Split the complex with_advisory_lock function into smaller, focused
functions to improve readability and reduce cyclomatic complexity:
- with_advisory_lock_in_transaction: Handles case when already in transaction
- with_advisory_lock_new_transaction: Handles case when starting new transaction
- execute_within_transaction: Executes function within transaction
- normalize_fun_result: Normalizes function results to consistent format
- handle_transaction_result: Handles transaction results and sends notifications
This fixes Credo warnings:
- Function body nested too deep (was 3, now max 2)
- Function too complex (was 11, now max 9)
Change validation to fail closed instead of fail open when types cannot
be loaded. This prevents inconsistent data states and provides clearer
error messages to users.
- Replace Enum.map |> Enum.join with Enum.map_join for efficiency
- Extract helper functions to reduce nesting depth from 4 to 2
- Rename is_current_cycle? to current_cycle? following Elixir conventions
- Implemented regenerate_cycles_on_type_change helper in Member resource
- Cycles that haven't ended yet (cycle_end >= today) are deleted and regenerated
- Paid and suspended cycles remain unchanged (not deleted)
- CycleGenerator reloads member with new membership_fee_type_id
- Adjusted tests to work with current cycles only (no future cycles)
- All integration tests passing
Phase 4 completed: Cycle regeneration on type change
- Remove Process.sleep calls from integration tests (tests run synchronously in SQL sandbox)
- Improve error handling: membership_fee_type_not_found now returns changeset error instead of just logging
- Clarify partial_failure documentation: successful_cycles are not persisted on rollback
- Update documentation: joined_at → join_date, left_at → exit_date
- Document PostgreSQL advisory locks per member (not whole table lock)
- Document gap handling: explicitly deleted cycles are not recreated
- Use return_notifications?: true when creating cycles within transaction
- Collect notifications and send them after transaction commits
- Prevents 'Missed notifications' warnings in test output
- Notifications are now properly sent via Ash.Notifier.notify/1
- Change algorithm to start from last existing cycle instead of start_date
- Deleted cycles (gaps) are no longer automatically filled
- Add test to verify gaps remain unfilled
- Update documentation to clarify gap handling behavior
- Handle Task crashes in async_stream with {:exit, reason}
- Return {:error, {:partial_failure, successes, errors}} when some cycles fail
- Previously returned {:ok, successful} even on partial failures
- Improves debuggability and allows callers to handle partial failures
- Remove exit_date filter from generate_cycles_for_all_members query
- Inactive members now get cycles generated up to their exit_date
- Add tests for inactive member processing and exit_date boundary
- Document exit_date == cycle_start behavior (cycle still generated)
- Add warning logging for unexpected errors (not missing prerequisites)
- Use CalendarCycles.interval() type instead of generic atom()
- Update moduledoc to reflect actual usage (no where clause needed)
- Log warnings when cycle generation fails in Member create/update
- Extract generate_fn to reduce code duplication
- Improves debuggability of silent failures
- Replace Application.get_env(:mv, :env) with :sql_sandbox config
- Remove redundant :env config from test.exs
- More explicit and less error-prone for test environment detection
- Escape SQL LIKE wildcards (% and _) to prevent pattern injection
- Limit search query length to 100 characters
- Apply sanitization in both :search action and linking filters
- FTS and fuzzy search use unsanitized query (wildcards not special there)
Custom field LIKE queries on JSONB are expensive (no index).
User linking only needs name/email search for autocomplete.
Custom fields are still searchable via main member search (uses FTS index).
Remove unnecessary credo:disable as function complexity is now acceptable.
The fields parameter was accepted but never used in the :search action.
Simplify API to only accept the query parameter.
Update @doc to reflect the actual functionality.
Explain the two-tier matching approach:
- % operator with server-wide threshold (0.3) for fast index scans
- similarity functions with configurable threshold (0.2) for edge cases
Add rationale for threshold value based on German name testing
- Replace 4 LIKE checks with 2 in build_custom_field_filter
- Simplify CASE blocks in migration trigger functions
- ->> operator always returns text, no need for -> + ::text fallback
- Performance improvement: 50% fewer LIKE operations
