Thread safe whitespace

[david-davies]

Closes #49

Re-implement mkSpace using an IORef to a map of threads to their white space parsers.

[j-mie6]

Eh, partition is still good: we will not "put back alive threads" into the map, just into the weak-thread map. We might miss a dead-thread who died during the alive list, but that's ok, the next thread along will pick them up. We can't be perfect here, as threads don't clean themselves up from the structure. We are just aiming to ensure some amount of fairness.

[j-mie6]

Ah right, you can clobber the threads who have joined the list since the partition...

[david-davies]

Strange bug atm.
In examples/ExprLang.hs, if I run parseProg "return (1 + 2)", the parser succeeds.
However, even though the program parser is defined in terms of expr, running parseExpr "(1 + 2)" fails (a now customised message) with the message:
Exception: GigaParsec.Internal.Token.Space.getMapRef: entry in `RefMap` not initialised for thread: 54

Somehow, the whitespace parsers are not being initialised in expr, but they are in program.
Should note that program is defined in terms of fully, which uses the lexical description, so perhaps that forces the initialisation.
Whereas expr is a precedence table that doesn't make an explicit reference to a lexical description.

[j-mie6]

Yes, without an explicit initSpace (or whatever it is), you need fully.

[david-davies]

Ahh right. Perhaps then we should make this explicit in the docs? (I don't think it is already)
Or is there a way to automatically ensure initSpace is called?

[j-mie6]

Can't ensure it, but the wiki docs would point this out, see here