P1029R1 [[move_relocates]] (Niall Douglas)

P1029R2 move = relocates (Niall Douglas)

The ABI break highlighted in R2 (when applying the feature to existing components) needs LEWG visibility.

What groups have seen this paper?

Can this bug please be updated with information about when and where it has been seen, and with links to the relevant minutes? Thanks.

According to the author, this paper is a subset of P1631 Object detachment and attachment #551.
P1144 Object relocation in terms of move plus destroy #43 is more ambitious than the present paper.

Cologne 2019-07 LEWGI Minutes

P1029R2 Move Relocation: Design Feedback

Champion: Niall Douglas

Minute Taker: Ruslan Arutyunyan

Start Overview: 07-17 15:19

Start Polling: 15:26

POLL: We understand and accept that = relocates changes the ABI of a type, and thus we will be unable to apply this feature to existing standard library facilities (such as unique_ptr) or remove it from new facilities without accepting a standard library ABI break.

Attendance: 23

End: 15:30

EWG-I in Belfast

Given WG21's limited time constraints, should we encourage the author to continue work in this direction?
|SF|F|N|A|SA|
|7|3|0|0|0|

Do we want an opt-in mechanism for propagation from bases/members (not necessarily in this paper)?
|SF|F|N|A|SA|
|3|5|2|0|0|

Should =relocates work with types that have virtual destructors?
|SF|F|N|A|SA|
|2|4|2|1|0|

Do we approve of the direction for permitting passing =relocates types into C functions?
|SF|F|N|A|SA|
|2|5|1|2|0|

Given the current state of the paper + our current feedback, do we believe this paper is ready for EWG?
|SF|F|N|A|SA|
|6|3|1|1|0|

P1029R3 move = bitcopies (Niall Douglas)

Please forward along with P1144 Object relocation in terms of move plus destroy #43. It would be good to merge papers.

Based on the email thread with @jfbastien, we should look at P1029 and P1144 together. I'm still waiting to hear back about whether a champion for P1144 will be available in Prague.

EWGI in Prague

We believe that P1029 and P1144 are
sufficiently different that they should be
advanced separately.

We concluded that P1029 needs a revision and another visit to EWGI, however it should not have its progress through the committee linked to P1144.

Through a mailing list discussion, the author has let me know that they do not intend to move this paper forward. I'm therefore closing this issue, with the expectation that we can re-open it if another person wants to champion it.

Part of the discussion revolved around [[trivial_abi]] ad the feedback received in this LLVM review and linked patches. There's substantial discussion to catch up to and summarize. Where it applies to this paper, the feedback needs to be addressed.