implement field visibility #345
Conversation
| withNewScope c: | ||
| # copy toplevel scope status for exported fields |
There was a problem hiding this comment.
Exporting a symbol is only allowed if the current scope has kind ToplevelScope, so this is propagated from object declarations in toplevel scopes to the scope of their fields so that the fields can be exported. The only other use of scope kinds seems to be for whether or not to make global or local syms in identToSym which already unconditionally makes global syms for field symbols, so I thought it was sufficient to change it.
Generics in general also open a new scope though (the other place where this is done) which is a bit more murky, this is supposed to work:
type Foo[T] = object
field*: TBut this would compile even though it doesn't add the symbols to the top level scope:
type Foo[T] = (; const field* = 123; T)Maybe a ToplevelFieldScope kind instead?
We rarely if ever find the time to clean up things so it's better to clean it up right now. That said, I don't find it particularly ugly. |
Code is a bit ugly, can be cleaned up if necessary
Currently checks for the line info of the field access to compare modules. An alternative might be to track the current instantiated routine symbol in a stack which would include templates
Imported modules don't seem to compile to C yet so had to disable testing the working cases
The type conversion in:
is a workaround, the type of
x.privateis theTin the original definition ofGeneric. This only errors for templates because other routines usecommonTypesince #320, which matches anything to typevar expected types. Maybex.privatecan substitute the type ofprivateaccording to theInvokeTparams of the type ofx, this would have to be done for inherited types too.