Why don't Pow and Div have hash field?

[bowenszhu]

SymbolicUtils.jl/src/types.jl

Lines 51 to 61 in dcbe861

	struct Div{T} <: BasicSymbolic{T}
	num::Any = 1
	den::Any = 1
	simplified::Bool = false
	arguments::Vector{Any} = EMPTY_ARGS
	end
	struct Pow{T} <: BasicSymbolic{T}
	base::Any = 1
	exp::Any = 1
	arguments::Vector{Any} = EMPTY_ARGS
	end

SymbolicUtils.jl/src/types.jl

Lines 228 to 256 in dcbe861

	function Base.hash(s::BasicSymbolic, salt::UInt)
	E = exprtype(s)
	if E === SYM
	hash(nameof(s), salt ⊻ 0x4de7d7c66d41da43)
	elseif E === ADD || E === MUL
	!iszero(salt) && return hash(hash(s, zero(UInt64)), salt)
	h = s.hash[]
	!iszero(h) && return h
	hashoffset = isadd(s) ? 0xaddaddaddaddadda : 0xaaaaaaaaaaaaaaaa
	h′= hash(hashoffset, hash(s.coeff, hash(s.dict, salt)))
	s.hash[] = h′
	return h′
	elseif E === DIV
	return hash(s.num, hash(s.den, salt ⊻ 0x334b218e73bbba53))
	elseif E === POW
	hash(s.exp, hash(s.base, salt ⊻ 0x2b55b97a6efb080c))
	elseif E === TERM
	!iszero(salt) && return hash(hash(s, zero(UInt)), salt)
	h = s.hash[]
	!iszero(h) && return h
	op = operation(s)
	oph = op isa Function ? nameof(op) : op
	h′ = hashvec(arguments(s), hash(oph, salt))
	s.hash[] = h′
	return h′
	else
	error_on_type()
	end
	end

For Add, Mul and Term, once hash() is called, the field hash is filled and then it doesn't need to recompute the hash down to the subtree when hash() is called later.

But Sym, Pow and Div don't share this benefit.

It would be inefficient for, for example, the following expression.

(((((a / b) ^ c) / d) ^ e) / f) ^ g