On Mon, Jun 29, 2026 at 8:02 AM Matt Borland <matt@mattborland.com> wrote:
Rediscover means that the information was lost and then found again. The S/R authors clearly knew about this information since they wrote about it. I would say that it's not rediscovery and should be framed more as convergence.
The information was lost. That is what the earlier papers in this series establish. * P4094R1 shows that the executor unification in P0443 collapsed the three continuation primitives dispatch, post, and defer, into a single execute(F&&). * P4095R1 shows that P1525R0 analyzed that unified operation under the work framing alone: "eagerly submits" a callable "on an execution agent that the executor creates for it." The continuation primitives do not appear. * P4096R1 shows P2464R0 did the same. The continuation framing from Kohlhoff's P0113R0 was erased from the API surface and absent from both analyses. I coined the term Rationale Loss to describe the phenomenon when tacit knowledge which informed a paper is lost in subsequent papers from different authors on the same subject. Those three papers predate P3941R4. They identified the framing distinction and showed that the continuation framing had been lost. P3941R4 then required infallibility for schedulers used with affine_on: exactly the shape the continuation framing predicts! P4286 connects the two: the prediction held. The committee arrived back at the shape that was already there in P0113R0. That is rediscovery, not convergence. You converge on something new. You rediscover something that was already known and lost from the analysis.
The paper's argument is the rediscovery of the shape of the executor in the form of infallible scheduler from S/R. The core issue is that the networking TS was blocked over the same problem.
The paper's argument is that the framing analysis from P4094, P4095, and P4096 is validated by P3941R4. Those papers identified two readings of execute(F&&) - work framing and continuation framing - and showed that P1525R0 and P2464R0 used only the work framing. Under the work framing, infallibility is a defect. Under the continuation framing, it is the correct shape. P3941R4 requires infallibility for the exact role - scheduling a continuation - where the continuation framing says it belongs. The core issue is not that the Networking TS was blocked over "the same problem." It is that the constraint called a defect in 2021 was analyzed under the wrong framing, and the committee re-derived the same constraint when it needed to schedule continuations in task. It is not the same problem because under the continuation framing it is not a problem. The paper could probably be reworded to make this crystal clear, since the point obviously did not land. Thank you for your feedback! Best