Unrelated module wrapped in closure with --inline=no only

[andreasnoack]

The following code snippet fails when julia is launched --inline=no but works fine in normal mode. This causes the DistributedArrays.jl tests to fail.

import FactCheck

addprocs(1)

using DistributedArrays

g = () -> begin
    f = x -> Int(x)
    A = [1,2]
    DA = distribute(A)
    mapreduce(f, +, A)
    mapreduce(f, +, DA)
end

g()

Notice that FactCheck is not used for anything in g but the worker fails with

WARNING: Module FactCheck not defined on process 2
fatal error on 2: ERROR: UndefVarError: FactCheck not defined
 in deserialize(::SerializationState{TCPSocket}, ::Type{Module}) at ./serialize.jl:542
 [inlined code] from ./promotion.jl:135
 in handle_deserialize(::SerializationState{TCPSocket}, ::Int32) at ./serialize.jl:515
 [inlined code] from ./sysimg.jl:48
 in deserialize_datatype(::SerializationState{TCPSocket}) at ./serialize.jl:740
 in handle_deserialize(::SerializationState{TCPSocket}, ::Int32) at ./serialize.jl:505
 [inlined code] from ./sysimg.jl:48
 in deserialize(::SerializationState{TCPSocket}, ::Type{SimpleVector}) at ./serialize.jl:522
 [inlined code] from ./promotion.jl:135
 in handle_deserialize(::SerializationState{TCPSocket}, ::Int32) at ./serialize.jl:515
 [inlined code] from ./sysimg.jl:48
 in deserialize_datatype(::SerializationState{TCPSocket}) at ./serialize.jl:747
 in handle_deserialize(::SerializationState{TCPSocket}, ::Int32) at ./serialize.jl:505
 in deserialize_array(::SerializationState{TCPSocket}) at ./serialize.jl:640
 in handle_deserialize(::SerializationState{TCPSocket}, ::Int32) at ./serialize.jl:503
 [inlined code] from ./sysimg.jl:48
...

I've tried to reduce it but this is as small as I could make it. E.g. if either Int is used directly or removed from the f then the error goes away. If one of the mapreduce lines is commented out or if the order is reversed the error also goes away.

I've tried with a few other packages instead of FactCheck and they don't fail. However, since this has different behavior with and without inlining, I think the bug report is appropriate here.

My system is

Julia Version 0.5.0-dev+3390
Commit a9e7e86* (2016-04-04 12:47 UTC)
DEBUG build
Platform Info:
  System: Darwin (x86_64-apple-darwin15.4.0)
  CPU: Intel(R) Core(TM) i7-4870HQ CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT NO_AFFINITY HASWELL)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.7.1 (ORCJIT, haswell)

[andreasnoack]

I've bisected this to f6e9f8d.

[andreasnoack]

This popped up again in https://travis-ci.org/JuliaParallel/DistributedArrays.jl/builds/125829660 and it can be reproduced consistently with --inline=no.

[yuyichao]

Ah, this is the issue. Close #15996 as a dup of this. I think this is caused by serialization of the lambdainfo/method.

[amitmurthy]

Same as #15451 ?

[vtjnash]

fixed on master, by #16692?

julia> import FactCheck

julia> addprocs(1)
1-element Array{Int64,1}:
 2

julia> using DistributedArrays

julia> using Primes

julia> g = () -> begin
           f = x -> Int(x)
           A = [1,2]
           DA = distribute(A)
           mapreduce(f, +, A)
           mapreduce(f, +, DA)
       end
(::#1) (generic function with 1 method)

julia> factor(1,)
Dict{Int64,Int64} with 0 entries

julia> g()
WARNING: specialized_binary(f::Function) is deprecated, use f instead.
 in depwarn(::String, ::Symbol) at ./deprecated.jl:64
 in specialized_binary(::Function) at deprecated.jl:50
 in mapreduce(::Function, ::Function, ::DistributedArrays.DArray{Int64,1,Array{Int64,1}}) at /Users/jameson/.julia/v0.5/DistributedArrays/src/DistributedArrays.jl:679
 in (::##1#3)() at REPL[5]:6
 in eval(::Module, ::Any) at ./boot.jl:234
 in eval(::Module, ::Expr) at REPL.jl:3
 in macro expansion; at REPL.jl:92 [inlined]
 in (::Base.REPL.##1#2{Base.REPL.REPLBackend})() at event.jl:46
while loading no file, in expression starting on line 0
3

[andreasnoack]

I think so