3263. Atomic waiting function calls should only be unblocked once

Section: 32.5.6 [atomics.wait] Status: New Submitter: Geoffrey Romer Opened: 2019-08-19 Last modified: 2020-09-06

Priority: 3

View other active issues in [atomics.wait].

View all other issues in [atomics.wait].

View all issues with New status.

Discussion:

It appears that in a conforming implementation, all but one wait() call on a given atomic object may block forever, regardless of any notify_one() calls, because in principle every notify_one() call could be considered to unblock the same single wait() call. Common sense suggests (and David Olsen confirms) that the intent is for each waiting function call to be (non-spuriously) unblocked by at most one notifying function call, but as far as I can tell the words never say that.

[2019-09-14 Priority set to 3 based on reflector discussion]

Proposed resolution:

This wording is relative to N4830.

1. Modify 32.5.6 [atomics.wait] as indicated:

   -?- All blocking and unblocking events on a single atomic object occur in a single total order that is consistent with the "happens before" partial order.

   -4- A call to an atomic waiting operation on an atomic object M is eligible to be unblocked by a call to an atomic notifying operation on M if it has not been unblocked, and there exist side effects X and Y on M such that:

   1. (4.1) — the atomic waiting operation has blocked after observing the result of X,

   2. (4.2) — X precedes Y in the modification order of M, and

   3. (4.3) — Y happens before the call to the atomic notifying operation.