api.Memory.Size() overflows when the memory has the maximum 65536 pages about wazero HOT 6 OPEN

Thank you @EclesioMeloJunior for reporting this, and I confirmed that this is actually a bug in our API. As you described, when the memory has the maximum of 65536 pages, the buffer length becomes 2^32, which is larger than the maximum of uint32 which is the return type of api.Memory Size.

We could do either

1. Changing the return type of api.Memory Size to uint64
   • This is somewhat breaking change, though the impact should be really minimum because this is just a integer type conversion.
2. Adds another for SizeSafe() uint64 or whatever it is while putting deprecation comments on the api.Memory Size.
   • This won't be a breaking change

thoughts? @evacchi @ncruces @achille-roussel

from wazero.

what about checking in the Grow method for this limit? I mean, instead of allowing the buffer size to reach 2^32 (which overflows uint32 only by 1) check and limit the buffer size/cap to be (2^32) - 1 then it will bound to uint32 limits (and the Size will not have problems to calc/return the correct size). I would assume that the problem with this approach is that the latest page will lose 1 byte (65535 instead of 65536), right?

no that's not our option as that means wazero is not compliant with the Wasm spec.

from wazero.

@mathetake thanks for the response, I have a question regards the SizeSafe, what it will return in a case where the buffer size overflows uint32 limits? Also, I would like to add that wasmtime memory size function returns a u64

I would like to add a third option:

• what about checking in the Grow method for this limit? I mean, instead of allowing the buffer size to reach 2^32 (which overflows uint32 only by 1) check and limit the buffer size/cap to be (2^32) - 1 then it will bound to uint32 limits (and the Size will not have problems to calc/return the correct size). I would assume that the problem with this approach is that the latest page will lose 1 byte (65535 instead of 65536), right?

from wazero.

Since memory size can't be (2^32) - 1, I'd return that and document the corner case?
I'd maybe call the other method Size64, in preparation for 64 bit memories?

from wazero.

@mathetake @ncruces thanks for the clarifications, I took the freedom to create a draft PR that introduces the new method #1856

from wazero.

So we are looking to release wazero 2.0 next month, and I think we can make the breaking change to this rather than having another confusing api like Size64.

Tomorrow I will raise a PR to do so, which supersedes #1856. Thank you for your help anyway @EclesioMeloJunior

from wazero.