generate_dataset.jl

using Distributed
using JuMP
using UnitCommitment

const DEFAULT_INPUT_ROOT = "instances/matpower"
const DEFAULT_OUTPUT_ROOT = "../UnitCommitment_Trajectory_Dataset"
const VARIANTS = ("hourly_noline", "hourly_withline", "subhourly_noline", "subhourly_withline")

function _build_noline_formulation()
    return UnitCommitment.Formulation(
        transmission = UnitCommitment.ShiftFactorsFormulation(
            precomputed_isf = zeros(0, 0),
            precomputed_lodf = zeros(0, 0),
        ),
    )
end

function _write_mps(model::JuMP.Model, path::String)
    mkpath(dirname(path))
    JuMP.write_to_file(model, path)
    return nothing
end

function _list_json_gz(case_dir::String)
    files = filter(f -> endswith(f, ".json.gz"), readdir(case_dir))
    sort!(files)
    return files
end

function discover_matpower_cases(input_root::String = DEFAULT_INPUT_ROOT)
    isdir(input_root) || error("Input directory does not exist: $input_root")

    case_dirs = filter(d -> isdir(joinpath(input_root, d)), readdir(input_root))
    sort!(case_dirs)

    return [
        (case_name, joinpath(input_root, case_name))
        for case_name in case_dirs
        if !isempty(_list_json_gz(joinpath(input_root, case_name)))
    ]
end

function _parse_case_filter()
    raw = strip(get(ENV, "UC_CASES", ""))
    isempty(raw) && return nothing
    return Set(strip.(split(raw, ",")))
end

function _is_truthy_env(name::String)
    value = lowercase(strip(get(ENV, name, "")))
    return value in ("1", "true", "yes", "y")
end

function _parse_positive_int_env(name::String, default::Int)
    raw = strip(get(ENV, name, ""))
    isempty(raw) && return default

    value = tryparse(Int, raw)
    if value === nothing || value < 1
        error("$name must be a positive integer, got: $raw")
    end
    return value
end

function _default_worker_count()
    return nprocs() > 1 ? length(workers()) : 1
end

function _requested_worker_count()
    return _parse_positive_int_env("UC_WORKERS", _default_worker_count())
end

function _selected_cases(input_root::String)
    cases = discover_matpower_cases(input_root)
    selected = _parse_case_filter()
    selected === nothing && return cases
    return filter(case -> case[1] in selected, cases)
end

function _ensure_worker_count!(requested_workers::Int)
    requested_workers <= 1 && return Int[]

    existing_workers = nprocs() > 1 ? workers() : Int[]
    extra_workers = requested_workers - length(existing_workers)

    if extra_workers > 0
        project_file = Base.active_project()
        if project_file === nothing
            addprocs(extra_workers)
        else
            addprocs(extra_workers; exeflags = `--project=$(dirname(project_file))`)
        end
    end

    return workers()[1:requested_workers]
end

function _load_script_on_workers!(worker_ids::Vector{Int})
    isempty(worker_ids) && return nothing

    script_path = abspath(@__FILE__)
    @sync for pid in worker_ids
        @async remotecall_wait(pid, script_path) do path
            include(path)
            return nothing
        end
    end

    return nothing
end

function _generate_one_instance!(
    case_name::AbstractString,
    date_tag::AbstractString,
    src_path::AbstractString,
    output_root::AbstractString,
    noline_formulation,
)
    inst_hourly = UnitCommitment.read(src_path)
    inst_hourly_noline = deepcopy(inst_hourly)
    empty!(inst_hourly_noline.scenarios[1].lines)

    model_hourly_noline = UnitCommitment.build_model(
        instance = inst_hourly_noline,
        formulation = noline_formulation,
        variable_names = true,
    )
    _write_mps(
        model_hourly_noline,
        joinpath(output_root, case_name, "hourly_noline", "$(case_name)_$(date_tag)_h_noline.mps"),
    )

    model_hourly_withline = UnitCommitment.build_model(
        instance = inst_hourly,
        variable_names = true,
    )
    _write_mps(
        model_hourly_withline,
        joinpath(output_root, case_name, "hourly_withline", "$(case_name)_$(date_tag)_h_withline.mps"),
    )

    inst_sub = UnitCommitment.convert_to_subhourly(inst_hourly, inst_hourly)
    inst_sub_noline = deepcopy(inst_sub)
    empty!(inst_sub_noline.scenarios[1].lines)

