Closed oscardssmith closed 1 month ago
Attention: Patch coverage is 0%
with 5 lines
in your changes are missing coverage. Please review.
Project coverage is 41.40%. Comparing base (
541e7fd
) to head (ccf79ed
).
Files | Patch % | Lines |
---|---|---|
src/integrator_interface.jl | 0.00% | 5 Missing :warning: |
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Previously, we would run the instability check no matter how badly the integrator had controlled the error. This caused issues, especially for the first timestep where the dt chosen was fairly arbitrary. We could end up in a situation very easily where the first timestep was big enough that the step would result in incredibly large (or NaN) u values. This step would get rejected, but because the values were incorrectly really big, we would get killed by instability detection. This is silly because just shrinking the
dt
would have made it stable, but we didn't get a chance to do so.With this PR, we will only do the instability check in the first place if we either roughly have the error controlled or the dt is already pretty small.