Drop support for Julia 1.9 (trixi-framework#703)

* Drop support for Julia 1.9

* Explicitly test Julia versions since 1.10

* Increase compat for PointNeighbors.jl

* Downgrade CairoMakie

Why did this work before!?

* Run Windows and macOS tests on versions min and 1

* Fix job names

* Actually run jobs on the specified operating systems

* Display Julia version

* Synchronize with PointNeighbors.jl CI

* Bump version number

* Require hot-fixed version of PointNeighbors.jl

* Update solution_saving.jl

* Fix example tests

* Fix properly

* Fix validation

* Make validation tests pass on macOS ARM

* Update tolerances for macOS-ARM

Author: efaulhaber

.github/workflows/ci.yml

@@ -27,40 +27,35 @@ on:
       - 'docs/**'
   workflow_dispatch:
 
-# Cancel redundant CI tests automatically
+# Skip intermediate builds: always.
+# Cancel intermediate builds: only if it is a pull request build.
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
-  cancel-in-progress: true
+  cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
 
 jobs:
   build:
-    name: Run Tests (Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }})
-    runs-on: ubuntu-latest
+    name: Run Tests (Julia ${{ matrix.version }} - ${{ matrix.os }})
+    runs-on: ${{ matrix.os }}
     strategy:
+      # Don't cancel all running jobs when one job fails
       fail-fast: false
       matrix:
-          version:
-            - '1.9'
-            - '1.10'
-            - '1'
-          os:
-            - ubuntu-latest
-          arch:
-            - x64
-          include:
-            # Also run tests on Windows and macOS-ARM, but only with the latest Julia version and 1.10
-            - version: '1.10'
-              os: windows-latest
-              arch: x64
-            - version: '1.10'
-              os: macos-14
-              arch: arm64
-            - version: '1'
-              os: windows-latest
-              arch: x64
-            - version: '1'
-              os: macos-14
-              arch: arm64
+        version:
+          - '1.10'
+          - '1'
+        os:
+          - ubuntu-latest
+        include:
+          # Also run tests on Windows and macOS-ARM, but only with the latest Julia release and minimum supported version
+          - version: 'min'
+            os: windows-latest
+          - version: 'min'
+            os: macos-14
+          - version: '1'
+            os: windows-latest
+          - version: '1'
+            os: macos-14
 
     steps:
       - name: Check out project
@@ -71,6 +66,9 @@ jobs:
         with:
           version: ${{ matrix.version }}
 
+      - name: Display Julia version
+        run: julia -e 'using InteractiveUtils; versioninfo(verbose=true)'
+
       - uses: julia-actions/cache@v2
 
       - name: Build package

Project.toml

@@ -1,7 +1,7 @@
 name = "TrixiParticles"
 uuid = "66699cd8-9c01-4e9d-a059-b96c86d16b3a"
 authors = ["erik.faulhaber <[email protected]>"]
-version = "0.2.4"
+version = "0.2.5-dev"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
@@ -32,19 +32,19 @@ TrixiBase = "9a0f1c46-06d5-4909-a5a3-ce25d3fa3284"
 WriteVTK = "64499a7a-5c06-52f2-abe2-ccb03c286192"
 
 [compat]
-Adapt = "3, 4"
+Adapt = "4"
 CSV = "0.10"
 DataFrames = "1.6"
 DelimitedFiles = "1"
-DiffEqCallbacks = "2, 3, 4"
+DiffEqCallbacks = "4"
 FastPow = "0.1"
 FileIO = "1"
 ForwardDiff = "0.10"
-GPUArraysCore = "0.1, 0.2"
+GPUArraysCore = "0.2"
 JSON = "0.21"
 KernelAbstractions = "0.9"
 MuladdMacro = "0.2"
-PointNeighbors = "0.4.2"
+PointNeighbors = "0.4.7"
 Polyester = "0.7.10"
 RecipesBase = "1"
 Reexport = "1"
@@ -54,4 +54,4 @@ StrideArrays = "0.1"
 TimerOutputs = "0.5.25"
 TrixiBase = "0.1.3"
 WriteVTK = "1"
-julia = "1.9"
+julia = "1.10"

README.md

@@ -21,7 +21,7 @@ TrixiParticles.jl focuses on the following use cases:
 - Development of new particle-based methods and models.
 - Easy setup of accessible simulations for educational purposes, including student projects, coursework, and thesis work.
 
-It offers intuitive configuration, robust pre- and post-processing, and vendor-agnostic GPU-support based on the Julia package [KernelAbstractions.jl](https://github.com/JuliaGPU/KernelAbstractions.jl). 
+It offers intuitive configuration, robust pre- and post-processing, and vendor-agnostic GPU-support based on the Julia package [KernelAbstractions.jl](https://github.com/JuliaGPU/KernelAbstractions.jl).
 
