Now Gauche starts to support continuation marks natively (only the basic srfi-157 is implemented and I'll expand it toward srfi-226). So far as I understood, continuation marks *is* the dynamic environment, without revealing too much internal details, isn't it? Previously, I managed various internal dynamic contexts (e.g. exception handlers, parameters) by dynamic-wind; when a continuation is invoked, before and after handlers work hard to restore the dynamic state. Once I reversed the design, that is, letting continuation frames to hold the tip of the current dynamic environment chain and simply restore it when popped, things seem to become a lot simpler.
On Tue, Oct 18, 2022 at 9:01 PM Shiro Kawai <shiro.ka...@gmail.com> wrote: > Now I started seeing the source of my confusion. The phrase "annotate the > most recent continuation frame": I interpret it as modifying the most > recent continuation frame to attach the annotation. But it's not---I only > need to keep the "current" dynamic list of marks, and when the next > continuation frame is pushed, that continuation frame remembers the tip of > the marks so that it will be restored when the frame is popped. > > It might be helpful to say "_effectively_ annotate the most recent > continuation frame" to indicate that annotation doesn't really involve > immediate direct connection between continuation frames and marks. > > If I understand "dynamically annotated", it is less likely that I would've > thought with-continuation-marks could modify the existing continuation > frames. But it might be helpful to add a hint that continuation frames are > immutable. > > > On Sun, Oct 16, 2022 at 7:59 PM Marc Nieper-Wißkirchen < > marc.nie...@gmail.com> wrote: > >> Am So., 16. Okt. 2022 um 12:35 Uhr schrieb Shiro Kawai < >> shiro.ka...@gmail.com>: >> > >> > On Sat, Oct 15, 2022 at 10:41 PM Marc Nieper-Wißkirchen < >> marc.nie...@gmail.com> wrote: >> >> >> >> Am So., 16. Okt. 2022 um 01:26 Uhr schrieb Shiro Kawai < >> shiro.ka...@gmail.com>: >> >> >> >> The sample implementation does it by not gluing the last activation >> >> record and the last set of marks to a continuation frame. This only >> >> happens, conceptually, for older frames. >> >> >> >> In the example below, CONT conceptually contains the latest frame >> >> without the continuation mark attached. Inside the evaluation of >> >> `with-continuation-marks`, the current continuation has changed, >> >> namely its latest frame has been replaced by one with a mark attached. >> > >> > >> > I see that is how reference implementation works, but is it the only >> interpretation of the spec? Notably, the behavior of >> with-continuation-marks replacing the latest frame is derived from the >> current spec? >> > >> > Probably I was not understanding "dynamically annotated" correctly. >> Does it mean that the annotation of the latest frame is effective within >> the dynamic extent of with-continuation-marks? >> >> Yes; this is what it is supposed to say. >> >> Continuation frames and continuations are effectively immutable. If >> (non-pure) mutation happens, I use language like "replace", "mutate", >> or "modify". >> >> If you think it will make things more transparent, I will explicitly >> add a sentence stating the immutability. >> >> Marc >> >
