remove unused roottask constant

[JeffBezanson]

This doesn't seem to be used anywhere and must be quite obsolete.

[JeffBezanson]

Removing an unused variable breaks everything. Classic 😂

It looks like the system image writer is depending on there being a reference to this; will fix it.

[vtjnash]

LOL, this would have made #31948 easier. I was wondering where that reference was popping up from. You could now just drop all references to root_task inside that file (unless you want to continue supporting packages creating references to it). I suppose you have to add a test now too to make sure it keeps working, since it's no longer implicitly tested by the sysimg build?

[mkitti]

In JavaCall.jl, we use the roottask constant to detect if the code is running on the root task. Is there a better way?

https://github.com/JuliaInterop/JavaCall.jl/blob/81029b7220803d743bf966a059f8981973cfad3c/src/jvm.jl#L230

[StefanKarpinski]

@vtjnash: is there an answer to @mkitti's question?

[mkitti]

As long as we are not removing Base.roottask, we should be able to get by at the moment.

I suggested in #35726 that the API should be a method, Base.root_task() that parallels Base.current_task(), which would allow some flexibility in the future. We could change the underlying implementation of the method if needed. An alternative is to implement Base.isroottask( task = Base.current_task() ).

There's some concern that the concept of a root Task is an implementation detail that should not be relied upon. However, with the current implementation there is a practical need to identify the root Task for interoperability with other software such as the Java Virtual Machine. This also arose in #36064 for embedding Julia into GAP. There the solution was just to record the current Task just after initializing Julia.

Should the concept of a root Task become obsolete, one could just make Base.root_task() an alias for Base.current_task() or have Base.isroottask() = true.

[Keno]

I just ran into this also. Having this magic object that needs special magic in the serializer is very unfortunate. We should get rid of it if we can.

[mkitti]

I just ran into this also. Having this magic object that needs special magic in the serializer is very unfortunate. We should get rid of it if we can.

We just need a way to detect if the current task is the root task to prevent JavaCall.jl from crashing Julia.

The following would be sufficient if jl_get_root_task always existed.

function is_root_task(task = current_task())
    task == ccall(:jl_get_root_task, Ref{Task}, ())
end

That depends on jl_get_root_task always existing:

julia/src/task.c

Lines 753 to 757 in 107901d

	JL_DLLEXPORT jl_value_t *jl_get_root_task(void)
	{
	jl_ptls_t ptls = jl_get_ptls_states();
	return (jl_value_t*)ptls->root_task;
	}

jl_get_root_task is currently only defined if JL_HAVE_ASYNCIFY.