Skip to content

v0.117.0

Choose a tag to compare

@github-actions github-actions released this 04 Nov 18:09
· 77 commits to main since this release
1265094

Oceananigans v0.117.0

Diff since v0.116.1

Breaking changes

  • New Open topology, distinct from Bounded: operators may read the physical halos beyond an Open direction as if they were part of the domain (#6276).
  • Open boundary conditions are reworked into part of time stepping. Each open boundary has its own storage and is advanced by initialize_boundary_conditions!, cache_boundary_conditions! and step_boundary_conditions!, no longer inside fill_halo_regions! (#6277).
  • The default JLD2Writer compression changes from Deflate to Shuffle + ZstdFilter(1). Both are lossless, but reading new files requires a JLD2 that supports Zstd (#5891).
  • The Cox-stencil IsopycnalSkewSymmetricDiffusivity is removed in favor of the triad implementation, so results change (#5955).
  • TripolarGrid cells beyond the southern wall are now inactive (#5564).
  • With Float32 data and Float64 times, FieldTimeSeries reads at exact time hits now return Float64, as interpolated reads already did (#6366).
  • The nodes of a regular coordinate no longer depend on its halo or partition (#6183).

Changes to solutions

  • The WENOVectorInvariant() vertical momentum flux is now consistent with the divergence cross term. This removes a spurious growing 2Δz mode (#6402).

New features

  • GridFittedVerticalBoundaries and PartialCellVerticalBoundaries (#6153).
  • at_maximum, at_minimum, and findmax(f, field) (#6323).
  • Calendar-year Cyclical, numeric schedules on DateTime clocks, and DateTime stop times (#6342).
  • Biogeochemistry checkpointing (#6382).
  • Distributed warns about misconfigured MPI/GPU setups (#6314).
  • Constant fields as JLD2Writer and NetCDFWriter outputs (#5177).

Merged pull requests:

Closed issues:

  • Reducing a Field with mean() is much slower than reducing its underlying array (#4633)
  • Add arch as an argument to heuristic_workgroup (#5207)
  • DelinearizeIndexingPass from fill_halo_regions (#5252)
  • Can't raise _compute_source_term! with Reactant (#5272)
  • QuasiAdamsBashforth2 Reactant extension correctness (#5301)
  • Bounded halo-filling bug? (#5319)
  • on_architecture(arch, ::GridFittedBottom{<:Field}) crashes on middle ranks of 1×N≥3 distributed TripolarGrid (#5627)
  • Reactant extension update_state! eager call (#5714)
  • Overwrite LinearAlgebra.norm for Reactant (#5834)
  • Most open boundary schemes aren't on the docs page (#6311)
  • Reactant run! traces the first time step twice for every timestepper; only Adams-Bashforth needs the peel (#6329)
  • update_field_time_series! on a window that holds every snapshot reads the time on the host, failing at a traced time (#6345)
  • TimeSeriesInterpolation's adapt_structure leaves the clock time unadapted, so a Field of it fails to compile on ReactantState (#6347)
  • First nested lazy reduction compiled in a julia process is about 1000x slower than subsequent runs (#6351)
  • FieldTimeSeries with Float32 data and Float64 times reads as Union{Float32, Float64} (#6364)
  • prognostic_state(model) skips model.biogeochemistry (#6380)
  • SplitRungeKutta time steppers never call update_lagrangian_particle_state!, so particle tracked_fields are never updated (#6381)
  • Picking up from a checkpoint fails when auxiliary_fields include an AbstractOperation or FunctionField (#6390)
  • WENOVectorInvariant has a spurious linear instability with a 2Δz structure: proof, mechanism, and a fix (#6400)
  • Extend methods explicitly: Module.f(...) = ... instead of import Module: f (#6405)
  • ShallowWaterModel allocates ~9 KiB per grid point per RK3 stage on the CPU (#6421)