Merge pull request #211 from barche/v07-further-fixes

lcw · web-flow · commit 5f605b77e4e6 · 2018-08-02T12:14:18.000-07:00
V07 further fixes
diff --git a/.gitignore b/.gitignore
@@ -7,3 +7,5 @@ libjuliampi.so
 libjuliampi.dylib
 *.cov
 *.mem
+Manifest.toml
+deps/build.log
diff --git a/.travis.yml b/.travis.yml
@@ -5,7 +5,6 @@ os:
   - linux
   - osx
 julia:
-  - 0.6
   - 0.7
   - nightly
 notifications:
@@ -21,8 +20,6 @@ before_install:
 # Work around OpenMPI attempting to create overly long temporary
 # file names - and erroring as a result
   - export TMPDIR=/tmp
-script:
-    - julia -e 'if VERSION < v"0.7.0-DEV.5183"; Pkg.clone(pwd()) ; else; using Pkg; Pkg.up(); end; Pkg.build("MPI"); Pkg.test("MPI"; coverage=true)'
 after_success:
-  - julia -e 'VERSION < v"0.7.0-DEV.5183" || using Pkg; cd(Pkg.dir("MPI")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())'
-  - julia -e 'VERSION < v"0.7.0-DEV.5183" || using Pkg; cd(Pkg.dir("MPI")); Pkg.add("Coverage"); using Coverage; Codecov.submit(Codecov.process_folder())'
+  - julia -e 'using Pkg; cd(Pkg.dir("MPI")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())'
+  - julia -e 'using Pkg; cd(Pkg.dir("MPI")); Pkg.add("Coverage"); using Coverage; Codecov.submit(Codecov.process_folder())'
diff --git a/Project.toml b/Project.toml
@@ -1,10 +1,13 @@
-authors = []
 name = "MPI"
 uuid = "da04e1cc-30fd-572f-bb4f-1f8673147195"
-version = "0.1.0"
+authors = []
+version = "0.7.0"
 
 [deps]
 BinDeps = "9e28174c-4ba2-5203-b857-d8d62c4213ee"
 Compat = "34da2185-b29b-5c13-b0c7-acf172513d20"
 Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+Sockets = "6462fe0b-24de-5631-8697-dd941f90decc"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
diff --git a/README.md b/README.md
@@ -118,10 +118,10 @@ The julia master process is NOT part of the MPI cluster. The main script should
 launched directly, MPIManager internally calls `mpirun` to launch julia/mpi workers.
 All the workers started via MPIManager will be part of the MPI cluster.
 
-`MPIManager(;np=Sys.CPU_CORES, mpi_cmd=false, launch_timeout=60.0)`
+`MPIManager(;np=Sys.CPU_THREADS, mpi_cmd=false, launch_timeout=60.0)`
 
 If not specified, `mpi_cmd` defaults to `mpirun -np $np`
-STDOUT from the launched workers is redirected back to the julia session calling `addprocs` via a TCP connection.
+`stdout` from the launched workers is redirected back to the julia session calling `addprocs` via a TCP connection.
 Thus the workers must be able to freely connect via TCP to the host session.
 The following lines will be typically required on the julia master process to support both julia and mpi
 
diff --git a/REQUIRE b/REQUIRE
@@ -1,3 +1,3 @@
-julia 0.6
+julia 0.7.0-beta
 BinDeps
 Compat 0.66
diff --git a/appveyor.yml b/appveyor.yml
@@ -1,11 +1,9 @@
 environment:
   matrix:
-  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x86/0.6/julia-0.6-latest-win32.exe"
-  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x64/0.6/julia-0.6-latest-win64.exe"
   - JULIA_URL: "https://julialangnightlies-s3.julialang.org/bin/winnt/x86/julia-latest-win32.exe"
   - JULIA_URL: "https://julialangnightlies-s3.julialang.org/bin/winnt/x64/julia-latest-win64.exe"
-  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x86/0.7/julia-0.7.0-alpha-win32.exe"
-  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x64/0.7/julia-0.7.0-alpha-win64.exe"
+  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x86/0.7/julia-0.7-latest-win32.exe"
+  - JULIA_URL: "https://julialang-s3.julialang.org/bin/winnt/x64/0.7/julia-0.7-latest-win64.exe"
 
