fixup! Add a watcher mechanism to detect when Distributed might be in use

Author: JamesWrigley

src/DistributedNext.jl

@@ -72,17 +72,22 @@ export
 # Used only by shared arrays.
     check_same_host
 
-distributed_module::Union{Module, Nothing} = nothing
-
-function _find_distributed_stdlib()
+function _check_distributed_active()
     # Find the Distributed module if it's been loaded
     distributed_pkgid = Base.PkgId(Base.UUID("8ba89e20-285c-5b6f-9357-94700520ee1b"), "Distributed")
-    global distributed_module = get(Base.loaded_modules, distributed_pkgid, nothing)
+    if !haskey(Base.loaded_modules, distributed_pkgid)
+        return false
+    end
+
+    if isdefined(Base.loaded_modules[distributed_pkgid].LPROC, :cookie) && inited[]
+        @warn "DistributedNext has detected that the Distributed stdlib may be in use. Be aware that these libraries are not compatible, you should use either one or the other."
+        return true
+    else
+        return false
+    end
 end
 
 function _require_callback(mod::Base.PkgId)
-    _find_distributed_stdlib()
-
     if Base.toplevel_load[] && myid() == 1 && nprocs() > 1
         # broadcast top-level (e.g. from Main) import/using from node 1 (only)
         @sync for p in procs()
@@ -132,16 +137,12 @@ function __init__()
     # cluster cookie has been set, which is most likely to have been done
     # through Distributed.init_multi() being called by Distributed.addprocs() or
     # something.
-    _find_distributed_stdlib()
     watcher_task = Threads.@spawn while true
-        if !isnothing(distributed_module)
-            if isdefined(distributed_module.LPROC, :cookie) && inited[]
-                @warn "DistributedNext has detected that the Distributed stdlib may be in use. Be aware that these libraries are not compatible, you should use either one or the other."
-                return
-            end
+        if _check_distributed_active()
+            return
         end
 
-        sleep(0.2)
+        sleep(1)
     end
     errormonitor(watcher_task)
 end

test/distributed_stdlib_detection.jl

@@ -2,21 +2,21 @@ using DistributedNext
 
 @testset "Distributed.jl detection" begin
     # Just loading Distributed should do nothing
-    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; sleep(2)'`
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; @assert !DistributedNext._check_distributed_active()'`
     stderr_buf = IOBuffer()
     run(pipeline(cmd; stderr=stderr_buf))
     stderr_str = String(take!(stderr_buf))
     @test isempty(stderr_str)
 
     # Only one of the two being active should also do nothing
-    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); sleep(2)'`
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); @assert !DistributedNext._check_distributed_active()'`
     stderr_buf = IOBuffer()
     run(pipeline(cmd; stderr=stderr_buf))
     stderr_str = String(take!(stderr_buf))
     @test isempty(stderr_str)
 
     # But both being active at the same time should trigger a warning
-    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); DistributedNext.init_multi(); sleep(2)'`
+    cmd = `$test_exename $test_exeflags -e 'using Distributed, DistributedNext; Distributed.init_multi(); DistributedNext.init_multi(); @assert DistributedNext._check_distributed_active()'`
     stderr_buf = IOBuffer()
     run(pipeline(cmd; stderr=stderr_buf))
     stderr_str = String(take!(stderr_buf))