(0.106.4) Move NetCDF output attribute methods from extension to Models module

[simone-silvestri]

Summary

• Moves default_output_attributes, default_velocity_attributes, and default_tracer_attributes method implementations from OceananigansNCDatasetsExt into src/Models/output_attributes.jl
• Keeps only a generic fallback default_output_attributes(model) = Dict() in OutputWriters
• Models defines specialized methods via dispatch on NonhydrostaticModel, HydrostaticFreeSurfaceModel, and ShallowWaterModel
• add_schedule_metadata! methods remain in OutputWriters (no external type dependencies)

Motivation

These functions don't depend on NCDatasets but were defined entirely inside the extension. This prevented downstream package extensions (e.g. ClimaSeaIceNCDatasetsExt) from adding methods at compile time, forcing them to use __init__ + @eval which breaks incremental compilation on Julia 1.12.

Moving the methods to Models (following the existing pattern for default_included_properties and checkpointer_address) makes them available to downstream packages via standard import.

Companion PR: CliMA/ClimaSeaIce.jl — ss/fix-netcdf-extension

Test plan

• Existing NetCDF writer tests pass
• ClimaSeaIce precompiles successfully with NumericalEarth on Julia 1.12

[giordano]

Related to my comments in CliMA/ClimaSeaIce.jl#127 (review) (including the use of import 🥲), feels like these methods should really live in OutputWriters, not in this extension.

[giordano]

And the fact that you tried to use @eval to circumvent the impossibility to properly define the methods is a strong indication that the methods just currently live in the wrong places (the extensions, where they have nothing specific to the extensions)

[glwagner]

what do you mean? they do live in OutputWriters; they are just extended here. What specifically do you suggest?

[glwagner]

you mean put this chunk of code in OutputWriters?

#####
##### Variable attributes
#####

default_velocity_attributes(::RectilinearGrid) = Dict(
    "u" => Dict("long_name" => "Velocity in the +x-direction.", "units" => "m/s"),
    "v" => Dict("long_name" => "Velocity in the +y-direction.", "units" => "m/s"),
    "w" => Dict("long_name" => "Velocity in the +z-direction.", "units" => "m/s"))

default_velocity_attributes(::LatitudeLongitudeGrid) = Dict(
    "u"            => Dict("long_name" => "Velocity in the zonal direction (+ = east).", "units" => "m/s"),
    "v"            => Dict("long_name" => "Velocity in the meridional direction (+ = north).", "units" => "m/s"),
    "w"            => Dict("long_name" => "Velocity in the vertical direction (+ = up).", "units" => "m/s"),
    "displacement" => Dict("long_name" => "Sea surface height displacement", "units" => "m"))

default_velocity_attributes(ibg::ImmersedBoundaryGrid) = default_velocity_attributes(ibg.underlying_grid)

default_tracer_attributes(::Nothing) = Dict()

default_tracer_attributes(::BuoyancyForce{<:BuoyancyTracer}) = Dict("b" => Dict("long_name" => "Buoyancy", "units" => "m/s²"))

default_tracer_attributes(::BuoyancyForce{<:SeawaterBuoyancy{FT, <:LinearEquationOfState}}) where FT = Dict(
    "T" => Dict("long_name" => "Temperature", "units" => "°C"),
    "S" => Dict("long_name" => "Salinity",    "units" => "practical salinity unit (psu)"))

default_tracer_attributes(::BuoyancyBoussinesqEOSModel) = Dict("T" => Dict("long_name" => "Conservative temperature", "units" => "°C"),
                                                               "S" => Dict("long_name" => "Absolute salinity",        "units" => "g/kg"))

function default_output_attributes(model)
    velocity_attrs = default_velocity_attributes(model.grid)
    buoyancy = model isa ShallowWaterModel ? nothing : model.buoyancy
    tracer_attrs = default_tracer_attributes(buoyancy)
    return merge(velocity_attrs, tracer_attrs)
end

[glwagner]

and

#####
##### Saving schedule metadata as global attributes
#####

add_schedule_metadata!(attributes, schedule) = nothing

function add_schedule_metadata!(global_attributes, schedule::IterationInterval)
    global_attributes["schedule"] = "IterationInterval"
    global_attributes["interval"] = schedule.interval
    global_attributes["output iteration interval"] = "Output was saved every $(schedule.interval) iteration(s)."

    return nothing
end

function add_schedule_metadata!(global_attributes, schedule::TimeInterval)
    global_attributes["schedule"] = "TimeInterval"
    global_attributes["interval"] = schedule.interval
    global_attributes["output time interval"] = "Output was saved every $(prettytime(schedule.interval))."

    return nothing
end

function add_schedule_metadata!(global_attributes, schedule::WallTimeInterval)
    global_attributes["schedule"] = "WallTimeInterval"
    global_attributes["interval"] = schedule.interval
    global_attributes["output time interval"] =
        "Output was saved every $(prettytime(schedule.interval))."

    return nothing
end

function add_schedule_metadata!(global_attributes, schedule::AveragedTimeInterval)
    global_attributes["schedule"] = "AveragedTimeInterval"
    global_attributes["interval"] = schedule.interval
    global_attributes["output time interval"] = "Output was time-averaged and saved every $(prettytime(schedule.interval))."

    global_attributes["time_averaging_window"] = schedule.window
    global_attributes["time averaging window"] = "Output was time averaged with a window size of $(prettytime(schedule.window))"

    global_attributes["time_averaging_stride"] = schedule.stride
    global_attributes["time averaging stride"] = "Output was time averaged with a stride of $(schedule.stride) iteration(s) within the time averaging window."

    return nothing
end

[giordano]

what do you mean? they do live in OutputWriters; they are just extended here.

The functions are declared in OutputWriters, the methods are defined in the extensions.

What specifically do you suggest?

To have the methods in OutputWriters, not in the extensions. I don't see anything specific to NCDatasets in those methods.

[glwagner]

yes you're right. there is no dependence on NCDatasets as a package; it is just that the methods are specific to NetCDF. But there is no functional barrier to putting them in OutputWriters. I can do that.

[giordano]

I guess these methods were initially needed only internally in the extension, and that's fair, but then when they grew to be needed in downstream packages having them in the extension became odd, and really not technically necessary.

[simone-silvestri]

there is the problem that OutputWriters is loaded before many other modules that these functions require. We could load OutputWriters after actually. I think functions in OutputWriters are only extended in Modules

[glwagner]

there is the problem that OutputWriters is loaded before many other modules that these functions require. We could load OutputWriters after actually. I think functions in OutputWriters are only extended in Modules

We can define the functions in OutputWriters, and then add methods in the modules. That's the canonical way.