Attempting to access `Vector{Core.Compiler.DomTreeNode} at index [0]` when optimizing a function

[KristofferC]

Repro:

julia> using GMT

julia> code_warntype(show, Tuple{Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, GMT.Gdal.IDataset}; optimize=true)
ERROR: BoundsError: attempt to access 124-element Vector{Core.Compiler.DomTreeNode} at index [0]
Stacktrace:
  [1] throw_boundserror(A::Vector{Core.Compiler.DomTreeNode}, I::Tuple{Int64})
    @ Core.Compiler ./essentials.jl:14
  [2] getindex
    @ ./essentials.jl:894 [inlined]
  [3] _dominates(domtree::Core.Compiler.GenericDomTree{false}, bb1::Int64, bb2::Int64)
    @ Core.Compiler ./compiler/ssair/domtree.jl:622
  [4] dominates
    @ ./compiler/ssair/domtree.jl:605 [inlined]
  [5] try_resolve_finalizer!(ir::Core.Compiler.IRCode, idx::Int64, finalizer_idx::Int64, defuse::Core.Compiler.SSADefUse, inlining::Core.Compiler.InliningState{…}, lazydomtree::Core.Compiler.LazyGenericDomtree{…}, lazypostdomtree::Core.Compiler.LazyGenericDomtree{…}, info::Core.Compiler.CallInfo)
    @ Core.Compiler ./compiler/ssair/passes.jl:1619
  [6] sroa_mutables!(ir::Core.Compiler.IRCode, defuses::Core.Compiler.IdDict{…}, used_ssas::Vector{…}, lazydomtree::Core.Compiler.LazyGenericDomtree{…}, inlining::Core.Compiler.InliningState{…})

Seems to be caused by #53739 because reverting that PR seems to fix it on the 1.11 backport branch.

[maleadt]

MWE:

mutable struct IFeatureLayer
    ptr::Ptr{Cvoid}
    function IFeatureLayer()
        layer = new()
        finalizer(destroy, layer)
    end
end
destroy(layer) = layer.ptr = 0
getgeomtype(layer) = ccall(0, UInt32, (Ptr{Cvoid},), layer.ptr)

function main()
    layer = IFeatureLayer()
    get(identity, getgeomtype(layer), getgeomtype(layer))
    layer.ptr
end
Base.code_ircode(main, Tuple{})

[maleadt]

Seems to be caused by #53739 because reverting that PR seems to fix it on the 1.11 backport branch.

Confirmed that by bisecting the above MWE on the master branch. (cc @topolarity)