Argument size computation for foreign ffi exports is likely wrong.

Summary

In GHC.HsToCore.Foreign.C:mkFExportCBits we compute the size of arguments like this:

aug_arg_size = sum [ widthInBytes (typeWidth rep) | (_,_,_,rep) <- aug_arg_info]

While this is accurate as far as I can tell the only thing we use this for is if we use stdcall foreign exports we annotate the function with the argument size. I believe stdcall expects arguments to be padded to multiples of the platforms word size, however the current code will give the exact size. So there might be a missmatch.

However as I understand it:

• stdcall is only a thing on 32bit
• It's basically only a thing on windows
• It's not the default calling convention
• I'm not sure the annotation is actually used/relevant for resolving symbols.
• If someone uses it they should easily be able to swap to another calling convention.

So I don't think it's worth investigating it further, but I figured I would document it in case anyone comes across this issue.