 [![YouTube](https://github.com/user-attachments/assets/dc2be627-a799-4bfd-9226-2077f737c4b0)](https://www.youtube.com/watch?v=V7FWl4YumcA&t=4667s)
 
@@ -66,7 +66,7 @@ We provide several example simulation setups in the `examples` folder (which can
 ## Installation
 If you have not yet installed Julia, please [follow the instructions for your
 operating system](https://julialang.org/downloads/platform/). TrixiParticles.jl works
-with Julia v1.9 and newer. We recommend using the latest stable release of Julia.
+with Julia v1.10 and newer. We recommend using the latest stable release of Julia.
 
 ### For users
 TrixiParticles.jl is a registered Julia package.

docs/Project.toml

@@ -13,5 +13,5 @@ Documenter = "1"
 DocumenterCitations = "1"
 OrdinaryDiffEq = "6"
 Plots = "1"
-PointNeighbors = "0.4"
+PointNeighbors = "0.4.7"
 TrixiBase = "0.1"

docs/src/install.md

@@ -3,7 +3,7 @@
 ## Setting up Julia
 If you have not yet installed Julia, please [follow the instructions on the
 official website](https://julialang.org/downloads/). TrixiParticles.jl works
-with Julia v1.9 and newer. We recommend using the latest stable release of Julia.
+with Julia v1.10 and newer. We recommend using the latest stable release of Julia.
 
 ## For users
 TrixiParticles.jl is a registered Julia package.

src/callbacks/solution_saving.jl

@@ -106,9 +106,7 @@ function SolutionSavingCallback(; interval::Integer=0, dt=0.0,
                                                -1, Ref("UnknownVersion"))
 
     if length(save_times) > 0
-        # See the large comment below for an explanation why we use `finalize` here.
-        # When support for Julia 1.9 is dropped, the `finalize` argument can be removed.
-        return PresetTimeCallback(save_times, solution_callback, finalize=solution_callback)
+        return PresetTimeCallback(save_times, solution_callback)
     elseif dt > 0
         # Add a `tstop` every `dt`, and save the final solution
         return PeriodicCallback(solution_callback, dt,
@@ -198,41 +196,22 @@ end
 # When `interval` is used, this is
 # `DiscreteCallback{<:SolutionSavingCallback,
 #                   <:SolutionSavingCallback,
-#                   typeof(TrixiParticles.initialize_save_cb!),
-#                   typeof(SciMLBase.FINALIZE_DEFAULT)}`.
+#                   typeof(TrixiParticles.initialize_save_cb!)}`.
 #
 # When `dt` is used, this is
 # `DiscreteCallback{DiffEqCallbacks.var"#99#103"{...},
 #                   DiffEqCallbacks.PeriodicCallbackAffect{<:SolutionSavingCallback},
-#                   DiffEqCallbacks.var"#100#104"{...}
-#                   typeof(SciMLBase.FINALIZE_DEFAULT)}`.
+#                   DiffEqCallbacks.var"#100#104"{...}}`.
 #
 # When `save_times` is used, this is
 # `DiscreteCallback{DiffEqCallbacks.var"#115#117"{...},
 #                   <:SolutionSavingCallback,
-#                   DiffEqCallbacks.var"#116#118"{...},
-#                   typeof(SciMLBase.FINALIZE_DEFAULT)}}`.
+#                   DiffEqCallbacks.var"#116#118"{...}}`.
 #
 # So we can unambiguously dispatch on
 # - `DiscreteCallback{<:SolutionSavingCallback, <:SolutionSavingCallback}`,
 # - `DiscreteCallback{<:Any, <:PeriodicCallbackAffect{<:SolutionSavingCallback}}`,
 # - `DiscreteCallback{<:Any, <:SolutionSavingCallback}`.
-#
-# WORKAROUND FOR JULIA 1.9:
-# When `save_times` is used, the `affect!` is also wrapped in an anonymous function:
-# `DiscreteCallback{DiffEqCallbacks.var"#110#113"{...},
-#                   DiffEqCallbacks.var"#111#114"{<:SolutionSavingCallback},
-#                   DiffEqCallbacks.var"#116#118"{...},
-#                   typeof(SciMLBase.FINALIZE_DEFAULT)}}`.
-#
-# To dispatch here, we set `finalize` to the callback itself, so that the fourth parameter
-# becomes `<:SolutionSavingCallback`. This is only used in Julia 1.9. 1.10 and later uses
-# a newer version of DiffEqCallbacks.jl that does not have this issue.
-#
-# To use the callback as `finalize`, we have to define the following function.
-# `finalize` is set to `FINALIZE_DEFAULT` by default in the `PresetTimeCallback`,
-# which is a function that just returns `nothing`.
-# We define the `SolutionSavingCallback` to do the same when called with these arguments.
 function (finalize::SolutionSavingCallback)(c, u, t, integrator)
     return nothing
 end
@@ -256,17 +235,12 @@ function Base.show(io::IO,
     print(io, "SolutionSavingCallback(dt=", solution_saving.interval, ")")
 end
 
