Errors can be reported as residual goals

[triska]

For example, in Scryer Prolog, we get:

?- throw(t).
   throw(t).
?- T =.. a.
   error(type_error(list,a),(=..)/2).
?- X =:= Y.
   error(instantiation_error,(is)/2).

Note that exceptions are reported "uniformly" as residual goals, much like constraints and answer substitutions, using the predicates error/2 and throw/1.

This may be worth considering in the API design. It is related to #4.

[guregu]

I'm not sure if we can express this in a nice way with the variable substitution format the response currently uses 🤔

IIRC, the way Tau does this is it reports either a map of substitutions or a goal corresponding to the thrown ball, but this leads to kind of an ugly API situation where the response shape can be totally different, so you have to use a helper function (pl.types.is_error in Tau). Being able to switch on the response status is a little bit nicer.

[guregu]

I never thought about errors being conceptually the same as residual goals though, that's actually kind of mind blowing, haha.

[guregu]

In the latest release you can use the Prolog format to get this.

% demo toplevel looks like this now
?- throw(t).
   throw(t).
?- T =.. a.
   error(type_error(list,a),(=..)/2).
?- X =:= Y.
   error(instantiation_error,number).

const goal = await pl.queryOnce("throw(t).", {format: "prolog"});
// goal = "throw(t)."

[guregu]

Closing as done because the Prolog format toplevel report answers the same as regular Trealla now. I would like to further improve the uniformity by splitting the variable dump: #20.