Am Mo., 26. Juli 2021 um 15:14 Uhr schrieb John Cowan <xxxxxx@ccil.org>:

On Mon, Jul 26, 2021 at 8:09 AM Marc Nieper-Wißkirchen <xxxxxx@gmail.com> wrote:

 

SRFIs normally give a user-directed specification of what these things do, and put the implementation strategy into the non-normative "Implementation" section.  This SRFI is almost all implementation strategy.

Can you clarify what you mean? I don't think I say anything about how to implement these things.

The idea of quasi syntax objects is a brilliant way of discussing both syntax-case and ER macros with a common vocabulary.  But unless it is your claim that they are the *only* way to correctly implement ER (in which case the implementation merges with the specification), then I don't see how passages like the following can be interpreted except as (broadly drawn) implementation specifications:

At first, the input form is recursively fully unwrapped preserving any shared or cyclic structure. Then proc is called with three arguments, the fully unwrapped input form and two procedures rename and compare. It is an error if it does not return a quasi-syntax object. The symbols at the leaves of the quasi-syntax object returned are then replaced by their injections to yield the output form as a syntax object.

Doesn't look operational semantics always this way? I don't care how, say, er-macro-transformer is implemented as long as it behaves operationally equivalent.

To give one example outside SRFI 225: If you take a look at 12.1 of the R6RS Standard Libraries, the expansion algorithm is operationally described through Dybvig's marks and substitutions algorithm. While the R6RS expander can be implemented in this way (Chez Scheme, psyntax, Unsyntax) by directly transforming the semantics into an algorithm, it doesn't have to be (Larceny, Chibi Scheme).

Also, note that the predicates of being "wrapped" and "unwrapped" syntax objects have to fulfill some axioms but they are not fully defined in the sense that they allow only one model. For example, in some implementations "fully unwrapped" syntax objects can be just the same as "wrapped" syntax objects.

 

By the way, I think the use of `quasisyntax` from R7RS as an analogue of quasiquote and the term "quasi-syntax objects" are unrelated, and if so, distinguishing them solely by a hyphen is a mistake.

Yes, that's not ideal. Do you have a better prefix to name "almost syntax objects" in mind?