Checking against signature (the shortcut) should make sure we dispatch to the right method

[ulysses4ever]

Code in question:

julia-type-stability-checker/src/StabilityCheck.jl

Lines 142 to 146 in ae53bd0

	# Step 2a: run type inference with the input type even if abstract
	# and party if we're concrete
	@debug "is_stable_method: check against signature (2a)"
	try
	if is_stable_call(func, sig_types)

is_stable_call will call code_typed with an abstract input type (potentially), which can return several results and this is not handled at all: is_stable_call will just take the first CodeInfo:

julia-type-stability-checker/src/utils.jl

Line 42 in ae53bd0

(_ #=code=#, res_type) = ct[1] # we ought to have just one method body, I think

This is sound for a concrete input type, but the shortcut uses an abstract one. So, is_stable_call should make sure that if there are several results from code_typed, it picks the right one.

[ulysses4ever]

I can't see why it would return several results when called with the signature types of a particular method. Only that method will be returned always, I think? I seem to recall that Jan V convinced me that it's a real issue, but now I can't recall how and why. We can run an experiment, of course. But priority High doesn't seem to be justified.

One idea: can it happen if the methods are unlmbiguous? But then the whole program doesn't make sense...