Am Di., 25. Aug. 2020 um 09:20 Uhr schrieb Alex Shinn <xxxxxx@gmail.com>:

> Both of these examples are (in theory) fixed by not rewriting quoted data, and I've just pushed this.

I think the needed change is more complicated. Consider a pattern of
the form (quasiquote (quote PAT)). Here, you still want to rewrite
PAT. Moreover, the problem A. does not go away when the identifier
appears quasiquoted.

> Neither of these examples require match-letrec, match-let is what you want here.

Sure. These examples are test cases.

> However the cycle example fails anyway because the current match-let implementation doesn't handle cycles either.

Can you explain what this means? I checked the same example with
match-let and it worked. The comparison is done with equal? and this
works with cycles (at least in R7RS).

Note that I am not talking about cycles outside quoted s-expressions.

> Cyclic literals in source are asking for trouble.  They are likely to break in many implementations in many situations.

If I understand you correctly this argument sounds a bit like using
hygienic macros is asking for trouble because there are
implementations that do not implement them properly.

> But we shouldn't spend a lot of time supporting this in a sample implementation for something which will never happen.

In my opinion, if possible, a sample implementation should be correct
and not just in 99% of all cases.

> And I don't want to spend time rewriting (well, wiring through an extra parameter into every line of the file) match.scm,
> in order to support match-letrec, at least until I first see a real-world example where someone actually wants it.

In some sense, this shows that, while possible, syntax-rules are
inadequate for writing complex macros like match when you later want
to extend them. Local changes in behavior often require global changes
in the code. It would be nice to see a rewrite by some volunteer of
the match macros in terms of SRFI 148, which compiles everything to
syntax-rules.