Lann Martin (Aug 27 2024 at 19:53): Lann Martin (Aug 27 2024 at 19:53):Is it valid for a module to import the same symbol multiple times?
Alex Crichton (Aug 27 2024 at 20:05):So long as they all have the same signature that should be fine yeah
Lann Martin (Aug 27 2024 at 20:05):Should wit-component be able to handle that?
Lann Martin (Aug 27 2024 at 20:06):It seems to explicitly validate against it: https://github.com/bytecodealliance/wasm-tools/blob/2fdc2ed0de54e55f57ae0bc88273d2b674ce9bf4/crates/wit-component/src/validation.rs#L172
Lann Martin (Aug 27 2024 at 20:08):context: Grain 0.6.6 has duplicate imports for a hello world module
Lann Martin (Aug 27 2024 at 20:11):causing this to fail: https://github.com/fermyon/spin/blob/82008342a3ec8de7d0ac6f3850828b463c692160/crates/componentize/src/lib.rs#L174-L178
Alex Crichton (Aug 27 2024 at 20:31):Ah ok different layer, no at the component level core modules must have a single import per module/name combo
Alex Crichton (Aug 27 2024 at 20:32):I thought you meant the LLD/linking level where this is allowed
Lann Martin (Aug 27 2024 at 20:34):OK so this is just something you can't handle by adapting?
Lann Martin (Aug 27 2024 at 20:34):wasmtime run can run the (preview1) module directly
Alex Crichton (Aug 27 2024 at 20:35):correct, the component model has a validation predicate for core wasm modules which disallows this
Alex Crichton (Aug 27 2024 at 20:35):yes that's where this is a valid core wasm thing just not a valid thing in the component model
Alex Crichton (Aug 27 2024 at 20:37):there's quite a long discussion about this at https://github.com/WebAssembly/design/issues/1402 which ended up with today's rule where core wasm wasn't changed but the component model has extra validation
Lann Martin (Aug 27 2024 at 20:43):I don't suppose there is existing code to dedupe imports?
Alex Crichton (Aug 27 2024 at 20:43):not currently, no, but with wasm-encoder's new support for reencoding it might not be too too hard to whip up something
Lann Martin (Aug 27 2024 at 20:44)::thumbs_up: conceptually it shouldn't be hard for this particular case; all the duplicate imports have the same type index
Alex Crichton (Aug 27 2024 at 20:57):this is a simple possible implementation, although I've not tested it much but shows the shape of what such a transform might be
Lann Martin (Aug 27 2024 at 20:59):Thanks!
Lann Martin (Aug 27 2024 at 21:00):Should I take that as an invitation to PR this into wasm-tools? :smile:
Alex Crichton (Aug 27 2024 at 21:00):heh perhaps yeah, I haven't thought really about this in the context of wasm-tools
Alex Crichton (Aug 27 2024 at 21:01):there's all sorts of possible transformations you can do on wasm files and I'm not sure we'd want to pack them all into one tool which would probably have relatively sparse testing
Alex Crichton (Aug 27 2024 at 21:01):that being said I also don't really see a reason to not do it either, and it's easy enough to add tests and it shouldn't be too much of a burden to keep things compiling
Alex Crichton (Aug 27 2024 at 21:02):it might make sense to implement this as like wasm-tools transform foo.wasm --name --of --various --passes --to --apply where CLI flags such as --dedupe-imports would be allowed (or something like that)
Alex Crichton (Aug 27 2024 at 21:02):or maybe --pass dedupe-imports,validate,...
Alex Crichton (Aug 27 2024 at 21:03):where each step would take wasm an output wasm and that way we could pack future things in there relatively easily
Alex Crichton (Aug 27 2024 at 21:05):one thing I implemented long long ago for wasm modules was a "gc" pass which would remove unneeded imports/functions/globals/etc and that might be kind of neat to reimplement with wasm-encoder nowadays and it might be much easier
Alex Crichton (Aug 27 2024 at 21:05):https://crates.io/crates/wasm-gc
Last updated: Nov 22 2024 at 16:03 UTC