code_warntype now shows unstable SSA useref values in red

[arhik]

Addresses #54028

Only function calls are covered. Other nodes like return, phi etc are not covered.

[arhik]

Not full work. Just want to have a quick feedback on the output and code changes.

[arhik]

Preview of work on Pi, Upsilon, Phi, PhiC, GotoIfNot, ReturnNodes.

[arhik]

source code for above output

using Revise
using Debugger
Revise.track(Core.Compiler)

function goo(t, w)
	return sum(t) + w
end

function foo(x, z)
	y = x[1] + goo(x, z)
	return goo([sin(y), sin(z)], z)
end

code_warntype(foo, (Vector{Any}, Int), optimize=true)

[Zentrik]

To avoid having two definitions to forward when no unstable_ssa is passed, can you not just do

function print_stmt(io::IO, idx::Int, @nospecialize(stmt), used::BitSet, maxlength_idx::Int, color::Bool, show_type::Bool, unstable_ssa::BitSet=BitSet())

I haven't looked at this in much detail so perhaps I missed why that won't work.

Ditto for the other functions where you do the same thing.

[arhik]

Thats how I started initially. Then I ran into stack overflow issues very often at multiple locations. In interest of time, I had to plant this variation in signature to bypass those issues. I might have to take a look at that approach again since I have a better idea of this work now than when I started.

[arhik]

I also learnt just today that optional arguments and kwargs are different. optional arguments generate method instances with/without an optional argument. That might explain stack overflow issues I had.