Replace Travis CI with GitHub Actions (#85)

* Replace Travis CI with GitHub Actions

Remove .travis.yml and add .github/workflows/ci.yml. Tests Julia 1
on Linux (x64) and macOS (Apple Silicon), and Julia pre on Linux.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Keep Grid reference in DistSparseMatrix and DistMultiVec

Prevent the default Grid from being garbage collected while
distributed matrices or vectors are still alive, matching the
pattern already used by DistMatrix.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Constrain MPICH_jll to v3/v4 to avoid ABI break in v5

MPICH 5.0 adopted the standard MPI ABI where MPI_Comm changed from
int to a pointer type. The pre-compiled Elemental binary is
incompatible with this new ABI, causing segfaults in Grid::VCSize().

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Constrain MPICH_jll to v3 to match libmpi.so.12 ABI

The Elemental binary was linked against libmpi.so.12 (MPICH 3.x).
MPICH 4.x may have internal ABI differences despite the same soname.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add Julia 1.10 to CI matrix to test older runtime

The Elemental binary segfaults on Julia 1.12 in Grid::VCSize()
regardless of MPICH version. Adding 1.10 to determine if this is
a Julia version compatibility issue.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Fix ccall signatures: pass Grid instead of MPI communicator

The C interface functions ElDistSparseMatrixCreate and
ElDistMultiVecCreate take an ElConstGrid (Grid pointer), not an
MPI_Comm. The Julia wrappers were incorrectly passing an ElComm
integer which got reinterpret_cast'd as a Grid pointer, causing
a null pointer dereference in Grid::VCSize().

Also revert the MPICH_jll version constraint and Julia 1.10 CI
entry since those were not the root cause.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Remove comm() for DistSparseMatrix/DistMultiVec, use grid instead

ElDistSparseMatrixComm and ElDistMultiVecComm don't exist in the
C API. Replace comm(A) calls with A.grid and remove the dead comm
functions.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Increase timeout for CI jobs and handle test process timeouts in runtests

* temp fix

* fix

* implemented suggestion

* Remove DistributedArrays import from runtests.jl

* Reduce matrix dimensions in lav.jl test to 12x12 for improved performance

* fix

---------

Co-authored-by: Andreas Noack <andreas@noack.dk>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: 3 people

.github/workflows/ci.yml

@@ -0,0 +1,39 @@
+name: CI
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+jobs:
+  test:
+    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }}
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 60
+    strategy:
+      fail-fast: false
+      matrix:
+        version:
+          - '1'
+        os:
+          - ubuntu-latest
+          - macos-latest
+        arch:
+          - x64
+        exclude:
+          - os: macos-latest
+            arch: x64
+        include:
+          - os: macos-latest
+            arch: aarch64
+            version: '1'
+          - os: ubuntu-latest
+            arch: x64
+            version: 'pre'
+    steps:
+      - uses: actions/checkout@v4
+      - uses: julia-actions/setup-julia@v2
+        with:
+          version: ${{ matrix.version }}
+      - uses: julia-actions/cache@v2
+      - uses: julia-actions/julia-buildpkg@v1
+      - uses: julia-actions/julia-runtest@v1

.travis.yml

@@ -15,11 +15,12 @@ Elemental_jll = "0.87"
 julia = "1.3"
 
 [extras]
+MPI = "da04e1cc-30fd-572f-bb4f-1f8673147195"
 MPIClusterManagers = "e7922434-ae4b-11e9-05c5-9780451d2c66"
 Primes = "27ebfcd6-29c5-5fa9-bf4b-fb8fc14df3ae"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 TSVD = "9449cd9e-2762-5aa3-a617-5413e99d722e"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Test", "MPIClusterManagers", "Primes", "TSVD", "Random"]
+test = ["Test", "MPI", "MPIClusterManagers", "Primes", "TSVD", "Random"]

Project.toml

@@ -1,5 +1,6 @@
 mutable struct DistMultiVec{T} <: ElementalMatrix{T}
     obj::Ptr{Cvoid}
+    grid::Grid # keep the grid around to avoid that it's freed before the matrix
 end
 
 for (elty, ext) in ((:ElInt, :i),
@@ -16,24 +17,16 @@ for (elty, ext) in ((:ElInt, :i),
             return nothing
         end
 
