LoadError: type Nothing has no field _conj

[Yue-Zhengyuan]

On Julia 1.11.1, I get the following error when building TensorKit:

(@v1.11) pkg> precompile
Precompiling project...
  ✗ TensorKit → TensorKitChainRulesCoreExt
  0 dependencies successfully precompiled in 1 seconds. 466 already precompiled.

ERROR: The following 1 direct dependency failed to precompile:

TensorKitChainRulesCoreExt 

Failed to precompile TensorKitChainRulesCoreExt [74f4665c-2d76-51f1-bd11-a86ebbb150d0] to "/Users/zhengyuanyue/.julia/compiled/v1.11/TensorKitChainRulesCoreExt/jl_uJjNv9".
ERROR: LoadError: type Nothing has no field _conj
Stacktrace:
 [1] getproperty(x::Nothing, f::Symbol)
   @ Base ./Base.jl:49
 [2] top-level scope
   @ ~/.julia/packages/TensorKit/59D34/ext/TensorKitChainRulesCoreExt/TensorKitChainRulesCoreExt.jl:19
 [3] include
   @ ./Base.jl:557 [inlined]
 [4] include_package_for_output(pkg::Base.PkgId, input::String, depot_path::Vector{String}, dl_load_path::Vector{String}, load_path::Vector{String}, concrete_deps::Vector{Pair{Base.PkgId, UInt128}}, source::Nothing)
   @ Base ./loading.jl:2790
 [5] top-level scope
   @ stdin:5
in expression starting at /Users/zhengyuanyue/.julia/packages/TensorKit/59D34/ext/TensorKitChainRulesCoreExt/TensorKitChainRulesCoreExt.jl:1
in expression starting at stdin:

Package versions:

[07d1fe3e] TensorKit v0.12.6
⌅ [6aa20fa7] TensorOperations v4.1.1

[lkdvos]

Aha, I see that my "extension hacking" has finally come to bite me in the backside...

It looks like julia 1.11.1 made some changes to the order of loading the package extensions. Because of that, the TensorKitChainRulesCoreExt module is not finding the TensorOperationsChainRulesCoreExt module, thus missing the required definitions. In principle this is not too hard to fix, it's a matter of porting over the definitions of _conj and trivtuple, but then we need to decide if it is worth it to backport this to a version 0.12.7, or if we wait for the new version of TensorKit. Any thoughts @Jutho ?

EDIT: It seems like this change affects the whole chain of packages, so things like PEPSKit no longer compile because of this. I'll open an issue to check if this is intended Julia behavior, but otherwise we should probably get this "hotfixed" and port it to a version 0.12.7

[Jutho]

I think this is due to this: JuliaLang/julia#55589

[Jutho]

I was also just looking into this. I do think porting over the required definitions is probably the quickest fix, as we will be starting to get many bug reports as people upgrade to Julia 1.11.1.

[lkdvos]

Also see JuliaLang/julia#56204, we are not the only ones affected by this. I have to say that whether or not this gets resolved at the Julia level, it's safest to just avoid it altogether.

[Jutho]

Although; it may be harder than that. We also use this approach to overload internal methods from TensorOperationsChainRulesCoreExt, which will require a more substantial fix.

[lkdvos]

Ah right, I forgot about that... We did not do this yet in the 0.12 version though, so it might be possible to hotfix that one and rethink the approach for the new tensorkit versions

[Yue-Zhengyuan]

Now that TensorKitChainRulesCoreExt can build under Julia 1.11.1, should I close this issue, or leave it open to track further fundamental fixes?