# Inconsistent hashing of types containing \`Set\` or \`Array\` fields
**URL:** https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900
**Category:** General Usage
**Created:** [March 22, 2018, 4:11pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900 "2018-03-22T16:11:20Z")
**Posts on this page:** 6
**Page:** 1
### Author:  [@ExpandingMan](https://discourse.julialang.org/u/ExpandingMan)
#### Post date: [March 22, 2018, 4:11pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/1 "2018-03-22T16:11:20Z")
Consider
```julia
struct TestType
S::Set{Int}
end
t1 = TestType(Set([1,2]))
t2 = TestType(Set([1,2]))
```
Then I do
```julia
julia> hash(t1)
0x9e50ff0d6c1bbf02
julia> hash(t2)
0x8a1b971008dc34e8
```
Well, that’s kind of annoying, but at this point I’m starting to imagine reasons why this might be the case, but then I do
```julia
julia> hash(t1.S)
0xb612ae5b24500075
julia> hash(t2.S)
0xb612ae5b24500075
```
I’ve now discovered that this happens with types containing `Array` (only custom types of course! I didn’t notice this until I created a few and tested them). Since that’s a pretty important use case, I’m imagining there is a good reason for this behavior. Any idea what that might be?
---
### Author:  [@rdeits](https://discourse.julialang.org/u/rdeits)
#### Post date: [March 22, 2018, 4:16pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/2 "2018-03-22T16:16:40Z")
The fallback method for `hash()` just uses `object_id`. Try:
```julia
@edit hash(t1, UInt(0))
```
Even though `t1` and `t2` have the same contents, they are different non-bits structs so they have different identities. You might not see this effect with a bitstype struct because identical bitstype structs seem to have the same `object_id()`:
```julia
julia> struct BitsType
x::Int
end
julia> b = BitsType(1)
BitsType(1)
julia> @edit hash(b, UInt(0))
julia> Base.object_id(b)
0x0df673c5a50bcfa2
julia> Base.object_id(BitsType(1))
0x0df673c5a50bcfa2
```
---
### Author:  [@mbauman](https://discourse.julialang.org/u/mbauman)
#### Post date: [March 22, 2018, 4:24pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/3 "2018-03-22T16:24:38Z")
At a slightly higher level:
Your custom object could care more about _behavior_ than _contents_. In your example, `t1` and `t2` are completely independent, and can — in the future — diverge. This means that they’re not programmatically indistinguishable… so we take the conservative route and say that they’re neither `==` nor do they `hash` equal.
If you want them to behave as their contents do, then you need to decide which fields are important in their equality. It may be that not all fields are the same.
```julia
julia> struct TestType
S::Set{Int}
end
julia> t1 = TestType(Set([1,2]))
TestType(Set([2, 1]))
julia> t2 = TestType(Set([1,2]))
TestType(Set([2, 1]))
julia> t1 == t2
false
julia> push!(t1.S, 3)
Set([2, 3, 1])
julia> t1, t2
(TestType(Set([2, 3, 1])), TestType(Set([2, 1])))
```
---
### Author:  [@ExpandingMan](https://discourse.julialang.org/u/ExpandingMan)
#### Post date: [March 22, 2018, 4:49pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/4 "2018-03-22T16:49:20Z")
I guess my concept of `==` needs adjusting. I was thinking of it as being like mathematical = wherever possible. What you guys are talking about sounds more like `===` to me. Just because two objects _may_ not be equal at some future t \> t\_{0} doesn’t mean that they are not equal at t\_{0}, so I’d expect `==` to reflect this.
I understand that performance could get to be an issue here for hashing of large data, but it seems to me that most users would be conscious of this, so again, I was expecting default hashing to respect the naive concept of equality, and that it would be up to users to replace the `hash` function if it becomes too computationally intensive.
---
### Author:  [@StefanKarpinski](https://discourse.julialang.org/u/StefanKarpinski)
#### Post date: [March 22, 2018, 4:49pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/5 "2018-03-22T16:49:21Z")
Welcome to [https://github.com/JuliaLang/julia/issues/4648](https://github.com/JuliaLang/julia/issues/4648).
---
### Author:  [@ExpandingMan](https://discourse.julialang.org/u/ExpandingMan)
#### Post date: [March 22, 2018, 4:51pm UTC](https://discourse.julialang.org/t/inconsistent-hashing-of-types-containing-set-or-array-fields/9900/6 "2018-03-22T16:51:26Z")
Oh god, an open issue from 2013 😨. I can see that I’ve walked into a minefield 😆.