P1031 Low level file i/o library

[wg21bot]

P1031R2 Low level file i/o library (Niall Douglas)

[tituswinters]

LEWG defers to LEWG-I on this.

[brycelelbach]

Cologne 2019-07 LEWGI Minutes

P1031R2 Low Level File I/O: Direction Review

Champion: Niall Douglas

Minute Taker: Ruslan Arutyunyan

Start Overview: 07-17 13:46

Start Review: 14:40

Start Polling: 14:45

POLL: We should promise more committee time to pursuing P1031R2, knowing that our time is scarce and this will leave less time for other work.

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
3	4	12	2	1

Attendance: 23

POLL: A modern I/O library should be one of the C++ committee's major goals, knowing that our time is scarce and this will leave less time for other work.

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
7	13	2	1	0

Attendance: 23

POLL: We can tolerate the use of polymorphic inheritance for handles in P1031R2 (which has standard library ABI implications).

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
0	1	11	8	2

Attendance: 23

POLL: Focus on the portable subset of this proposal (which will be a much more limited subset).

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
0	6	8	5	4

Attendance: 23

POLL: Knowing what we know now, we should promise more committee time to evaluating P1031R2, knowing that our time is scarce and this will leave less time for other work.

Strongly For	Weakly For	Neutral	Weakly Against	Strongly Against
4	5	9	3	1

Attendance: 23

End: 15:14

CONSENSUS: Bring a revision of P1031R2, with no polymorphic class hierarchy for handles, to LEWGI for further direction review. Christopher Kohlhoff will help the author prepare a presentation for the Belfast meeting.

[ogiroux]

SG1 in Cologne
P1031 includes a large SG1 aspect, to even just survey for issues
SF F N A SA
8 5 0 0 0 - Unanimous

[ned14]

@ogiroux Can I ask what SG1's interest is in reviewing this paper in Belfast? Happy to do so, but we shall need to choose a specific aspect to focus upon.

[ogiroux]

Your paper implies a consistency model, observable from multiple execution agents. If you intend to spell it out more in a subsequent revision, we can wait for this. If you don't know what to write, we could discuss that. If you don't know what to discuss, then I can remove you from the schedule. It is up to you, at what rate you want to make progress.

[ned14]

Your paper implies a consistency model, observable from multiple execution agents. If you intend to spell it out more in a subsequent revision, we can wait for this.

As I mentioned in my last appearance at SG1, we cannot standardise what we do not control. POSIX defines the consistency model for concurrent i/o to POSIX files. P1031 simply doesn't get in the way of any such concurrency semantics as chosen by the implementation, it does not define them.

file_handle looks like it will pass LEWG-I, and be forwarded to LEWG. map_handle implements shared memory, but before it has a chance of leaving LEWG-I we shall need some mechanism for telling the abstract machine when the content within a range of bytes holding a live object may change arbitrarily.

SG12 gave strong feedback on how best to implement that, so I reckon it ought to go back to SG12 for approval before proceeding.

I do think SG1 will get looped in, but it remains a bit early, and in any case any individual class may get rejected or showstoppered. It seems inefficient for SG1 to do work which may be wasted.

That said, all additional committee time is welcome, at any stage. If you have nothing higher priority to spend SG1 time upon, I am very sure I can find stuff for you to review.

[brycelelbach]

Your paper implies a consistency model, observable from multiple execution agents. If you intend to spell it out more in a subsequent revision, we can wait for this.

As I mentioned in my last appearance at SG1, we cannot standardise what we do not control. POSIX defines the consistency model for concurrent i/o to POSIX files. P1031 simply doesn't get in the way of any such concurrency semantics as chosen by the implementation, it does not define them.

file_handle looks like it will pass LEWG-I, and be forwarded to LEWG. map_handle implements shared memory, but before it has a chance of leaving LEWG-I we shall need some mechanism for telling the abstract machine when the content within a range of bytes holding a live object may change arbitrarily.

SG12 gave strong feedback on how best to implement that, so I reckon it ought to go back to SG12 for approval before proceeding.

I do think SG1 will get looped in, but it remains a bit early, and in any case any individual class may get rejected or showstoppered. It seems inefficient for SG1 to do work which may be wasted.

That said, all additional committee time is welcome, at any stage. If you have nothing higher priority to spend SG1 time upon, I am very sure I can find stuff for you to review.

Note that LEWG won't look at P1031 or P1883 until SG1 has forwarded them.