Conversation
Benchmark Results (Julia v1.10)Time benchmarks
Memory benchmarks
|
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
Would it make sense to move the JET.jl tests to a new CI job - or multiple of them - to fix this? Alternatively, it seems to be fine to use JET.jl only with a specific version, e.g., the LTS or the newest release. |
Would that work? We need to have some compat bound for JET.jl in test/Project.toml. If it is something like v0.9.9 the environment for Julia v1.12 cannot instantiate and if it is something like v0.10, the environment for Julia <v1.12 cannot instantiate even if we do not run any JET.jl tests in either of the tests. If I understand correctly, we would also need different environments for the different jobs or am I missing something? |
|
@ranocha, do you have an idea how to fix the world age issue with |
Yes - we could have a specific setup that installs JET.jl only for one of the Julia versions. |
|
Does it help to use DispersiveShallowWater.jl/src/util.jl Line 91 in a8617f0 (and does it change any performance significantly)? |
Ok, so this would mean we don't have JET.jl in the test/Project.toml, but install it within the action yaml file (e.g., as a new action independent of the current CI.yaml)? Yeah, that would probably work, but is a more complicated setup than I would wish. But if it's the only option we see, I can try to set that up. |
Let's try in 14307b4. |
l2_error, linf_error = @invokelatest mod.analysis_callback(mod.sol)gives ERROR: LoadError: UndefVarError: `sol` not defined in `Main.var"##625"`
The binding may be too new: running in world age 44347, while current world is 44365.instead of ERROR: LoadError: UndefVarError: `analysis_callback` not defined in `Main.var"##624"`
The binding may be too new: running in world age 44340, while current world is 44358.Any idea how to solve this? |
|
Maybe another |
|
This fixed the world age issues 👍 This will probably also become relevant for Trixi.jl with Julia v1.12. |
|
Did the plain |
|
I tested it in Julia v1.11. So it worked. I didn't try from the REPL in Julia v1.12. But I can try. |
It fails under Julia v1.12 in the REPL. The |
Pull Request Test Coverage Report for Build 18346023930Details
💛 - Coveralls |
JoshuaLampert
left a comment
There was a problem hiding this comment.
CI passes on v1.12 and pre-v1.12 🥳
I would say, we can merge this once v1.12 is released, which shouldn't be in the too far future.
@ranocha, was the separate workflow for JET.jl roughly what you had in mind with
Would it make sense to move the JET.jl tests to a new CI job - or multiple of them - to fix this?
Problems:
│ WARNING: Constructor for type "Int16" was extended inVectorizationBasewithout explicit qualification or import.from VectorizationBase.jl, fixed by a new release of VectorizationBase.jl., but there are still warnings from LoopVectorization.jl and also SummationByPartsOperators.jl (probably also due to LoopVectorization.jl):set_approximation_variables!(q, mesh, equations, solver) = nothingto make JET.jl happy? Done in f16c15f.convergence_test: https://github.com/NumericalMathematics/DispersiveShallowWater.jl/actions/runs/15508851178/job/43667087583?pr=216#step:7:22099serre_green_naghdi_manufactured.jl with bathymetry_variable: https://github.com/NumericalMathematics/DispersiveShallowWater.jl/actions/runs/17231862744/job/48887502063?pr=216#step:7:8710,kdv_1d_nondim.jl: https://github.com/NumericalMathematics/DispersiveShallowWater.jl/actions/runs/17231862744/job/48887502063?pr=216#step:7:16570,hyperbolic_serre_green_naghdi_manufactured.jl: https://github.com/NumericalMathematics/DispersiveShallowWater.jl/actions/runs/17231862744/job/48887502063?pr=216#step:7:23401. Fixed by loosening the relevant tolerances.