Not a good practice to put seed as the 1st argument in hash function

[bowenszhu]

SymbolicUtils.jl/src/types.jl

Lines 254 to 269 in e9a96bd

	const ADD_SALT = 0xaddaddaddaddadda % UInt
	const SUB_SALT = 0xaaaaaaaaaaaaaaaa % UInt
	const DIV_SALT = 0x334b218e73bbba53 % UInt
	const POW_SALT = 0x2b55b97a6efb080c % UInt
	function Base.hash(s::BasicSymbolic, salt::UInt)::UInt
	E = exprtype(s)
	if E === SYM
	hash(nameof(s), salt ⊻ SYM_SALT)
	elseif E === ADD || E === MUL
	!iszero(salt) && return hash(hash(s, zero(UInt)), salt)
	h = s.hash[]
	!iszero(h) && return h
	hashoffset = isadd(s) ? ADD_SALT : SUB_SALT
	h′ = hash(hashoffset, hash(s.coeff, hash(s.dict, salt)))
	s.hash[] = h′
	return h′

Note line 267 in the above code. hashoffset of type UInt, serving as a seed, is the first argument of the hash function. Although it still works, it somehow violates the original design principle of 2-argument hash functions.

For hash(x::Int64, h::UInt), Julia Base focuses on the first argument and merely does a scalar multiplication and subtraction to the second argument, which is probably not what we want.
https://github.com/JuliaLang/julia/blob/57b9b591e28181aee8ff985d0ace661b408577d7/base/hashing.jl#L88
https://github.com/JuliaLang/julia/blob/57b9b591e28181aee8ff985d0ace661b408577d7/base/hashing.jl#L79
https://github.com/JuliaLang/julia/blob/57b9b591e28181aee8ff985d0ace661b408577d7/base/hashing.jl#L44

[nsajko]

hash(::UInt, ::UInt) just combines the arguments in a noncommutative way, with a multiplication and subtraction as you say. So it's quite cheap, but you could save a cycle or two by using ⊻ for mixing hashoffset with the computed hash instead of calling hash(::UInt, ::UInt).