compat: Support the TypeEq type-system refactor (JuliaLang/julia#61915)

[Keno]

The TypeEq refactor changes typeof(Type{T}) from DataType to the new Core.TypeEq. As a consequence, a Type{T} value is no longer an instance of DataType (isa(Type{T}, DataType) is now false), even though Type{T} <: DataType still holds and Base.isType(Type{T}) is true.

JLD2 dispatches its DataType serialization machinery on typeof(x), so on such a Julia a value whose type is a Type{T} (for example the field of a struct declared x::Type{Float64}) was routed through the generic compound path instead of the dedicated DataType path. This surfaced as a MethodError in commit and downstream fieldcount/convert errors when writing such values.

Introduce a typeof_serializable helper that maps Core.TypeEq-typed values back to DataType (guarded by isdefined(Core, :TypeEq) so the code still loads on released Julia), and relax the remaining value-level DataType annotations and isa(x, DataType) checks to also accept Type{T} values. These changes keep the on-disk format identical to older Julia versions.

This change was made with the assistance of generative AI (Claude Code).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.39%. Comparing base (5d7a409) to head (e282cb5).
⚠️ Report is 7 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #718      +/-   ##
==========================================
+ Coverage   87.38%   87.39%   +0.01%     
==========================================
  Files          37       37              
  Lines        4588     4593       +5     
==========================================
+ Hits         4009     4014       +5     
  Misses        579      579              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[JonasIsensee]

Thanks for providing the update!

Codecov complains that the isdefined(Core, :TypeEq) == true branch is not hit as part of the tests.
Is this change not yet merged into julia nightly ?

+ @static if isdefined(Core, :TypeEq)
+     @nospecializeinfer typeof_serializable(@nospecialize(x)) =
+         isa(x, Core.TypeEq) ? DataType : typeof(x)
+ else
+     @nospecializeinfer typeof_serializable(@nospecialize(x)) = typeof(x)
+ end

[Keno]

Codecov complains that the isdefined(Core, :TypeEq) == true branch is not hit as part of the tests.
Is this change not yet merged into julia nightly ?

Merged after the PR, but before your comment ;)

[Keno]

That said, the nightly build with it probably isn't done yet.

[JonasIsensee]

Let's wait until then, retrigger CI, and merge when tests pass with the relevant code change covered.

[nhz2]

I don't think a change is needed here because this is inside an if isconcretetype(T) block.

[nhz2]

Suggested change

	elseif (isa(T, DataType) || Base.isType(T)) && sizeof(T) ≤ MAX_INLINE_SIZE
	elseif isa(T, DataType) && sizeof(T) ≤ MAX_INLINE_SIZE

[nhz2]

The tests are still missing coverage for these changes.

Examples of things still broken:

Unions are not saved correctly.

julia> jldsave("union_new.jld2"; x = Union{Nothing, Type{Int}})

julia> load("union_new.jld2")
Dict{String, Any} with 1 entry:
  "x" => Nothing

Type{Int} values cannot be read in older julia versions.

In Julia nightly

julia> jldsave("datatypes_new.jld2"; x = Type{Int})

julia> load("datatypes_new.jld2")
Dict{String, Any} with 1 entry:
  "x" => Type{Int64}

Loading this file in Julia 1.12.6

julia> load("datatypes_new.jld2")
┌ Warning: custom serialization of Core.TypeEq encountered, but the type does not exist in the workspace; the data will be read unconverted
└ @ JLD2 ~/.julia/packages/JLD2/qcxKY/src/data/reconstructing_datatypes.jl:112
Dict{String, Any} with 1 entry:
  "x" => UnknownType{Symbol("Core.TypeEq"), Tuple{Int64}}

There is a 2x file size regression storing arrays of Type{Int}, not sure this is practical problem, but it does show there was some change to the representation.

1.12.6:

julia> jldsave("typearray.jld2"; x = fill(Type{Int}, 100000))

julia> filesize("typearray.jld2")
8801445

nightly:

julia> jldsave("typearray.jld2"; x = fill(Type{Int}, 100000))

julia> filesize("typearray.jld2")
16379568