Fix CI: gate MATLAB-runtime tests on engine availability; raise MATLAB floor to 0.8.1 (#83)

The Core and Downgrade jobs have been red because the MATLAB runtime is not
available on CI runners:

- Core (julia 1 / lts): matlab_runtime_tests.jl calls solve(prob, ode45()),
  which opens a MATLAB MSession. On any runner without a usable MATLAB engine
  (all GitHub-hosted runners, and self-hosted runners without a licensed
  engine) MSession throws UndefRefError, erroring the test. The interface and
  JET tests, which exercise this package's own logic without an engine, pass.
  Gate the engine-dependent solve calls on an actual capability probe
  (MATLAB.libmx_size > 0 plus a real MSession(0) attempt): run them where a
  working MATLAB engine exists (and assert they succeed), skip with an @info
  notice where it does not. The package logic tests still run unconditionally.

- Downgrade (lts): the MATLAB compat floor "0.8" resolves MATLAB v0.8.0, whose
  __init__ unconditionally runs `which matlab` and errors when MATLAB is absent,
  so MATLABDiffEq cannot even precompile. v0.8.1 introduced the deps.jl /
  check_deps build mechanism with a CI fallback (libmx_size = 0) that loads
  gracefully without MATLAB. Raise the floor to "0.8.1" (drops only the broken
  v0.8.0 patch) so Downgrade resolves to a loadable MATLAB.

Verified locally on Julia 1.12 (the "julia 1" channel) with CI=true and no
MATLAB present: Core tests pass (interface 40/40, jet 34/34, runtime tests
skipped with the @info notice). MATLAB v0.8.1 confirmed to load gracefully
(libmx_size == 0) on Julia 1.10 with CI=true and no MATLAB installed.

Co-authored-by: ChrisRackauckas-Claude <accounts@chrisrackauckas.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: 3 people

Project.toml

@@ -1,6 +1,6 @@
 name = "MATLABDiffEq"
 uuid = "e2752cbe-bcf4-5895-8727-84ebc14a76bd"
-version = "1.5.0"
+version = "1.5.1"
 
 [deps]
 DiffEqBase = "2b5f629d-d688-5b77-993f-72d75c75574e"
@@ -11,7 +11,7 @@ Reexport = "189a3867-3050-52da-a836-e630ba90ab69"
 
 [compat]
 DiffEqBase = "6.122, 7"
-MATLAB = "0.8, 0.9, 0.10"
+MATLAB = "0.8.1, 0.9, 0.10"
 ModelingToolkit = "8, 9, 10, 11"
 ParameterizedFunctions = "5"
 PrecompileTools = "1.1"

test/matlab_runtime_tests.jl

@@ -1,37 +1,64 @@
-# These tests require MATLAB runtime to be available.
-# They test the actual ODE solving functionality.
-using DiffEqBase, MATLABDiffEq, ParameterizedFunctions
+# These tests exercise the actual ODE solving functionality, which requires a
+# usable MATLAB engine. They are gated on the engine being startable: on CI
+# runners (and any machine without a licensed/working MATLAB) starting an
+# MSession is impossible, so the engine-dependent smoke tests are skipped with a
+# notice rather than failing. Where a working MATLAB engine is present the full
+# solve path runs and must succeed.
+using DiffEqBase, MATLABDiffEq, ParameterizedFunctions, MATLAB, Test
 
-f = @ode_def_bare LotkaVolterra begin
-    dx = a * x - b * x * y
-    dy = -c * y + d * x * y
-end a b c d
-p = [1.5, 1, 3, 1]
-tspan = (0.0, 10.0)
-u0 = [1.0, 1.0]
-prob = ODEProblem(f, u0, tspan, p)
-sol = solve(prob, MATLABDiffEq.ode45())
-
-function lorenz(du, u, p, t)
-    du[1] = 10.0(u[2] - u[1])
-    du[2] = u[1] * (28.0 - u[3]) - u[2]
-    return du[3] = u[1] * u[2] - (8 / 3) * u[3]
+# `libmx_size == 0` means MATLAB.jl was built without a MATLAB installation
+# (the build-time CI fallback), so the engine libraries were never loaded.
+# Even when libraries are present, starting an MSession can still fail (no
+# license, no display), so probe an actual session before running the solves.
+function matlab_engine_available()
+    MATLAB.libmx_size > 0 || return false
+    return try
+        MATLAB.MSession(0)
+        true
+    catch
+        false
+    end
 end
-u0 = [1.0; 0.0; 0.0]
-tspan = (0.0, 100.0)
-prob = ODEProblem(lorenz, u0, tspan)
-sol = solve(prob, MATLABDiffEq.ode45())
 
-algs = [
-    MATLABDiffEq.ode23
-    MATLABDiffEq.ode45
-    MATLABDiffEq.ode113
-    MATLABDiffEq.ode23s
-    MATLABDiffEq.ode23t
-    MATLABDiffEq.ode23tb
-    MATLABDiffEq.ode15s
-]
+@testset "MATLAB runtime solve" begin
+    if !matlab_engine_available()
+        @info "MATLAB engine not available on this machine; skipping engine-dependent solve tests."
+    else
+        f = @ode_def_bare LotkaVolterra begin
+            dx = a * x - b * x * y
+            dy = -c * y + d * x * y
+        end a b c d
+        p = [1.5, 1, 3, 1]
+        tspan = (0.0, 10.0)
+        u0 = [1.0, 1.0]
+        prob = ODEProblem(f, u0, tspan, p)
+        sol = solve(prob, MATLABDiffEq.ode45())
+        @test length(sol.t) > 0
+
+        function lorenz(du, u, p, t)
+            du[1] = 10.0(u[2] - u[1])
+            du[2] = u[1] * (28.0 - u[3]) - u[2]
+            return du[3] = u[1] * u[2] - (8 / 3) * u[3]
+        end
+        u0 = [1.0; 0.0; 0.0]
+        tspan = (0.0, 100.0)
+        prob = ODEProblem(lorenz, u0, tspan)
+        sol = solve(prob, MATLABDiffEq.ode45())
+        @test length(sol.t) > 0
+
+        algs = [
+            MATLABDiffEq.ode23
+            MATLABDiffEq.ode45
+            MATLABDiffEq.ode113
+            MATLABDiffEq.ode23s
+            MATLABDiffEq.ode23t
+            MATLABDiffEq.ode23tb
+            MATLABDiffEq.ode15s
+        ]
 
-for alg in algs
-    sol = solve(prob, alg())
+        for alg in algs
+            sol = solve(prob, alg())
+            @test length(sol.t) > 0
+        end
+    end
 end