Small tweak for Julia 1.12 world age change

[Keno]

Julia 1.12 will clean up when implicit world age increments happen. In particular, there will no longer be implicit increments within the same statement at toplevel. However, we are likely retaining implicit increments before each statement within @testset. This PR makes a small rearrangements to be compatible with this change. See JuliaLang/julia#56509.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 67.35%. Comparing base (b70e247) to head (e56d18e).
Report is 7 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #128   +/-   ##
=======================================
  Coverage   67.35%   67.35%           
=======================================
  Files          21       21           
  Lines         729      729           
=======================================
  Hits          491      491           
  Misses        238      238           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[THinnerichs]

Hey @Keno, thanks for pointing that out. Sorry for the naive questions, but does that only affect @testset?
We do quite some Julia magic with eval(..) ( see e.g.

HerbSearch.jl/src/program_iterator.jl

Line 74 in b70e247

check = :(eval($mod.$super) <: HerbSearch.ProgramIterator ||

)
and I am not sure what the difference to the fixed @testsets is.

[Keno]

Hey @Keno, thanks for pointing that out. Sorry for the naive questions, but does that only affect @testset?

It's not specific to @testset. The current situation is:

statement1(#= Maybe World Age Increment =#; statement2());
#= Maybe World Age Increment =# statement3();

The Maybe here is a problem, because it's unpredictable and depends on whether we're running in the interpreter or the compiler. 1.12 gives it definite semantics, which are:

At toplevel:

statement1(#= No World Age Increment =#; statement2());
#= Yes World Age Increment =# statement3();

Within begin/end at top level:

being
statement1(#= No World Age Increment =#; statement2());
#= No World Age Increment =# statement3();
end

but @testset goes back to pretending it's toplevel:

@testset being
statement1(#= No World Age Increment =#; statement2());
#= Yes World Age Increment =# statement3();
end

Which is why despite being inside a begin/end block, just making it separate statements is sufficient for getting the implicit increment. You can also always get an explicit increment with @Core.latestworld. This is all at toplevel of course. Both pre- and post- 1.12, there's never an implicit world age increment in local scope.

That said, tests passed with just the changes in this PR. You are correct that if you want to observe the return of the eval inside the :block you linked within the same top-level statement, you need an explicit world-age increment afterwards. Since I didn't see it break, I'm assuming you're not relying on that. My suggestion would be to let the 1.12 changes shake out a bit further and if you notice any world-age related issues in practice, let me know and we can take a look at whether there's any further changes required.

[ReubenJ]

Thanks @Keno for making all of these PRs related to the change—much appreciated!