As we modularize the WASI APIs (see #98 for a strawman), it will be nice to transition using the global namespace for typenames into modules.
Permitting a typename construct inside module is easy. The bigger issue is dealing with scope, because suddenly there are non-local names you may want to refer to, e.g. we may want to move the definition of (typename $fd ...) into (module $wasi_filesystem ...), but other modules may define additional functions that take file descriptors.
I propose the following scoping rules:
- for now, all names are public. any module can use a typename defined in any other module. I don't anticipate needing rules for private identifiers, since witx is a description language for interfaces, and privacy is typically for the encapsulation of implementations.
: is a scope separator inside identifiers.$:fully:qualified:identifier has a leading :. $:wasi_filesystem:fd would refer to the fd name inside the wasi_filesystem module.- Identifiers without a leading
: search the local scope.
If we adopt those rules, it would make sense to introduce a leading : in identifiers inside modules that refer to typenames defined at the top-level scope. That is a pretty simple change.
This change will be more invasive to witx crate users. The witx::ast::Id type will be changed to contain a Vec<String> with the fully qualified name. The validator will need some work to track scopes, and hopefully we can keep the define-before-use rules of identifiers for now. (Recursive and mutually recursive types are ruled out by this rule during validation.) Luckily, there are only two witx crate users that I know of, and both of them have been good about merging my PRs when I break their stuff by changing witx, so we'll just stick with that process for now.
As we modularize the WASI APIs (see #98 for a strawman), it will be nice to transition using the global namespace for
typenames into modules.Permitting a
typenameconstruct insidemoduleis easy. The bigger issue is dealing with scope, because suddenly there are non-local names you may want to refer to, e.g. we may want to move the definition of(typename $fd ...)into(module $wasi_filesystem ...), but other modules may define additional functions that take file descriptors.I propose the following scoping rules:
:is a scope separator inside identifiers.$:fully:qualified:identifierhas a leading:.$:wasi_filesystem:fdwould refer to thefdname inside thewasi_filesystemmodule.:search the local scope.If we adopt those rules, it would make sense to introduce a leading
:in identifiers inside modules that refer to typenames defined at the top-level scope. That is a pretty simple change.This change will be more invasive to
witxcrate users. Thewitx::ast::Idtype will be changed to contain aVec<String>with the fully qualified name. The validator will need some work to track scopes, and hopefully we can keep the define-before-use rules of identifiers for now. (Recursive and mutually recursive types are ruled out by this rule during validation.) Luckily, there are only two witx crate users that I know of, and both of them have been good about merging my PRs when I break their stuff by changing witx, so we'll just stick with that process for now.