"""
Derived-measure formulas must fail loudly, not generate zeros.

Regression: `_eval_formula` caught every exception and returned 0.0, and
`_eval_node` resolved an unknown identifier to 0.0. So a formula whose column
name did not match — `REVENUE - COGS` when the measures are actually
TOTAL_REVENUE / TOTAL_COGS — evaluated to `0 - 0 = 0` for every single row.

The column loaded 100% zero, the liveboard tile rendered blank, and the run
died much later in the post-load quality gate with no indication of WHICH
formula was wrong. Yodeck failed that gate with 12/12 dead derived columns and
no retry — the run just ended.

Formulas are now validated before a single row is generated, and every problem
is reported at once.

Run with: pytest tests/test_formula_validation.py -v
"""

import pytest

from demoprep_app.dataset.engine import (
    FormulaError,
    _eval_formula,
    validate_fact_formulas,
)
from demoprep_app.scenario.blueprint import BlueprintMeasure, FactSpec


def fact(*measures, name="FACT_SALES"):
    return FactSpec(name=name, grain="one row per order",
                    dimension_names=[], measures=list(measures))


def base(name):
    return BlueprintMeasure(name=name, kind="base")


def derived(name, formula):
    return BlueprintMeasure(name=name, kind="derived", formula=formula)


class TestTheYodeckCondition:
    def test_typod_measure_name_is_reported_not_zeroed(self):
        """The exact defect: formula names don't match the measures."""
        spec = fact(
            base("TOTAL_REVENUE"),
            base("TOTAL_COGS"),
            derived("GROSS_MARGIN", "REVENUE - COGS"),
        )
        with pytest.raises(FormulaError) as exc:
            validate_fact_formulas(spec)

        msg = str(exc.value)
        assert "GROSS_MARGIN" in msg
        assert "'REVENUE'" in msg
        assert "unknown" in msg

    def test_suggests_the_intended_measure(self):
        spec = fact(
            base("TOTAL_REVENUE"),
            derived("NET", "TOTAL_REVENUEE * 2"),
        )
        with pytest.raises(FormulaError) as exc:
            validate_fact_formulas(spec)
        assert "did you mean 'TOTAL_REVENUE'" in str(exc.value)

    def test_reports_every_problem_at_once(self):
        """Five typos must not cost five regenerate cycles."""
        spec = fact(
            base("A"),
            derived("D1", "NOPE1 + A"),
            derived("D2", "NOPE2 - A"),
            derived("D3", "NOPE3 * A"),
        )
        with pytest.raises(FormulaError) as exc:
            validate_fact_formulas(spec)
        msg = str(exc.value)
        assert "3 unusable derived measure formula(s)" in msg
        for name in ("NOPE1", "NOPE2", "NOPE3"):
            assert name in msg


class TestOtherStructuralFaults:
    def test_derived_without_formula(self):
        spec = fact(base("A"), derived("D", ""))
        with pytest.raises(FormulaError, match="declared derived but has no formula"):
            validate_fact_formulas(spec)

    def test_unparseable_formula(self):
        spec = fact(base("A"), derived("D", "A +"))
        with pytest.raises(FormulaError, match="does not parse"):
            validate_fact_formulas(spec)

    def test_function_call_rejected(self):
        spec = fact(base("A"), derived("D", "SUM(A)"))
        with pytest.raises(FormulaError, match="calls a function"):
            validate_fact_formulas(spec)

    def test_unsupported_operator_rejected(self):
        spec = fact(base("A"), derived("D", "A % 2"))
        with pytest.raises(FormulaError, match="unsupported operator"):
            validate_fact_formulas(spec)

    def test_comparison_rejected(self):
        spec = fact(base("A"), base("B"), derived("D", "A > B"))
        with pytest.raises(FormulaError, match="unsupported expression"):
            validate_fact_formulas(spec)


class TestValidFormulasPass:
    def test_simple_arithmetic(self):
        spec = fact(
            base("NET_SALES"),
            base("COST_AMOUNT"),
            derived("MARGIN", "NET_SALES - COST_AMOUNT"),
        )
        validate_fact_formulas(spec)  # must not raise

    def test_derived_referencing_derived(self):
        spec = fact(
            base("REVENUE"),
            base("COGS"),
            derived("MARGIN", "REVENUE - COGS"),
            derived("MARGIN_RATE", "MARGIN / REVENUE"),
        )
        validate_fact_formulas(spec)

    def test_constants_and_parens(self):
        spec = fact(base("A"), derived("D", "(A * 100) / 2 + 1"))
        validate_fact_formulas(spec)

    def test_base_measures_are_not_checked(self):
        """A base measure with a stray formula string is not a derived defect."""
        m = BlueprintMeasure(name="A", kind="base", formula="WHATEVER")
        validate_fact_formulas(fact(m))


class TestPerRowEvaluation:
    def test_division_by_zero_is_still_a_legitimate_zero(self):
        """Per-row arithmetic edge cases must stay tolerant."""
        assert _eval_formula("REVENUE / ORDERS", {"REVENUE": 100.0, "ORDERS": 0.0}) == 0.0

    def test_unknown_name_at_eval_time_raises(self):
        with pytest.raises(FormulaError, match="not an available measure"):
            _eval_formula("REVENUE - MISSING", {"REVENUE": 100.0})

    def test_normal_arithmetic(self):
        assert _eval_formula("A - B", {"A": 10.0, "B": 4.0}) == 6.0
        assert _eval_formula("-A", {"A": 3.0}) == -3.0