P3481 R3 std::execution::bulk() issues

[wg21bot]

P3481R0 Summarizing std::execution::bulk() issues (Lucian Radu Teodorescu, Lewis Baker, Ruslan Arutyunyan)

[ogiroux]

SG1 in Wroclaw

If we have chunking, we need to expose a control on the chunk size
SF F N A SA
1 2 3 1 1
No consensus
SA: in a standard facility, I think this can only lead to bad cross-platform experiences, we have
plenty of elite tools for the elite to go use to get whatever they really want

We need a version of the bulk API that presents the chunk to the callable in order to implement the
parallel algorithms
SF F N A SA
4 2 1 0 1
Consensus
SA: any quality implementation of the parallel algorithms on a scheduler will want to be customized
for that scheduler

We need a version of the bulk API that creates an execution agent per iteration
SF F N A SA
2 3 3 0 0
Unanimous consent

We believe / desire:

1. Bulk needs an execution policy to describe the callable, IF the scheduler also has an execution
   policy (for debugging for example) then a conservative choice should be used (seq is more
   conservative than par)
2. No SG1 concerns with the proposed exception handling
3. No change to status quo on default implementation of bulk being serial
4. No change to status quo on bulk having a predecessor
5. Forms of bulk needed:
   bulk -> f(i) -> bulk_chunked
   bulk_chunked -> f(b, e)
   bulk_unchunked -> f(i) "execution agent per iteration"
   SF F N A SA
   5 2 1 0 0
   Unanimous consent