Show macro expansions for undeclared identifiers about saltwater HOT 5 OPEN

This is a good idea, but it's going to be really hard. Locations will have to be expanded to keep track of each macro along the way, which means they can no longer be Copy and there will suddenly be a lot of lifetime errors.

Maybe there should be some sort of lookup table for locations so that they don't have to be passed along with each struct? I don't think HashMap<dyn Hash, Location> is a thing, though.

from saltwater.

The Locations of #defines are stored in a map somewhere already, I assume? So could we instead add an Option<InternedStr> member to Token::Id which points to where it which macro replaced into it? Then if there's an error, we can do a recursive lookup until None.

from saltwater.

Or if the Locations of #defines aren't already stored somewhere, then lex::replace::Definitions could be changed from

pub type Definitions = HashMap<InternedStr, Definition>;

to

pub type Definitions = HashMap<InternedStr, (Definition, Location)>;

Not sure if this will mess with lifetimes.

from saltwater.

@hdamron17 sure, making it a HashMap<InternedStr, Locatable<Definition>> isn't hard. The hard part is tracing back from the replaced version to the original. By the time this error occurs in the parser, all context about macros has been lost, there's only tokens left and you can't tell if they originated from a macro or not.

from saltwater.

I guess a linked list isn't the worst thing since most replacements shouldn't be highly nested. I'm a little concerned about memory usage but we can always worry about performance later.

from saltwater.