 branches:
   only:
@@ -33,11 +31,7 @@ install:
   - set PATH=C:\Program Files\Microsoft MPI\Bin;%PATH%
 
 build_script:
-# Need to convert from shallow to complete for Pkg.clone to work
-  - IF EXIST .git\shallow (git fetch --unshallow)
-  - C:\projects\julia\bin\julia -e "VERSION < v\"0.7.0-DEV\" || (using InteractiveUtils);
-    versioninfo(); pkg=\"MPI\";
-    if VERSION < v\"0.7.0-DEV.5183\"; Pkg.clone(pwd(), pkg); else; using Pkg; Pkg.up(); end; Pkg.build(\"MPI\")"
+  - C:\projects\julia\bin\julia -e "using Pkg; pkg\"activate .\"; pkg\"build\""
 
 test_script:
-  - C:\projects\julia\bin\julia --check-bounds=yes -e "VERSION < v\"0.7.0-DEV.5183\" || using Pkg; Pkg.test(\"MPI\")"
+  - C:\projects\julia\bin\julia --check-bounds=yes -e "using Pkg; pkg\"activate .\"; pkg\"test\""
diff --git a/deps/CMakeLists.txt b/deps/CMakeLists.txt
@@ -115,65 +115,14 @@ include_directories(${MPI_Fortran_INCLUDE_PATH})
 set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${MPI_C_COMPILE_FLAG}")
 set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ${MPI_Fortran_COMPILE_FLAG}")
 
-file(WRITE ${CMAKE_BINARY_DIR}/version.h.in
-"\#define VERSION \"@VERSION@\"\n"
-"\#define INSTALL \"@CMAKE_INSTALL_PREFIX@\"\n"
-)
-
 if(GIT_FOUND)
   set(USE_GIT 1)
 else(GIT_FOUND)
   set(USE_GIT 0)
 endif(GIT_FOUND)
-file(WRITE ${CMAKE_BINARY_DIR}/Version.cmake
-"if(${USE_GIT} EQUAL 1)
-  execute_process(
-       COMMAND ${GIT_EXECUTABLE} describe --tags --dirty
-       WORKING_DIRECTORY \${WORKING_DIRECTORY}
-       OUTPUT_VARIABLE GIT_DESCRIBE_VERSION
-       RESULT_VARIABLE GIT_DESCRIBE_RESULT
-       ERROR_VARIABLE GIT_DESCRIBE_ERROR
-       OUTPUT_STRIP_TRAILING_WHITESPACE
-   )
-else(${USE_GIT} EQUAL 1)
-  set(GIT_DESCRIBE_RESULT -1)
-endif(${USE_GIT} EQUAL 1)
-
-if(GIT_DESCRIBE_RESULT EQUAL 0)
-  file(WRITE ${CMAKE_SOURCE_DIR}/VERSION \"\${GIT_DESCRIBE_VERSION}\")
-else(GIT_DESCRIBE_RESULT EQUAL 0)
-  FILE(READ  ${CMAKE_SOURCE_DIR}/VERSION_NO_GIT GIT_DESCRIBE_VERSION)
-endif(GIT_DESCRIBE_RESULT EQUAL 0)
 
-string(REGEX REPLACE \"v([0-9]*)\\\\.([0-9]*)\\\\.(.*)\"
-       \"\\\\1.\\\\2.\\\\3\" VERSION \"\${GIT_DESCRIBE_VERSION}\" )
-
-string(REGEX REPLACE \"v([0-9]*)\\\\.([0-9]*)\\\\.(.*)\"
-       \"\\\\1\" VERSION_MAJOR \"\${GIT_DESCRIBE_VERSION}\")
-string(REGEX REPLACE \"v([0-9]*)\\\\.([0-9]*)\\\\.(.*)\"
-       \"\\\\2\" VERSION_MINOR \"\${GIT_DESCRIBE_VERSION}\")
-string(REGEX REPLACE \"v([0-9]*)\\\\.([0-9]*)\\\\.(.*)\"
-       \"\\\\3\" VERSION_PATCH \"\${GIT_DESCRIBE_VERSION}\")
-")
-
-file(WRITE ${CMAKE_BINARY_DIR}/FillVersion.cmake
-"include(Version.cmake)
-configure_file(\${SRC} \${DST} @ONLY)
-")
 include_directories(${CMAKE_BINARY_DIR})
-add_custom_target(
-    version
-    ${CMAKE_COMMAND} -D SRC=${CMAKE_BINARY_DIR}/version.h.in
-                     -D DST=${CMAKE_BINARY_DIR}/version.h
-                     -D WORKING_DIRECTORY=${CMAKE_SOURCE_DIR}
-                     -P ${CMAKE_BINARY_DIR}/FillVersion.cmake
-)
 set(WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}")
-include(${CMAKE_BINARY_DIR}/Version.cmake)
-set(CPACK_PACKAGE_VERSION "${VERSION}")
-set(CPACK_PACKAGE_VERSION_MAJOR "${VERSION_MAJOR}")
-set(CPACK_PACKAGE_VERSION_MINOR "${VERSION_MINOR}")
-set(CPACK_PACKAGE_VERSION_PATCH "${VERSION_PATCH}")
 set(CPACK_SOURCE_IGNORE_FILES
 "~$"
 ".sw[a-z]$"
@@ -200,8 +149,6 @@ EXECUTE_PROCESS(COMMAND cpack -G TGZ --config CPackSourceConfig.cmake
     )
 endif(UNIX)
 
