Large allocations with MTK event

[klinders]

Describe the bug 🐞
Adding an event to the MTK model will slow down the solve significantly due to excessive allocations.

Expected behavior

Minimal performance impact when using events.

Minimal Reproducible Example 👇

using ModelingToolkit
using ModelingToolkit: t_nounits as t, D_nounits as D

function affect!(mod,obs,ctx,int)
    ModelingToolkit.terminate!(int)

    return (;)
end

function FOL_event(;name)
    @parameters begin
        τ = 3.0 # parameters
    end
    @variables begin
        x(t) = 0.0 # dependent variables
    end
    eq=[
        D(x) ~ (1 - x) / τ
    ]

    events = [x ~ 0.6]=>(affect!,(;))

    return System(eq, t; name=name, continuous_events=events)
end

function FOL(;name)
    @parameters begin
        τ = 3.0 # parameters
    end
    @variables begin
        x(t) = 0.0 # dependent variables
    end
    eq=[
        D(x) ~ (1 - x) / τ
    ]

    return System(eq, t; name=name)
end

using OrdinaryDiffEq
@mtkbuild fol = FOL()
@mtkbuild fol_event = FOL_event()

print("Starting problems\n")
print("No event:")
@time prob = ODEProblem(fol, [], (0.0, 10.0), dense=true,save_everystep=false)
print("With event:")
@time prob_event = ODEProblem(fol_event, [], (0.0, 10.0), dense=true, save_everystep=false)

print("Solving problems\n")
print("No event:")
@time sol = solve(prob,Tsit5())
print("With event:")
@time sol_event = solve(prob_event, Tsit5())

Simulation output ⚠️
I ran the code above in a REPL twice to get rid of the compilation time.

Starting problems
No event:  0.009388 seconds (39.02 k allocations: 2.201 MiB)
With event:  0.007077 seconds (36.27 k allocations: 1.683 MiB)
Solving problems
No event:  0.000044 seconds (88 allocations: 6.578 KiB)
With event:  0.119570 seconds (144.21 k allocations: 8.894 MiB, 99.95% compilation time: 100% of which was recompilation)

Environment (please complete the following information):

• Output of using Pkg; Pkg.status()

Status `~/Documents/Simulations/Julia/BPackSimulator/Project.toml`
  [e9eeebe6] BatteryPeck v0.1.0 `src/BatteryPeck`
  [634d3b9d] DrWatson v2.19.1
  [961ee093] ModelingToolkit v10.26.1
  [16a59e39] ModelingToolkitStandardLibrary v2.25.0
  [1dea7af3] OrdinaryDiffEq v6.103.0
  [91a5bcdd] Plots v1.41.1
  [295af30f] Revise v3.12.0

• Output of versioninfo()

Julia Version 1.10.10
Commit 95f30e51f41 (2025-06-27 09:51 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 12 × 13th Gen Intel(R) Core(TM) i7-1365U
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, goldmont)
Threads: 1 default, 0 interactive, 1 GC (on 12 virtual cores)

Additional context
The topic seems similar to Significant allocations with Callbacks (Tsit5) , but since those fixes were merged a long time ago, I decided to open a new issue.

[klinders]

I'm starting to think the issue is not with MTK. It seems like DifferentialEquations.jl has the same disease.

using ModelingToolkit
using ModelingToolkit: t_nounits as t, D_nounits as D
using OrdinaryDiffEq

function affect!(integrator)
    terminate!(integrator)

    # return (;)
end

function FOL(;name)
    @parameters begin
        τ = 3.0 # parameters
    end
    @variables begin
        x(t) = 0.0 # dependent variables
    end
    eq=[
        D(x) ~ (1 - x) / τ
    ]

    return System(eq, t; name=name)
end

@mtkbuild fol = FOL()

cb = DiscreteCallback((u,t,integrator)-> t==1, affect!)

print("Starting problems\n")
print("No event:")
@time prob = ODEProblem(fol, [], (0.0, 10.0))
print("With event:")
@time prob_event = ODEProblem(fol, [], (0.0, 10.0))

print("Solving problems\n")
print("No event:")
@time sol = solve(prob,Tsit5())
print("With event:")
@time sol_event = solve(prob, Tsit5(), callback=cb, tstops=[1.0])

Output:

Starting problems
No event:  0.008206 seconds (38.60 k allocations: 2.169 MiB)
With event:  0.008437 seconds (28.83 k allocations: 1.325 MiB)
Solving problems
No event:  0.000042 seconds (196 allocations: 14.922 KiB)
With event:  0.611402 seconds (1.65 M allocations: 111.943 MiB, 5.67% gc time, 99.98% compilation time)