Surprising world age error in Julia 1.12rc1

[RJDennis]

The following MWE produces a world age problem that does not occur in Julia 1.11 or 1.10, etc.

struct store{T<:AbstractFloat}
    num::T
end

path = "c:/users/work/desktop/test_file.txt" # You will need to change this to align with where you put test_file.txt

function retrieve_file(path::Q) where {Q<:AbstractString}

    include(path)

    if length(n) != 0
        st = store(x)
    else
        st = store(x)
    end

    return st

end

st = retrieve_file(path) # Produces world age error
st = retrieve_file(path) # Running it again works

test_file.txt is attached.

test_file.txt

[Seelengrab]

I'm pretty sure that this is expected, since include is not really intended for use in a local scope.

[RJDennis]

It's not expected because it worked without error in previous Julia 1.x versions, isn't that how semver works? And it works when the last line is executed the second time, which wouldn't happen if it should error.

[KristofferC]

This is kind of breaking change yes, you will need to use invokelatest to access bindings introduced like that. https://pretalx.com/juliacon-2025/talk/TJXV9L/ was a talk about it at JuliaCon.

[timholy]

It's not expected because it worked without error in previous Julia 1.x versions, isn't that how semver works?

Semver is not as straightforward as that. Obligatory link https://xkcd.com/1172/, not to say that your example is in that category, but just to point out that the gap between "bug fix" or "new feature" vs "breaking change" is not nearly as clear as you might imagine it to be. There's plenty of code that breaks with new releases, but often it's clearly the fault of the broken code (e.g., it accessed internals of Julia that weren't public, etc.) This is a more complicated case.

There is a perspective from which it was a bug that your code ever worked at all: that really is kinda an abuse of include, at least in how I think about it. But it also makes sense that it worked given how we resolved bindings in previous Julia versions, so I also see your point.

[RJDennis]

I don't claim the MWE to represent exemplary code. But it doesn't play with Julia's internals and it only uses functions that are part of Julia's public API. Perhaps it should never have worked, but it did, so the fact that it errors in Julia 1.12 looks to me like a violation of semver.

Now, of course, I will look at the links provided above so thanks for those.

[vtjnash]

We could probably fix it, but the conclusion of #58511 was not to

[adienes]

see also: #39136