-add_dependencies(dist version)
-
 add_library(juliampi SHARED juliampi.c test_mpi.f90)
 
 if(MPI_Fortran_LINK_FLAGS)
@@ -221,7 +168,6 @@ else()
 endif()
 
 add_executable(gen_functions gen_functions.c)
-add_dependencies(gen_functions version)
 
 add_executable(gen_constants gen_constants.f90)
 
diff --git a/deps/gen_functions.c b/deps/gen_functions.c
@@ -2,7 +2,6 @@
 #include <stddef.h>
 #include "jlmpi_f2c.h"
 #include "mpi.h"
-#include "version.h"
 
 // check that the size of MPI_Aint is consistent
 typedef int check_sizeof_MPI_Aint[sizeof(MPI_Aint) == sizeof(ptrdiff_t) ? 1 : -1];
@@ -14,7 +13,6 @@ int main(int argc, char *argv[]) {
   printf("using Compat\n");
   printf("import Compat.String\n");
   printf("\n");
-  printf("const WRAPPER_VERSION = \"%s\"\n", VERSION);
   printf("\n");
   printf("const _mpi_functions = Dict{Symbol, String}(\n");
   printf("    :MPI_ABORT              => \"%s\",\n", STRING(MPI_ABORT));
diff --git a/examples/05-juliacman.jl b/examples/05-juliacman.jl
@@ -11,10 +11,10 @@ println("Running 01-hello as part of a Julia cluster")
 @mpi_do manager (include("01-hello-impl.jl"); do_hello())
 
 # Interspersed julia parallel call
-nheads = @parallel (+) for i=1:10^8
+nheads = @distributed (+) for i=1:10^8
   Int(rand(Bool))
 end
-println("@parallel nheads $nheads")
+println("@distributed nheads $nheads")
 
 println("Running 02-broadcast as part of a Julia cluster")
 @mpi_do manager (include("02-broadcast-impl.jl"); do_broadcast())
diff --git a/src/MPI.jl b/src/MPI.jl
@@ -33,7 +33,7 @@ function __init__()
 
         # look up all symbols ahead of time
         for (jname, fname) in _mpi_functions
-            eval(:(const $jname = Libdl.dlsym(libmpi_handle, $fname)))
+            Core.eval(MPI, :(const $jname = Libdl.dlsym(libmpi_handle, $fname)))
         end
     end
 
diff --git a/src/cman.jl b/src/cman.jl
@@ -3,7 +3,7 @@ export MPIManager, launch, manage, kill, procs, connect, mpiprocs, @mpi_do
 export TransportMode, MPI_ON_WORKERS, TCP_TRANSPORT_ALL, MPI_TRANSPORT_ALL
 using Compat
 using Compat.Distributed
-import Compat.Sockets: connect, listenany, accept, getipaddr, IPv4
+import Compat.Sockets: connect, listenany, accept, IPv4, getsockname
 
 
 
@@ -42,6 +42,7 @@ mutable struct MPIManager <: ClusterManager
 
     # TCP Transport
     port::UInt16
+    ip::UInt32
     stdout_ios::Array
 
     # MPI transport
@@ -54,7 +55,7 @@ mutable struct MPIManager <: ClusterManager
     sending_done::Channel{Nothing}
     receiving_done::Channel{Nothing}
 
