InteractiveUtils: access bindings in latest world

[timholy]

InteractiveUtils provides a number of utilities for querying and accessing the contents of modules. Because these modules can gain new bindings, some callers of InteractiveUtils functions produce warnings

WARNING: Detected access to binding ... prior to its definition world.

One dramatic example occurs with the sequence

using OptimizationProblems: OptimizationProblems
using OptimizationProblems.ADNLPProblems
using ADNLPModels: ADNLPModel

when running with Revise, where it can produce hundreds of warnings. ADNLPProblems loads new problems via Requires, and starting with v3.13 Revise queries InteractiveUtils.subtypes to cache dependent fieldtypes. This triggers the warning, and because it runs in a separate task, the --depwarn=error trick unfortunately failed to produce a stacktrace. Thus this was quite difficult to track down.

Fixes timholy/Revise.jl#993

CC @aviatesk, @Keno

[timholy]

The main alternative to this approach might be to give names the option to cut off names at the current task's world age.

[Keno]

I think it'd prefer if all of these gained a world kwarg (ok to default to latest) so that the query happens consistently in one particular world (possibly captured on function entry).

[timholy]

@Keno updated. Perhaps the main thing to decide is whether get_world_counter or tls_world_age is the right default for InteractiveUtils. An argument to use get_world_counter (which is the current state of this PR) is that interactive tools might want to be comprehensive by default, whereas code callers (which have a lower penalty on verbosity) are more likely to want to narrow the range.

[Keno]

get_world_counter is the right default for InteractiveUtils.

[timholy]

Please check---world wasn't working as intended without it, pre-definition queries were still returning the name.

[Keno]

It's a behavioral change. E.g. for export'd bindings that were not assigned.

[timholy]

Better?

[Keno]

Maybe? Undefined and public is still an option. This is not really about what's correct or not, more about what these options map to. My general preference would be not to change that since external packages are surprisingly sensitive to this list.

[Keno]

If you want to filter something specific I'd rather add another boolean flag.

[timholy]

Fundamentally the issue is that the guard entries for newly-created bindings in a module get backdated to the module age:

julia> module M
       public z
       end
Main.M

julia> age = Int(Base.get_world_counter())
39879

julia> b = GlobalRef(M, :z).binding
Binding Main.M.z
   39874:∞ - undefined binding - guard entry

julia> age2 = Int(Base.get_world_counter())
39881

julia> @eval M z=1
1

julia> b = GlobalRef(M, :z).binding
Binding Main.M.z
   39882:∞ - global variable with type Any
   39874:39881 - undefined binding - guard entry

julia> @eval M y=1
1

julia> b = GlobalRef(M, :y).binding
Binding Main.M.y
   39883:∞ - global variable with type Any
   39874:39882 - undefined binding - guard entry

Assuming this is necessary, I'm not sure how to identify the world age at time of creation.

[timholy]

xref #61200