MethodError: no method matching active_module() (leading to world age error), on Julia 1.11

[charleskawczynski]

We're seeing an odd MethodError on Julia 1.11 in ClimaCore CI (here):

field2arrays Unit Tests: Error During Test at /home/runner/work/ClimaCore.jl/ClimaCore.jl/test/MatrixFields/field2arrays.jl:6
  Got exception outside of a @test
  MethodError: no method matching active_module()
  You may have intended to import Base.active_module
  The applicable method may be too new: running in world age 26680, while current world is 27104.
  
  Closest candidates are:
    active_module(::REPL.LineEditREPL) (method too new to be called from this world context.)
     @ REPL /opt/hostedtoolcache/julia/1.11.0-beta2/x64/share/julia/stdlib/v1.11/REPL/src/REPL.jl:589
    active_module() (method too new to be called from this world context.)
     @ REPL /opt/hostedtoolcache/julia/1.11.0-beta2/x64/share/julia/stdlib/v1.11/REPL/src/REPL.jl:585
    active_module(::REPL.REPLDisplay) (method too new to be called from this world context.)
     @ REPL /opt/hostedtoolcache/julia/1.11.0-beta2/x64/share/julia/stdlib/v1.11/REPL/src/REPL.jl:591
    ...
  
  Stacktrace:
    [1] #invokelatest#2
      @ ./essentials.jl:1031 [inlined]
    [2] invokelatest
      @ ./essentials.jl:1028 [inlined]
    [3] active_module
      @ ./show.jl:519 [inlined]
    [4] show_type_name(io::IOBuffer, tn::Core.TypeName)
      @ Base ./show.jl:1058
    [5] show_datatype(io::IOBuffer, x::DataType, wheres::Vector{TypeVar})
      @ Base ./show.jl:1185
    [6] show_datatype
      @ ./show.jl:1094 [inlined]
    [7] _show_type(io::IOBuffer, x::Type)
      @ Base ./show.jl:978
    [8] show(io::IOBuffer, x::Type)
      @ Base ./show.jl:970
    [9] print(io::IOBuffer, x::Type)
      @ Base ./strings/io.jl:35
   [10] print_to_string(::String, ::Vararg{Any})
      @ Base ./strings/io.jl:148
   [11] string
      @ ./strings/io.jl:189 [inlined]
   [12] #s14#16
      @ ~/work/ClimaCore.jl/ClimaCore.jl/src/DataLayouts/DataLayouts.jl:172 [inlined]
   [13] var"#s14#16"(S::Any, Name::Any, ::Any, data::Any, ::Any)
      @ ClimaCore.DataLayouts ./none:0
   [14] (::Core.GeneratedFunctionStub)(::UInt64, ::LineNumberNode, ::Any, ::Vararg{Any})
      @ Core ./boot.jl:708
   [15] getproperty
      @ ~/work/ClimaCore.jl/ClimaCore.jl/src/DataLayouts/DataLayouts.jl:182 [inlined]
   [16] coordinates_data
      @ ~/work/ClimaCore.jl/ClimaCore.jl/src/Spaces/Spaces.jl:83 [inlined]
   [17] 

This same code works on Julia 1.10.

[vtjnash]

It looks like there was an issue with #46866, where is wrongly used invokelatest instead of using applicable to determine if a particular method was safe to call

[vtjnash]

Note that the thing this generated function may be doing (string-ify a Type) may be causing UB as well, since that operation accesses a significant amount of non-constant global state (to pretty-ify the result), which may trigger arbitrary UB.

[JeffBezanson]

This doesn't seem to happen now in ClimaCore --- did you work around it? Still needs to be fixed, just checking.

[charleskawczynski]

I didn’t do anything specifically, no 🤷🏻‍♂️, but you’re right in that the error doesn’t seem to be showing up anymore.

[charleskawczynski]

If it's helpful, I could open a pull request in ClimaCore that reverts to the last commit where we saw the error.

[JeffBezanson]

Yes that would be helpful, thanks.

[charleskawczynski]

Alright, I opened CliMA/ClimaCore.jl#1904, let's see what happens.

Hm, it needs some fixes due to changes in upstream packages, I'll keep pushing fixes so long they're simple, and see if we can reproduce it. I'll comment on this issue again once it's working. Out of the office tomorrow, though, so I may not get back to it until Monday.

[sethaxen]

I'm seeing this error as well in https://github.com/arviz-devs/PosteriorStats.jl/blob/176faa772079c1e57b35fe6c5aa2aeb965de2354/src/summarize.jl#L404-L408. An example CI log is https://github.com/arviz-devs/PosteriorStats.jl/actions/runs/10247536730/job/28346942153#step:11:496

I don't have the nightly installed, but this should reproduce the error:

using PosteriorStats, Statistics, Test
x = randn(1_000, 4, 3)
stats1 = @inferred PosteriorStats.summarize(x, mean, std, median)

[vtjnash]

While we will likely look into fixes for this in some contexts, note that calling string on a Type from inside @generated has always been in the documented list of things that are UB to appear there and permitted to do anything (including corrupt your program and data). It looks like you might just want nameof there instead (without the generated function nonsense), with a helpful error if the user passes in a custom callable object that doesn't subtype Function

[charleskawczynski]

I'm seeing this error as well

Good to know, I've been struggling to get our reproducer working again since there seem to be a bunch of required changes in the way.

[JeffBezanson]

Should be fixed by #55316