Sam Clegg

Looking at the tests, it also looks like they are not actually written against the wasi API at all, but rather against some higher liker `std` API which kind of...

@devsnek, sure, I guess that makes sense. I do think its important to have a set of low level unit tests though that are written directly against the wasi API....

Having just looked at the wasi-sdk tests again I realize they are also written against wasi-libc higher level APIs, not directly against wasi APIs :(

Nice! Thanks for working on this. wasm-interp from the wabt repro is also able to run wasi (via uvwasi).. its only really been tested with the was-sdk tests so its...

> > Nice! Thanks for working on this. > > Hopefully we'll eventually have a canonical test suite from this. > > > wasm-interp from the wabt repro is also...

From https://reviews.llvm.org/D81689: "This new behavior is disabled when the input has an explicit call to __wasm_call_ctors, indicating code not expecting new-style command support." This means that if you export a...

Putting it another way: Assuming the new behaviour can easily be avoided for those that what the more transitional single-entry command (with extra exports that are not entry point) would...

IIRC wasi-libc's `_start` function should trigger the old behavior because it calls `__wasm_call_ctors` itself: https://github.com/WebAssembly/wasi-libc/blob/2057ce9262f76f7ef5a2002fa16da219e2176896/libc-bottom-half/crt/crt1.c#L9 Perhaps I'm missing something here? @sunfishcode ?

Ah I see, I that case perhaps its wasi-sdk, or clang it self that should have some kind of option for force the old style crt1 to be chosen.

Yes, clearly canonical_abi_mallc/realloc would never want to be wrapped in calls to ctors/dtors. Linker support for canonical ABI stuff is yet to be decided IIUC.. its still being thought about/worked...