SharedMemory Not usable in pythonAPI

[xavierYou]

this line “linker.define(store_r, "env", "memory", SharedMemory(engine, MemoryType(Limits(32768, 32768), shared=True)))” will get exception “incompatible types, LP_LP_wasmtime_sharedmemory instance instead of LP_wasmtime_sharedmemory instance”。

[alexcrichton]

Thanks for the report! Probably just a bug in the bindings here in this repository and a hole in the test coverage. Happy to take a PR to adjust this as necessary!

[alexcrichton]

Er sorry did not mean to close.

[xavierYou]

Thank you very much! really hope to see the fix soon! And by the way, are there any alternative ways to use SharedMemory? I believe the purpose of designing it was to enable zero-copy data exchange between different Stores. But without SharedMemory, right now I'm working within the same Store by restricting the starting addresses of different modules so that they can still access the same memory block. However, this is quite cumbersome and prone to crashes, so I'm really looking forward to seeing the fix soon. Thank you again!!

[alexcrichton]

I don't know of an alternative to that use case, no. This is just a bug that needs fixing, and shouldn't be too hard of a fix in theory.

[xavierYou]

Hello, I accidentally discovered that the following operation can solve the previous bug,

def _as_extern_patched(self):
union = ffi.wasmtime_extern_union(sharedmemory=self.ptr())
return ffi.wasmtime_extern_t(ffi.WASMTIME_EXTERN_SHAREDMEMORY, union)
SharedMemory._as_extern = _as_extern_patched

Then in PyAPI, you can create it using the following method
SharedMemory(engine, MemoryType(Limits(min_pages, max_pages), shared=True))
And imported it into store by linker.define，Do you think this method is okay? Will it bring other problems，thank u！

[alexcrichton]

I wouldn't recommend monkeypatching classes personally. Would you be up for sending that change as a PR?

[xavierYou]

ok i will try to do so，after i test it in my production environment