    model_sub_noline = UnitCommitment.build_model(
        instance = inst_sub_noline,
        formulation = noline_formulation,
        variable_names = true,
    )
    _write_mps(
        model_sub_noline,
        joinpath(output_root, case_name, "subhourly_noline", "$(case_name)_$(date_tag)_s_noline.mps"),
    )

    model_sub_withline = UnitCommitment.build_model(
        instance = inst_sub,
        variable_names = true,
    )
    _write_mps(
        model_sub_withline,
        joinpath(output_root, case_name, "subhourly_withline", "$(case_name)_$(date_tag)_s_withline.mps"),
    )

    return nothing
end

function _build_generation_tasks(cases, output_root::AbstractString)
    tasks = NamedTuple[]
    case_count = length(cases)

    for (case_index, (case_name, case_dir)) in enumerate(cases)
        files = _list_json_gz(case_dir)
        instance_count = length(files)

        for (instance_index, file_name) in enumerate(files)
            date_tag = split(file_name, ".")[1]
            src_path = joinpath(case_dir, file_name)
            push!(
                tasks,
                (;
                    case_index,
                    case_count,
                    case_name,
                    instance_index,
                    instance_count,
                    date_tag,
                    src_path,
                    output_root,
                ),
            )
        end
    end

    return tasks
end

function _prepare_output_dirs!(cases, output_root::AbstractString)
    for (case_name, _) in cases
        for variant in VARIANTS
            mkpath(joinpath(output_root, case_name, variant))
        end
    end
    return nothing
end

function _print_case_plan(cases)
    for (case_index, (case_name, case_dir)) in enumerate(cases)
        println("[$case_index/$(length(cases))] $case_name ($(_list_json_gz(case_dir) |> length) instances)")
    end
    return nothing
end

function _print_task_done(task, pid::Int)
    println(
        "  [$(task.case_index)/$(task.case_count) $(task.case_name)] " *
        "$(task.instance_index)/$(task.instance_count) $(task.date_tag) pid=$pid",
    )
    flush(stdout)
    return nothing
end

function _generate_task!(task)
    noline_formulation = _build_noline_formulation()

    _generate_one_instance!(
        task.case_name,
        task.date_tag,
        task.src_path,
        task.output_root,
        noline_formulation,
    )

    GC.gc()
    _print_task_done(task, myid())
    return nothing
end

function _generate_parallel!(tasks, worker_ids::Vector{Int})
    pool = CachingPool(worker_ids)
    pmap(_generate_task!, pool, tasks; batch_size = 1)
    return nothing
end

function generate_dataset(;
    input_root::String = get(ENV, "UC_INPUT_ROOT", DEFAULT_INPUT_ROOT),
    output_root::String = get(ENV, "UC_OUTPUT_ROOT", DEFAULT_OUTPUT_ROOT),
)
    cases = _selected_cases(input_root)
    isempty(cases) && error("No cases selected under $input_root. Check UC_CASES or the input directory.")

    mkpath(output_root)
    requested_workers = _requested_worker_count()

    total_instances = sum(length(_list_json_gz(case_dir)) for (_, case_dir) in cases)
    println("Input root:  $input_root")
    println("Output root: $output_root")
    println("Cases:       $(length(cases))")
    println("Instances:   $total_instances")
    println("Variants:    $(length(VARIANTS))")
    println("MPS files:   $(total_instances * length(VARIANTS))")
    println("Workers:     $requested_workers")

    if _is_truthy_env("UC_DRY_RUN")
        println("\nDry run only. Set UC_DRY_RUN=0 or remove it to generate MPS files.")
        for (case_name, case_dir) in cases
            println("  $case_name: $(length(_list_json_gz(case_dir))) instances")
        end
        return nothing
    end

    _prepare_output_dirs!(cases, output_root)
    tasks = _build_generation_tasks(cases, output_root)

    println()
    _print_case_plan(cases)
    println()

    if requested_workers <= 1
        for task in tasks
            _generate_task!(task)
        end
    else
        worker_ids = _ensure_worker_count!(requested_workers)
        _load_script_on_workers!(worker_ids)
        println("Parallel worker pids: $(join(worker_ids, ", "))")
        _generate_parallel!(tasks, worker_ids)
    end

    println("\nDone. output_root=$output_root")
    return nothing
end

if myid() == 1 && abspath(PROGRAM_FILE) == @__FILE__
    generate_dataset()
end