Also, generatize space.enforest and space.transformer, and add functions like expr_meta.relative_precedence and expr_meta.ends_parse.
Few Rhombus users will care about these mechansisms, but the upshot is that the |> pipeline operator parses it's right-hand side as an "immediate callee" so that it can pass along static information from it's left-hand side. This enables more propagation of static information than would normally happen with a function call. For example,
"apple" |> fun (x): x[0]
and
"apple" |> (_[0])
and
"apple" |> (((fun (x): x[0])))
all work with use_static, because fun and #%parens (both in functon-shorthand mode and in grouping mode) are bound as immediate callee forms that end up communicating to x that it's a string.
Much of this commit is about generalizing the enforestation and transformer pipelines to support syntax-class arguments and deliver those arguments to macros. That generalization turned out not to be necessary for immediate callees, since they fit into the transformer protocol, amd enough plumbing was in place already for transformers. Nevertheless, the changes and exploration toward immediate callees suggested the generalization, and it's bound to be useful in the near future.
The protocol for parsing immediate callees and falling back to expression mode does not fit nicely into the Rhombus parsing framework. For example, in x |> (_ + 1)(2), the (_ + 1) is just to the right of |>, but it's not the immediate callee. To make these kinds of forms work out, #%parens and similar need expr_meta.ends_parse to "look ahead" for parsing. The functionality seems likely to be useful in other contexts.
Also, generatize
space.enforestandspace.transformer, and add functions likeexpr_meta.relative_precedenceandexpr_meta.ends_parse.Few Rhombus users will care about these mechansisms, but the upshot is that the
|>pipeline operator parses it's right-hand side as an "immediate callee" so that it can pass along static information from it's left-hand side. This enables more propagation of static information than would normally happen with a function call. For example,and
and
all work with
use_static, becausefunand#%parens(both in functon-shorthand mode and in grouping mode) are bound as immediate callee forms that end up communicating toxthat it's a string.Much of this commit is about generalizing the enforestation and transformer pipelines to support syntax-class arguments and deliver those arguments to macros. That generalization turned out not to be necessary for immediate callees, since they fit into the transformer protocol, amd enough plumbing was in place already for transformers. Nevertheless, the changes and exploration toward immediate callees suggested the generalization, and it's bound to be useful in the near future.
The protocol for parsing immediate callees and falling back to expression mode does not fit nicely into the Rhombus parsing framework. For example, in
x |> (_ + 1)(2), the(_ + 1)is just to the right of|>, but it's not the immediate callee. To make these kinds of forms work out,#%parensand similar needexpr_meta.ends_parseto "look ahead" for parsing. The functionality seems likely to be useful in other contexts.