https://github.com/dotnet/coreclr/blob/master/src/jit/lower.cpp#L1723 assumed the caller has the same number of arguments as the callee to be fastTailCalled. There are many different permutations of caller and callee that will break this assumption while still being decided as a fastTailCall case.

An example on unix amd64 would be:

// 9 arguments
// 8 in registers
// 1 on stack (16 bytes)
caller(float, float, float, float, float, float, float, float, { structOfSize 16 })

// 10 arguments
// 8 in registers
// 2 on stack (16 bytes)
callee(float, float, float, float, float, float, float, float, { structOfSize 8 }, { structOfSize 8} )

category:correctness
theme:tail-call
skill-level:intermediate
cost:small