-        function DistMultiVec(::Type{$elty}, cm::ElComm = MPI.CommWorld[])
+        function DistMultiVec(::Type{$elty}, grid::Grid = DefaultGrid[])
             obj = Ref{Ptr{Cvoid}}(C_NULL)
             ElError(ccall(($(string("ElDistMultiVecCreate_", ext)), libEl), Cuint,
-                (Ref{Ptr{Cvoid}}, ElComm),
-                obj, cm))
-            A = DistMultiVec{$elty}(obj[])
+                (Ref{Ptr{Cvoid}}, Ptr{Cvoid}),
+                obj, grid.obj))
+            A = DistMultiVec{$elty}(obj[], grid)
             finalizer(destroy, A)
             return A
         end
 
-        function comm(A::DistMultiVec{$elty})
-            cm = Ref{ElComm}()
-            ElError(ccall(($(string("ElDistMultiVecComm_", ext)), libEl), Cuint,
-                (Ptr{Cvoid}, Ref{ElComm}),
-                A.obj, cm))
-            return cm[]
-        end
-
         function get(x::DistMultiVec{$elty}, i::Integer = size(x, 1), j::Integer = 1)
             v = Ref{$elty}()
             ElError(ccall(($(string("ElDistMultiVecGet_", ext)), libEl), Cuint,
@@ -117,8 +110,8 @@ end
 
 getindex(x::DistMultiVec, i, j) = get(x, i, j)
 
-function similar(::DistMultiVec, ::Type{T}, sz::Dims, cm::ElComm = MPI.CommWorld[]) where {T}
-    A = DistMultiVec(T, cm)
+function similar(::DistMultiVec, ::Type{T}, sz::Dims, grid::Grid = DefaultGrid[]) where {T}
+    A = DistMultiVec(T, grid)
     resize!(A, sz...)
     return A
 end

src/core/distmultivec.jl

@@ -1,5 +1,6 @@
 mutable struct DistSparseMatrix{T} <: ElementalMatrix{T}
     obj::Ptr{Cvoid}
+    grid::Grid # keep the grid around to avoid that it's freed before the matrix
 end
 
 for (elty, ext) in ((:ElInt, :i),
@@ -16,24 +17,16 @@ for (elty, ext) in ((:ElInt, :i),
             return nothing
         end
 
-        function DistSparseMatrix(::Type{$elty}, comm::ElComm = MPI.CommWorld[])
+        function DistSparseMatrix(::Type{$elty}, grid::Grid = DefaultGrid[])
             obj = Ref{Ptr{Cvoid}}(C_NULL)
             ElError(ccall(($(string("ElDistSparseMatrixCreate_", ext)), libEl), Cuint,
-                (Ref{Ptr{Cvoid}}, ElComm),
-                obj, comm))
-            A = DistSparseMatrix{$elty}(obj[])
+                (Ref{Ptr{Cvoid}}, Ptr{Cvoid}),
+                obj, grid.obj))
+            A = DistSparseMatrix{$elty}(obj[], grid)
             finalizer(destroy, A)
             return A
         end
 
-        function comm(A::DistSparseMatrix{$elty})
-            cm = Ref{ElComm}()
-            ElError(ccall(($(string("ElDistSparseMatrixComm_", ext)), libEl), Cuint,
-                (Ptr{Cvoid}, Ref{ElComm}),
-                A.obj, cm))
-            return cm[]
-        end
-
         function globalRow(A::DistSparseMatrix{$elty}, iLoc::Integer)
             i = Ref{ElInt}(0)
             ElError(ccall(($(string("ElDistSparseMatrixGlobalRow_", ext)), libEl), Cuint,
@@ -112,8 +105,8 @@ for (elty, ext) in ((:ElInt, :i),
 end
 
 # The other constructors don't have a version with dimensions. Should they, or should this one go?
-function DistSparseMatrix(::Type{T}, m::Integer, n::Integer, comm::ElComm = MPI.CommWorld[]) where {T}
-    A = DistSparseMatrix(T, comm)
+function DistSparseMatrix(::Type{T}, m::Integer, n::Integer, grid::Grid = DefaultGrid[]) where {T}
+    A = DistSparseMatrix(T, grid)
     resize!(A, m, n)
     return A
 end

src/core/distsparsematrix.jl

@@ -23,7 +23,7 @@ function leastSquares(A::DistMatrix{T}, B::DistMatrix{T};
 end
 function leastSquares(A::DistSparseMatrix{T}, B::DistMultiVec{T};
     orientation::Orientation = NORMAL) where {T}
-    X = DistMultiVec(T, comm(A))
+    X = DistMultiVec(T, A.grid)
     return leastSquares!(A, B, X, orientation = orientation)
 end
 