-    function MPIManager(; np::Integer = Sys.CPU_CORES,
+    function MPIManager(; np::Integer = Sys.CPU_THREADS,
                           mpirun_cmd::Cmd = `mpiexec -n $np`,
                           launch_timeout::Real = 60.0,
                           mode::TransportMode = MPI_ON_WORKERS)
@@ -86,13 +87,15 @@ mutable struct MPIManager <: ClusterManager
         if mode != MPI_TRANSPORT_ALL
             # Start a listener for capturing stdout from the workers
             port, server = listenany(11000)
+            ip = getsockname(server)[1].host
             @async begin
                 while true
                     sock = accept(server)
                     push!(mgr.stdout_ios, sock)
                 end
             end
             mgr.port = port
+            mgr.ip = ip
             mgr.stdout_ios = IO[]
         else
             mgr.rank2streams = Dict{Int,Tuple{IO,IO}}()
@@ -133,7 +136,7 @@ function Distributed.launch(mgr::MPIManager, params::Dict,
                 throw(ErrorException("Reuse of MPIManager is not allowed."))
             end
             cookie = string(":cookie_",Distributed.cluster_cookie())
-            setup_cmds = `using MPI\;MPI.setup_worker'('$(getipaddr().host),$(mgr.port),$cookie')'`
+            setup_cmds = `using MPI\;MPI.setup_worker'('$(mgr.ip),$(mgr.port),$cookie')'`
             mpi_cmd = `$(mgr.mpirun_cmd) $(params[:exename]) -e $(Base.shell_escape(setup_cmds))`
             open(detach(mpi_cmd))
             mgr.launched = true
@@ -151,7 +154,7 @@ function Distributed.launch(mgr::MPIManager, params::Dict,
             end
 
             # Traverse all worker I/O streams and receive their MPI rank
-            configs = Array{WorkerConfig}(mgr.np)
+            configs = Array{WorkerConfig}(undef, mgr.np)
             @sync begin
                 for io in mgr.stdout_ios
                     @async let io=io
@@ -199,12 +202,12 @@ function setup_worker(host, port, cookie)
 
     # Hand over control to Base
     if cookie == nothing
-        Base.start_worker(io)
+        Distributed.start_worker(io)
     else
         if isa(cookie, Symbol)
             cookie = string(cookie)[8:end] # strip the leading "cookie_"
         end
-        Base.start_worker(io, cookie)
+        Distributed.start_worker(io, cookie)
     end
 end
 
@@ -279,8 +282,8 @@ end
 # case
 function start_send_event_loop(mgr::MPIManager, rank::Int)
     try
-        r_s = BufferStream()
-        w_s = BufferStream()
+        r_s = Base.BufferStream()
+        w_s = Base.BufferStream()
         mgr.rank2streams[rank] = (r_s, w_s)
 
         # TODO: There is one task per communication partner -- this can be
@@ -292,7 +295,7 @@ function start_send_event_loop(mgr::MPIManager, rank::Int)
             reqs = MPI.Request[]
             while !isready(mgr.initiate_shutdown)
                 # When data are available, send them
-                while nb_available(w_s) > 0
+                while bytesavailable(w_s) > 0
                     data = take!(w_s.buffer)
                     push!(reqs, MPI.Isend(data, rank, 0, mgr.comm))
                 end
@@ -307,7 +310,7 @@ function start_send_event_loop(mgr::MPIManager, rank::Int)
         end
         (r_s, w_s)
     catch e
-        Base.show_backtrace(STDOUT, catch_backtrace())
+        Base.show_backtrace(stdout, catch_backtrace())
         println(e)
         rethrow(e)
     end
@@ -334,11 +337,15 @@ function start_main_loop(mode::TransportMode=TCP_TRANSPORT_ALL;
             # Create manager object
             mgr = MPIManager(np=size-1, mode=mode)
             mgr.comm = comm
+            # Needed because of Julia commit https://github.com/JuliaLang/julia/commit/299300a409c35153a1fa235a05c3929726716600
+            if isdefined(Distributed, :init_multi)
+                Distributed.init_multi()
+            end
             # Send connection information to all workers
             # TODO: Use Bcast
             for j in 1:size-1
                 cookie = VERSION >= v"0.5.0-dev+4047" ? Distributed.cluster_cookie() : nothing
-                MPI.send((getipaddr().host, mgr.port, cookie), j, 0, comm)
+                MPI.send((mgr.ip, mgr.port, cookie), j, 0, comm)
             end
             # Tell Base about the workers
             addprocs(mgr)
@@ -363,6 +370,9 @@ function start_main_loop(mode::TransportMode=TCP_TRANSPORT_ALL;
 
