Define a method for hash(::Type, ::UInt)

[ararslan]

Currently, hash(::Type, ::UInt) uses objectid, which can have some odd behavior for types: in particular, subsequent identical type-valued variable definitions can have objectids which differ from the first such definition. This has some bizarre downstream effects when e.g. using types as the values of a Set or the keys of a Dict. See #49620 for examples.

There is an internal type_hash C function used for caching types but isn't exposed to Julia, as Jameson pointed out in the linked issue. This commit exposes it as jl_type_hash which is then used via ccall to define a method hash(::Type, ::UInt). This method then fixes #49620. Note, however, that this does not affect the differing objectids for otherwise identical types.

I've marked this as a bugfix and notably the linked issue is present at least as far back as Julia 1.6, so it could be worth backporting to release-1.6, release-1.9, and perhaps release-1.8 if we're going to do any more of those. I'm not sure whether that's allowed though for changes that add to the C API.

[vtjnash]

error looks unrelated (to this PR)

Error in testset error:
Test Failed at C:\buildkite-agent\builds\win2k22-amdci6-3\julialang\julia-master\julia-9249904fd3\share\julia\test\error.jl:116
  Expression: value <: Exception
   Evaluated: Main.Test70Main_errorshow.TypeCompareError <: Exception

[ararslan]

Marking for backport to 1.6 and 1.9 since it's a bugfix but feel free to remove the labels if it's not appropriate to backport due to the addition to the C API

[vtjnash]

I thought the C API for this was fairly recent, but if not, it seems worthwhile to backport

[ararslan]

I thought the C API for this was fairly recent, but if not, it seems worthwhile to backport

If it is then how do we fix #49620 for prior versions?

[KristofferC]

Is this intended:

julia> typeof(a)
Destroy{FockSpace{Symbol}, Symbol, Int64, Nothing}

julia> typeof(a')
Create{FockSpace{Symbol}, Symbol, Int64, Nothing}

julia> hash(a)
0xc35039ec05f5ae3b

julia> hash(a')
0xc35039ec05f5ae3b

julia> hash(a) == hash(a')
true

It causes https://github.com/qojulia/QuantumCumulants.jl/blob/7a00cbbdbc3c702668689900f96303607ed65c6a/test/test_fock.jl#LL8C1-L13C34 to fail with this backported.

[ararslan]

Is this intended:

I assume not, and I'm not sure why that's happening. @vtjnash, any ideas?

[vtjnash]

It is not wrong, but is fairly unexpected. We might not be bixmixing with the TypeName hash enough?

[KristofferC]

I droped this PR from backports 1.9 until the effects are better understood (#49975).

[nalimilan]

This triggers an annoying bug when building StatsBase docs (#49620) so it would be nice to find a way to backport it.