Hi, On Mon, Oct 28, 2024, at 6:19 PM, Arthur A. Gleckler wrote: > Marc Nieper-Wißkirchen, author of SRFI 248 > <https://srfi.schemers.org/srfi-248/>: *Minimal delimited continuations*, has > agreed to announcing a *last call* for this SRFI. He believes that it is > ready for finalization, but would like to get corrections and feedback before > we finalize it. *Feedback from Scheme implementers and maintainers is > particularly welcome.* >
I have briefly looked at the document, sample implementation, and discussion. One immediate question jumps out at me: what is the purpose of `empty-continuation?`? It doesn't seem to be implemented by the sample implementation or used in the test suite. Is it a necessary part of this SRFI? If an implementation provides built-in delimited continuations but doesn't provide `empty-continuation?` as a primitive, it seems like it might be difficult or inefficient to emulate. I also want to clarify the semantics of `with-unwind-handler`. The specification says that the anonymous exception handler it installs "applies *handler* on *obj* and *k* in the continuation and dynamic environment of the call to `with-unwind-handler`." That seems to mean that the anonymous handler aborts to the prompt before applying *handler* (though after capturing *k*)—and indeed, that's what the the sample implementation seems to do. But it is subtly different than the R6RS `with-exception-handler`, where the handler in called in the dynamic environment of the call to `raise`, "except that the current exception handler is the one that was in place when the handler being called was installed" (plus details about the handler returning that differ for `raise` and `raise-continuable`). (As an aside, I would love some clearer way to distinguish "handler" from the "anonymous handler.") Based on the note in https://srfi-email.schemers.org/srfi-248/msg/25615855/, it sounds like the semantics should also clarify that *k* is a composable continuation. Similarly, I'd like more precision about "the portion of the current continuation up to and including the call to `with-unwind-handler`". The use of "including" is confusingly similar to (but IIUC distinct from) the notion of continuation capture (not) including a prompt as in https://www-old.cs.utah.edu/plt/publications/icfp07-fyff.pdf: > With respect to the many composable control operators in the > literature, we have opted for a variant where capture does not > include the delimiting prompt, and composition does not introduce > a prompt. Previous work explores the design space related to this > choice in depth (Shan 2004; Kiselyov 2005a; Biernacki et al. 2006; > Dybvig et al. 2006); when taking into account space complexity, > only the design that does not include or add a prompt is known to > express the others (Dybvig et al. 2006). Hope this is useful, Philip