             # Send the cookie over. Introduced in v"0.5.0-dev+4047". Irrelevant under MPI
             # transport, but need it to satisfy the changed protocol.
+            if isdefined(Distributed, :init_multi)
+                Distributed.init_multi()
+            end
             MPI.bcast(Distributed.cluster_cookie(), 0, comm)
             # Start event loop for the workers
             @async receive_event_loop(mgr)
@@ -376,7 +386,7 @@ function start_main_loop(mode::TransportMode=TCP_TRANSPORT_ALL;
             mgr.comm = comm
             # Recv the cookie
             cookie = MPI.bcast(nothing, 0, comm)
-            Base.init_worker(cookie, mgr)
+            Distributed.init_worker(cookie, mgr)
             # Start a worker event loop
             receive_event_loop(mgr)
             MPI.Finalize()
@@ -394,7 +404,7 @@ function receive_event_loop(mgr::MPIManager)
         (hasdata, stat) = MPI.Iprobe(MPI.ANY_SOURCE, 0, mgr.comm)
         if hasdata
             count = Get_count(stat, UInt8)
-            buf = Array{UInt8}(count)
+            buf = Array{UInt8}(undef, count)
             from_rank = Get_source(stat)
             MPI.Recv!(buf, from_rank, 0, mgr.comm)
 
@@ -403,7 +413,7 @@ function receive_event_loop(mgr::MPIManager)
                 # This is the first time we communicate with this rank.
                 # Set up a new connection.
                 (r_s, w_s) = start_send_event_loop(mgr, from_rank)
-                Base.process_messages(r_s, w_s)
+                Distributed.process_messages(r_s, w_s)
                 num_send_loops += 1
             else
                 (r_s, w_s) = streams
@@ -459,7 +469,7 @@ end
 function mpi_do(mgr::MPIManager, expr)
     !mgr.initialized && wait(mgr.cond_initialized)
     jpids = keys(mgr.j2mpi)
-    refs = Array{Any}(length(jpids))
+    refs = Array{Any}(undef, length(jpids))
     for (i,p) in enumerate(Iterators.filter(x -> x != myid(), jpids))
         refs[i] = remotecall(expr, p)
     end
@@ -490,7 +500,7 @@ end
 macro mpi_do(mgr, expr)
     quote
         # Evaluate expression in Main module
-        thunk = () -> (eval(Main, $(Expr(:quote, expr))); nothing)
+        thunk = () -> (Core.eval(Main, $(Expr(:quote, expr))); nothing)
         mpi_do($(esc(mgr)), thunk)
     end
 end
diff --git a/test/runtests.jl b/test/runtests.jl
@@ -30,13 +30,8 @@ if Compat.Sys.iswindows()
     end
 end
 
-if VERSION > v"0.7.0-DEV.2005"
-  push!(excludedfiles, "test_cman_julia.jl")
-  push!(excludedfiles, "test_cman_mpi.jl")
-  push!(excludedfiles, "test_cman_tcp.jl")
-end
 function runtests()
-    nprocs = clamp(Sys.CPU_CORES, 2, 4)
+    nprocs = clamp(Sys.CPU_THREADS, 2, 4)
     exename = joinpath(BINDIR, Base.julia_exename())
     testdir = dirname(@__FILE__)
     istest(f) = endswith(f, ".jl") && startswith(f, "test_")
diff --git a/test/test_bcast.jl b/test/test_bcast.jl
diff --git a/test/test_cman_julia.jl b/test/test_cman_julia.jl
diff --git a/test/test_cman_mpi.jl b/test/test_cman_mpi.jl
diff --git a/test/test_cman_tcp.jl b/test/test_cman_tcp.jl
diff --git a/test/test_finalize_atexit.jl b/test/test_finalize_atexit.jl
diff --git a/test/test_spawn.jl b/test/test_spawn.jl
