"""Add ``cmdb_cost_facts.auto_owner_source`` so the dashboard can
distinguish rule-driven attribution from scope-owner fallback.

Revision ID: 145
Revises: 144
Create Date: 2026-05-10

Why:

* PR #240 left attribution wholly rule-driven: the ingest walked
  ``cmdb_cost_allocation_rules`` in priority order and stamped the
  first match's ``cost_owner_person_id`` onto the fact row. When no
  rule matched, the fact landed with ``cost_owner_person_id=NULL``
  and surfaced as "未分摊" in the dashboard.
* Production has 0 allocation rules — admins haven't authored any,
  and the auto-grant flows that *would* create rules (Azure OpenAI,
  Bedrock principal-tag) have not run for the existing 4 keys. Net:
  10,502 facts (~$7,332.99 / 90d) were 100% unallocated even though
  every fact had ``scope_id`` set and every active scope had
  ``owner_person_id`` populated. The chain fact → scope → owner →
  department was already complete in the data; the rule engine
  simply didn't walk it.
* The fallback added in the same PR as this migration walks that
  chain when the rule ladder misses, but the dashboard needs to be
  honest about *which path* attributed each fact: rule-driven
  (admin-authored override), scope-owner-fallback (subscription /
  project / sub-account default owner), or unattributed (no
  scope_id or scope had no owner). Without this column, the
  recompute summary cannot tell the operator "X rules matched, Y
  facts attributed via fallback, Z still unattributed".

Schema:

* ``auto_owner_source VARCHAR(16) NOT NULL DEFAULT 'none'`` — chosen
  over ENUM so a future addition (``iam_owner``, ``tag_owner``,
  ``rg_owner`` for Azure RG-level overrides) doesn't require an
  ALTER TYPE on TiDB. Allowed values are documented in the column
  doc and validated at the service layer; storing as a free string
  matches the broader codebase convention.
* Default ``'none'`` so existing rows backfill safely without
  scanning every row — TiDB handles this as a metadata-only DDL.
* Index on ``(auto_owner_source, date)`` because the dashboard's
  "未分摊 KPI tooltip" needs to count facts where source='none' in
  the chosen window without a full table scan; existing
  ``ix_facts_owner_date`` already covers ``cost_owner_person_id``
  but not the source dimension.

Downgrade: drop the index and the column. No data loss because the
column is purely a derived classification — recompute_allocation
re-stamps it from scratch on next run.
"""
from __future__ import annotations

import sqlalchemy as sa

from alembic import op

revision = "145"
down_revision = "144"
branch_labels = None
depends_on = None


def _table_exists(bind: sa.engine.Connection, name: str) -> bool:
    inspector = sa.inspect(bind)
    return name in inspector.get_table_names()


def _column_exists(
    bind: sa.engine.Connection, table: str, column: str,
) -> bool:
    inspector = sa.inspect(bind)
    if table not in inspector.get_table_names():
        return False
    return any(c["name"] == column for c in inspector.get_columns(table))


def _index_exists(
    bind: sa.engine.Connection, table: str, index: str,
) -> bool:
    inspector = sa.inspect(bind)
    if table not in inspector.get_table_names():
        return False
    return any(ix["name"] == index for ix in inspector.get_indexes(table))


def upgrade() -> None:
    bind = op.get_bind()
    if not _table_exists(bind, "cmdb_cost_facts"):
        return
    if not _column_exists(bind, "cmdb_cost_facts", "auto_owner_source"):
        op.add_column(
            "cmdb_cost_facts",
            sa.Column(
                "auto_owner_source",
                sa.String(16),
                nullable=False,
                server_default=sa.text("'none'"),
            ),
        )
    if not _index_exists(bind, "cmdb_cost_facts", "ix_facts_source_date"):
        op.create_index(
            "ix_facts_source_date",
            "cmdb_cost_facts",
            ["auto_owner_source", "date"],
        )


def downgrade() -> None:
    bind = op.get_bind()
    if not _table_exists(bind, "cmdb_cost_facts"):
        return
    if _index_exists(bind, "cmdb_cost_facts", "ix_facts_source_date"):
        op.drop_index("ix_facts_source_date", table_name="cmdb_cost_facts")
    if _column_exists(bind, "cmdb_cost_facts", "auto_owner_source"):
        op.drop_column("cmdb_cost_facts", "auto_owner_source")