-# With `save_times`, also working in Julia 1.9.
-# When support for Julia 1.9 is dropped, this can be changed to
-# `DiscreteCallback{<:Any, <:SolutionSavingCallback}`, and the `finalize` argument
-# in the constructor of `SolutionSavingCallback` can be removed.
+# With `save_times`
 function Base.show(io::IO,
-                   cb::DiscreteCallback{<:Any, <:Any, <:Any, <:SolutionSavingCallback})
+                   cb::DiscreteCallback{<:Any, <:SolutionSavingCallback})
     @nospecialize cb # reduce precompilation time
 
-    # This has to be changed to `cb.affect!` when support for Julia 1.9 is dropped
-    # and finalize is removed from the constructor of `SolutionSavingCallback`.
-    solution_saving = cb.finalize
+    solution_saving = cb.affect!
     print(io, "SolutionSavingCallback(save_times=", solution_saving.save_times, ")")
 end
 
@@ -323,15 +297,13 @@ end
 
 # With `save_times`. See comments above.
 function Base.show(io::IO, ::MIME"text/plain",
-                   cb::DiscreteCallback{<:Any, <:Any, <:Any, <:SolutionSavingCallback})
+                   cb::DiscreteCallback{<:Any, <:SolutionSavingCallback})
     @nospecialize cb # reduce precompilation time
 
     if get(io, :compact, false)
         show(io, cb)
     else
-        # This has to be changed to `cb.affect!` when support for Julia 1.9 is dropped
-        # and finalize is removed from the constructor of `SolutionSavingCallback`.
-        solution_saving = cb.finalize
+        solution_saving = cb.affect!
         cq = collect(solution_saving.custom_quantities)
 
         setup = [

test/Project.toml

@@ -16,7 +16,7 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [compat]
 CSV = "0.10"
-CairoMakie = "0.13"
+CairoMakie = "0.12"
 DataFrames = "1.6"
 GLM = "1.9"
 Glob = "1.3"

test/count_allocations.jl

@@ -5,6 +5,8 @@ struct NoUpdateNeighborhoodSearch{NHS}
     nhs::NHS
 end
 
+@inline Base.ndims(nhs::NoUpdateNeighborhoodSearch) = ndims(nhs.nhs)
+
 # Copy a `Semidiscretization`, but wrap the neighborhood searches with
 # `NoUpdateNeighborhoodSearch`.
 function copy_semi_with_no_update_nhs(semi)

test/validation/validation.jl

@@ -47,18 +47,26 @@
         @test sol.retcode == ReturnCode.Success
         @test count_rhs_allocations(sol, semi) == 0
 
-        if VERSION == v"1.10"
+        if Sys.ARCH === :aarch64
+            # MacOS ARM produces slightly different pressure values than x86.
+            # Note that pressure values are in the order of 1e5.
+            @test isapprox(error_edac_P1, 0, atol=8e-10)
+            @test isapprox(error_edac_P2, 0, atol=7e-12)
+            @test isapprox(error_wcsph_P1, 0, atol=18.0)
+            @test isapprox(error_wcsph_P2, 0, atol=0.004)
+        elseif VERSION == v"1.10"
+            # Reference values are computed with 1.10
             @test isapprox(error_edac_P1, 0, atol=eps())
             @test isapprox(error_edac_P2, 0, atol=eps())
             @test isapprox(error_wcsph_P1, 0, atol=eps())
             @test isapprox(error_wcsph_P2, 0, atol=eps())
         else
-            # 1.9 causes a large difference in the solution
-            # TODO 1.11 requires a performance hotfix which will likely change these results again
-            @test isapprox(error_edac_P1, 0, atol=4e-9)
-            @test isapprox(error_edac_P2, 0, atol=3e-11)
-            @test isapprox(error_wcsph_P1, 0, atol=26.3)
-            @test isapprox(error_wcsph_P2, 0, atol=8.2e-3)
+            # 1.11 produces slightly different pressure values than 1.10.
+            # Note that pressure values are in the order of 1e5.
+            @test isapprox(error_edac_P1, 0, atol=eps())
+            @test isapprox(error_edac_P2, 0, atol=eps())
+            @test isapprox(error_wcsph_P1, 0, atol=0.07)
+            @test isapprox(error_wcsph_P2, 0, atol=8e-6)
         end
 
         # Ignore method redefinitions from duplicate `include("../validation_util.jl")`