[shipmill-romamo]

Summary

Builds S-007-12 and S-007-13 of docs/specs/007-data-copy-schemas.md.

• PostgresSchemaName now takes exactly the spec's names: ^[a-z_][a-z0-9_]*$, at most 63 characters (read with fullmatch, so a trailing newline can't pass). Braces, quotes, semicolons and control characters fall outside it, so the Copy only the named schemas in data copy (#124) #134 guard against Ansible templating stays.
• DataCopySchemas refuses a name given twice instead of silently dropping the repeat.
• Behaviour change: a malformed or repeated --source-schema (CLI and the data_copy MCP tool) now exits PRECONDITION (exit 4, context code data_copy_schema_invalid) from the command, before any playbook, as the spec's Behaviour and S-007-12 say. Since Copy only the named schemas in data copy (#124) #134 it was a usage error (exit 2) raised from DataMigrateArgs.__post_init__. One exception: a value holding a newline is still refused by the CLI framework's own argv guard as ARG_ERROR (exit 2) before the command runs; no playbook runs either way.
• cloudfall migrate --data-schema DATABASE=NAME (repeatable, MCP data_schema) narrows the data:<database> step to the named schemas, passing the same DataCopySchemas to migrate_data that data copy --source-schema does; the plan step's description names them. A --data-schema for a database with no --data exits PRECONDITION with code migrate_data_schema_without_data, in plan mode as with --yes, before any step. A --data-schema value without = is a usage error, as a malformed --data already is.
• The Neon and Render guides name the rule and --data-schema.

Plan

• Serves: G-3
• Files: sdk/src/cloudfall/domain.py, sdk/src/cloudfall/lifecycle.py, sdk/src/cloudfall/app.py, sdk/src/cloudfall/migrate.py, sdk/tests/test_lifecycle.py, sdk/tests/test_data_migration.py, sdk/tests/test_migrate.py, sdk/tests/test_agent_server.py, sdk/tests/test_app.py, docs/neon-migration-guide.md, docs/render-migration-guide.md, changelog.d/126-source-schema-checks.md
• Order: failing tests first (S-007-12 over argv and the MCP data_copy and migrate tools, S-007-13 on the migrate data step's playbook argv), then PostgresSchemaName to ^[a-z_][a-z0-9_]*$ and at most 63 characters, DataCopySchemas refusing a repeat, both answered as PRECONDITION in the command instead of a usage error, then migrate --data-schema DATABASE=NAME through MigrateOptions to the data:<database> step; Behaviour as in spec 007, "Choosing the schemas"
• Proof: S-007-12, S-007-13

Test plan

• S-007-12: test_s007_12_data_copy_refuses_a_malformed_or_repeated_schema (Bad;Name, app app, empty, Jinja, tab, upper case, leading digit, quote, 64 characters; with and without --yes: exit 4, no inventory render), test_s007_12_data_copy_takes_the_longest_schema_name, test_s007_12_cli_migrate_refuses_a_bad_data_schema_before_any_playbook (no --data, malformed, repeated; with --yes: no plan file, no inventory), test_s007_12_migrate_refuses_a_data_schema_without_its_data (no engine step started), test_s007_12_data_copy_schemas_refuse_a_repeat_and_a_malformed_name, test_s007_12_the_data_tools_refuse_a_bad_schema_as_the_cli_does (over the real MCP stdio transport)
• S-007-13: test_s007_13_migrate_copies_only_the_data_schema_as_data_copy_does (the resumed data step's playbook argv equals data copy --source-schema public's), test_s007_13_migrate_groups_each_data_schema_under_its_database
• uv run ruff check ., uv run mypy, ansible-lint engine/ansible, and the touched test files pass locally; the full suite is left to CI
• Not run against a PostgreSQL server: nothing here changes the engine, and the copy itself is Copy only the named schemas in data copy (#124) #134's

Fixes #126

🤖 Generated with Claude Code