Illegal inttoptr when using PrecompileTools on HPC

[pat-alt]

This is related to timholy/SnoopCompile.jl#368 (comment).

When trying to load a package using PrecompileTools on our university's HPC, I get the following error:

(v1.8) pkg> add "https://github.com/JuliaTrustworthyAI/CounterfactualExplanations.jl.git"#302-add-precompiletools

[...]

Precompiling project...
  ✗ CounterfactualExplanations
  0 dependencies successfully precompiled in 217 seconds. 379 already precompiled.
  1 dependency errored. To see a full report either run `import Pkg; Pkg.precompile()` or load the package

julia> using CounterfactualExplanations
[ Info: Precompiling CounterfactualExplanations [2f13d31b-18db-44c1-bc43-ebaf2cff0be0]
┌ Warning: Package artifacts have been serialized on a different Julia version and may not be compatible with your version.
└ @ CounterfactualExplanations ~/.julia/packages/CounterfactualExplanations/zmii4/src/artifacts_setup.jl:9
Illegal inttoptr
	  %magicptr = ptrtoint {} addrspace(10)* %129 to i64, !dbg !14

signal (6): Aborted
in expression starting at /home/paltmeyer/.julia/packages/CounterfactualExplanations/zmii4/src/precompile.jl:3

This happens when trying to use the branch linked to JuliaTrustworthyAI/CounterfactualExplanations.jl#303. Using a version of the same package that doesn't use PrecompileTools works without issues.

@timholy is this right place to open this issue or should I move to JuliaLang/julia?

[KristofferC]

is this right place to open this issue or should I move to JuliaLang/julia?

This is more a Julia issue so I moved it.

[vchuravy]

What is versioninfo() and how did the Julia version get built/installed?

[pat-alt]

We are running 1.8.2 on the cluster. It's installed as part of a software stack:

02:26:15 paltmeyer@login01 julia-hpc-for-dummies ±|main|→ module spider julia/1.8.2

------------------------------------------------------------------------------------------------------------------------------------------------------
  julia: julia/1.8.2
------------------------------------------------------------------------------------------------------------------------------------------------------

    You will need to load all module(s) on any one of the lines below before the "julia/1.8.2" module is available to load.

      2023r1
      2023r1-gcc11
      2023rc1
      2023rc1-gcc11
 
    Help:
      The Julia Language: A fresh approach to technical computing

I'm not familiar with the details but I could ask our cluster administrator.

Do you need the full stack trace? I didn't post that because I wasn't sure if I might be accidentally sharing sensitive stuff.

[vtjnash]

The cluster might be using a non-supported build of Julia, since this bug should be fixed in llvm-13.0.1+3 (6ba1d23) in the 1.8 release. Although it looks somewhat like the correct binary name for libLLVM in that stacktrace: llvm-13.0.1-luuedcob23h45ategu2xuvx4frzgch7u/lib/libLLVM-13jl.so

Duplicate of #45702

[vtjnash]

looking closer, I see that this was fixed in Backports for Julia 1.8.6, and you are only using 1.8.2

[pat-alt]

OK, thanks for looking into this!

[timholy]

@pat-alt, can you close this when you've verified it's working?

[pat-alt]

@timholy yes, will do that as soon as I've had a chance to try in on a newer Julia version on our cluster

[pat-alt]

I've installed version 1.9.3 on our cluster and can confirm this works now.