@@ -49,7 +49,7 @@ function bpdn(A::DistMatrix{T}, B::DistMatrix{T}, lambda::T) where {T}
     return bpdn!(A, B, lambda, X)
 end
 function bpdn(A::DistSparseMatrix{T}, B::DistMultiVec{T}, lambda) where {T}
-    X = DistMultiVec(T, comm(A))
+    X = DistMultiVec(T, A.grid)
     return bpdn!(A, B, lambda, X)
 end
 
@@ -78,6 +78,6 @@ function ridge(A::DistMatrix{T}, B::DistMatrix{T}, gamma::T; ka...) where {T}
     return ridge!(A, B, gamma, X; ka...)
 end
 function ridge(A::DistSparseMatrix{T}, B::DistMultiVec{T}, gamma::T; ka...) where {T}
-    X = DistMultiVec(T, comm(A))
+    X = DistMultiVec(T, A.grid)
     return ridge!(A, B, gamma, X; ka...)
 end

src/lapack_like/euclidean_min.jl

@@ -33,11 +33,11 @@ function lav(A::DistMatrix{T}, b::DistMatrix{T}) where {T<:Union{Float32,Float64
     return lav!(A, b, x)
 end
 function lav(A::DistSparseMatrix{T}, b::DistMultiVec{T}) where {T<:Union{Float32,Float64}}
-    x = DistMultiVec(T, comm(A))
+    x = DistMultiVec(T, A.grid)
     return lav!(A, b, x)
 end
 
 function lav(A::DistSparseMatrix{T}, b::DistMultiVec{T}, ctrl::LPAffineCtrl{T}) where {T<:Union{Float32,Float64}}
-    x = DistMultiVec(T, comm(A))
+    x = DistMultiVec(T, A.grid)
     return lav!(A, b, x, ctrl)
 end

src/optimization/models.jl

@@ -94,17 +94,7 @@ if display
     # El.print(b, "b")
 end
 
-ctrl = El.LPAffineCtrl(Float64,
-    mehrotraCtrl=El.MehrotraCtrl(Float64,
-        solveCtrl=El.RegSolveCtrl(Float64,
-                                  progress=true),
-        print=true,
-        outerEquil=true,
-        time=true)
-    )
-
-# elapsedLAV = @elapsed x = El.lav(A, b);#Elemental.print(A, "Matrix A")
-elapsedLAV = @elapsed x = El.lav(A, b, ctrl)
+elapsedLAV = @elapsed x = El.lav(A, b)
 
 if El.MPI.commRank(El.MPI.CommWorld[]) == 0
     println("LAV time: $elapsedLAV seconds")
@@ -165,4 +155,4 @@ end
 
 # Require the user to press a button before the figures are closed
 # commSize = El.mpi.Size( El.mpi.COMM_WORLD() )
-# El.Finalize()
+El.Finalize()

test/lav.jl

@@ -1,8 +1,7 @@
 using Test
-using Elemental_jll.MPICH_jll: mpiexec
+using MPI: mpiexec
 
 # Import all of our external dependencies to make sure they're compiled serially.
-using DistributedArrays
 using TSVD
 using Primes
 using MPIClusterManagers
@@ -15,9 +14,12 @@ function runtests_mpirun()
     @info "Running Elemental.jl tests"
     for f in testfiles
         try
-            mpiexec() do exec
-                run(`$exec -np $nprocs $(Base.julia_cmd()) $(joinpath(@__DIR__, f))`)
+            proc = run(`$(mpiexec()) -np $nprocs $(Base.julia_cmd()) $(joinpath(@__DIR__, f))`, wait=false)
+            if timedwait(() -> !process_running(proc), 300.0) === :timed_out
+                kill(proc)
+                error("Test $f timed out after 5 minutes")
             end
+            wait(proc)
             Base.with_output_color(:green,stdout) do io
                 println(io,"\tSUCCESS: $f")
             end
@@ -43,7 +45,12 @@ function runtests_repl()
             # FixMe! We temporarily run Finalize() explictly on the workers because the atexit hook
             # doesn't seem to be correctly triggered on workers as of 31 October 2018.
             cmdstr = "using Distributed, MPIClusterManagers; man = MPIManager(np = $nprocs); addprocs(man); include(\"$(joinpath(@__DIR__, f))\"); asyncmap(p -> remotecall_fetch(() -> Elemental.Finalize(), p), workers())"
-            run(`$exename -e $cmdstr`)
+            proc = run(`$exename -e $cmdstr`, wait=false)
+            if timedwait(() -> !process_running(proc), 300.0) === :timed_out
+                kill(proc)
+                error("Test $f timed out after 5 minutes")
+            end
+            wait(proc)
             Base.with_output_color(:green,stdout) do io
                 println(io,"\tSUCCESS: $f")
             end