Erroneous warning when using types in keyword argument function declaration

[mmiller-max]

If I have a function with a nested type delegation (two wheres), I get the following incorrect warning

julia> foo(;a::T) where {T<:AbstractVector{N}} where {N} = zero(N)
WARNING: method definition for foo at REPL[5]:1 declares type variable N but does not use it.
WARNING: method definition for foo##kw at REPL[5]:1 declares type variable N but does not use it.
foo (generic function with 1 method)

It is being used, as can be seen by these two executions of the function

julia> foo(a=[1.0])
0.0

julia> foo(a=[1])
0

If I do something similar without nesting the type, it works fine with no warning

julia> bar(;a::N) where {N} = zero(N)
bar (generic function with 1 methods)

Ran on 1.8.5 and latest 1.9 RC

julia> versioninfo()
Julia Version 1.8.5
Commit 17cfb8e65ea (2023-01-08 06:45 UTC)
Platform Info:
  OS: macOS (arm64-apple-darwin21.5.0)
  CPU: 8 × Apple M1
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, apple-m1)
  Threads: 1 on 4 virtual cores

julia> versioninfo()
Julia Version 1.9.0-rc2
Commit 72aec423c2a (2023-04-01 10:41 UTC)
Platform Info:
  OS: macOS (arm64-apple-darwin21.3.0)
  CPU: 8 × Apple M1
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, apple-m1)
  Threads: 1 on 4 virtual cores
Environment:
  JULIA_IMAGE_THREADS = 1

[vtjnash]

It seems to be an issue with the way lowering decided if a variable gets used (not checking recursively). If we try some other combinations, we can see it handles it correctly if it is entirely unused:

julia> foo(;a::N,b::N) where {M, N, T<:AbstractVector{N}} = zero(N)
WARNING: method definition for #foo#11 at REPL[12]:1 declares type variable T but does not use it.
WARNING: method definition for #foo#11 at REPL[12]:1 declares type variable M but does not use it.
WARNING: method definition for foo at REPL[12]:1 declares type variable N but does not use it.
WARNING: method definition for foo at REPL[12]:1 declares type variable N but does not use it.
foo (generic function with 1 method)

[fredrikekre]

This issue seem to apply also to optional positional arguments, not just keyword arguments:

julia> f(a::T = Int[]) where {N, T <: AbstractVector{N}} = zero(N)
WARNING: method definition for f at REPL[1]:1 declares type variable N but does not use it.
f (generic function with 2 methods)

[eford]

FWIW, I also encountered this issue.

[c42f]

This issue seem to apply also to optional positional arguments

I just noticed this bug while porting the code to JuliaLowering.jl and set out to prove it ... and landed here :)

This is fixed in the JuliaLowering.jl implementation - the fix, conceptually, is to traverse the graph of symbolic dependencies between type variables, only picking up type variables which are used directly in the function arguments, or transitive dependencies of such.

The broken flisp lowering code for default args case is here:

julia/src/julia-syntax.scm

Line 655 in 858cb62

(define (filter-sparams expr sparams)