern "C++".

Dawes then explained the proposed restrictions on what signal handler functions could do (in answer to a German ballot comment). There were no questions.

Plum presented some work on whether C library routines should be in global namespace or in std namespace.

• Does using the new <cstdio> header allow users to declare, e.g., printf as a variable? No. The clarification is that such (external linkage) names will be reserved in global namespace (as well as std namespace).

There was quite a bit of discussion about why C library names were put into std in the first place.

• Does using the new <cstddef> header allow users to declare, e.g., size_t as a type? No. The clarification is that such names will be reserved in global namespace (as well as std namespace) and if the name is also defined as ::T it will be identical to std::T.

Straw vote in favor of adding both clarifications passed: 22 yes, 1 no, 11 abstain.

Plum volunteered to write this up as a formal proposal.

Lajoie presented N1094 = 97-0056 which proposes that operator new() and operator delete() can be declared static or inline.

Straw vote passed J16: 14 yes, 4 no, 16 abstain. Straw vote passed WG21: 3 yes, 0 no, 4 abstain.

Glassborow suggested that we defer a decision until we have Ireland's feedback on the proposal. Plum asked for a volunteer to liaise with Ireland's technical experts about the issue. Miller volunteered.

Lajoie explained the proposed clarification of when copies of objects can be elided, noting this answered an Australian comment. She said the WP is too broad and allows copies to be elided when they can still be used.

Lajoie said the two cases they wanted to explicitly allow are:

1. elision of copy of temporary values,
2. elision of copy of return value.

Preference straw vote:

Lajoie said Core would conduct further work to look at other optimizations.

Gibbons explained the proposal to remove template template arguments. The background is that having discovered some problems, the Core III WG investigated the issue and then became concerned that there were more problems lurking. Brück spoke out against removing a language feature at this stage and feels it would be better to address the German comment by specifying the semantics more fully. Unruh said that as originator of the comment, he would be satisfied by the addition of a single bullet that clarified how template template arguments interact with partial ordering.

Straw vote to remove template template arguments failed: 13 yes, 18 no.

Gibbons said the feature stays so we still have to address the German comment.

Gibbons also mentioned two issues which were explicitly rejected, both of which were considered unnecessary extensions.

Schwarz presented the resolutions from his group. He said they have resolved every issue and produced a list of all the differences. They have also fully specified money_punct in a paper that has been prepared at this meeting but it is too early to go for a straw vote on it yet.

Myers discussed an issue arising from overloading enums. Nelson showed a real example:

(enumval) == (charval) // is ambiguous because:
string::operator==( charval );
string::string( charval );

So the result was string( enumval ).operator==( charval ); A straw vote indicated that Core II should attempt to resolve this.

Plum asked WG chairs to draft responses to the NB comments (or delegate someone to do it for them). For comments with which we agreed, this will be easy, but for comments that we rejected the responses should indicate the due care and attention that the committee applied to the issue.

Myers asked what we should do about errors we find in the document. Dawes said they should go on an issue list for later resolution. Unruh asked that any major issues be raised more visibly. Clamage suggested the reflector be used to discuss any such issues.

A list of the papers which would constitute the formal motions was distributed. The majority of the papers consisted of troff diffs that should be applied to the WP.

Clamage said this had been a very productive meeting and thanked everyone for their cooperation.

Koenig asked Clamage to confirm that the next meeting agenda will include a vote to approve the WP once at the end of the meeting since we are producing an interim WP only. Clamage agreed. Clamage then counted J16 voting members (29 out of a possible 34) and WG21 voting members (8).

Core I

Core II

Core III

Library I

Library II

The following issues have been deferred and need resolution at or before the New Jersey meeting in November:

• Resolution of Germany's comments 9 and 10 about template template parameters, action Gibbons.
• Resolution of Ireland's comment 1 about the linkage of operator new and delete, action Lajoie, Miller.
• Further work on copy elision for the forwarding function optimization, action Lajoie.

Koenig reminded the committee of the outstanding problem with enum and overloading. This is a core issue (Core 889) and the WG21 representatives present in London were unanimous that the problem needs a resolution.

Dawes said that the LWG still needs to add explicit to some signatures (the exception and fpos constructors). This change is previously approved but not yet in the draft and addresses Germany's comment 191 and USA comment CD2-27-036.

Dawes noted there are also half a dozen "editorial" comments outstanding which include a couple of substantive but minor issues.

The committee will also consider further issues from the reflector and private email subject to unanimous consent of WG21.

Unruh raised the issue of auto_ptr again since this was controversial within WG21. He also noted that the resolution regarding C library and namespaces might not satisfy Germany.

Someone noted that the rvalue/lvalue overload resolution adjustment needs to be revisited after feedback from National Bodies.

None.

Koenig said the next meeting would be 9-14 November at Headquarters Plaza Hotel in Morristown, NJ hosted by AT&T. Clamage said the WG will complete resolution of CD2 ballot comments and produce the DIS for submission.

Miller said the usual schedules apply: 1^st August is the deadline for post-London mailing.

Plum said it is hard to decide an agenda for the March '98 meeting yet since we will not know the result of the JTC1 letter ballot on the FDIS until just before the meeting, which will be in Sofia Antipolis, SE France. The meeting will be hosted by the University of Nice.

Motion by Corfield/Dawes:

Motion passed WG21+J16: lots yes, 0 no.

The committees adjourned at 10:20 on Friday.

Minutes of the U.S. TAG Meeting (recorded by William M. Miller)

The U.S. TAG convened immediately following the adjournment of the joint meeting of J16 and WG21, chaired by J16 IR Clark Nelson. Miller volunteered to record the minutes, since J16 secretary Corfield is not a member of the U.S. TAG. 25 of the 29 TAG members with voting rights were present, satisfying quorum requirements.

Nelson stated that the only business for the TAG at this meeting was to select the U.S. delegation for the next WG21 meeting. He explained that there had been requests that Beman Dawes be added to the delegation to represent the Library subgroup and that Benito had volunteered to leave the delegation. The suggested delegation of Clamage, Dawes, Koenig, Nelson, and Stroustrup was approved unanimously.

The U.S. TAG meeting adjourned at 10:25 a.m.