&-sigiled terms an possibly other emergent "features"

[gfldex]

The following example works in HEAD and likely did so for quite some time.

class __px {}
multi postcircumfix:<{ }>(__px, @a) { "interpolation of @a[]" }

constant term:<&px> = __px;

say "&px<1 2 3>"

Here, an &-sigiled term mimics a subroutine call where there is no subroutine to be called. I can't find any mentioning of &-sigiled terms in Roast. In fact, term: seems to be underspecced in general. This leaves the quetion, how we deal with such emergent "features". As I understand it, Roast was meant as a starting point and common ground for alternate/competing compiler implementations. One could argue that we already got one with new-disp/RakuAST. Should we just spec this and expect other implementations to follow?

I'm asking, because I just showed this example in #raku-beginner and am not sure if unspecced constructs belong there. It somehow feels wrong to do so.

If this is specced, we need to solve the ENODOC as well.

[raiph]

Aiui we want to avoid nailing anything down in roast unless we are very confident we want to do that for any given "feature" / test.

---

As for the particular example you've provided....

First, simplifying a hair:

multi postcircumfix:<{ }>(Nil, @a) { "interpolation of @a[]" }

constant term:<&px> = Nil;

say "&px<1 2>"

Aiui, in strings being interpolated with :f set to True (eg per default double quoted rules), a valid identifier string starting with a & is treated as being possibly a function call, and Raku then parses from the & onwards until code no longer validly parses as a Raku expression. Then it tries to compile it; if it turns out there's no symbol matching the putative identifier, then the expression is treated as literal text, not as Raku code. If there is a matching symbol, but it turns out not to be a routine, it's replaced by whatever it is bound to and then the expression is evaluated.

Imo this is adequately documented on the Q Lang page and there's no need to update roast.