Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.3.0] - 2026-08-27

Added

• CredoExceptionSwallow.Checks.Warning.LoggedAndDroppedError (EX9002): detects {:error, _} branches that log a warning and then return a success-shaped value. SilentRescue covers exceptions; this covers the far more common shape in idiomatic Elixir, where nothing is ever raised — the batch finishes as {:ok, stats}, the row is gone, and the only trace is a log line nobody reads.

  The rule is deliberately narrow: only branches that already log at warning or error level are considered. Reaching for Logger.warning is the author saying the failure mattered; a branch that handles the error some other way never made that claim and is left alone.

  Covers case, with/else, try/else, fn, receive and multi-pattern clauses (every Enum.reduce callback has more than one), guarded clauses, and function heads — guarded or not — that pattern-match the failure.

  A branch is accepted when it reports (ErrorReporter.*, Sentry.capture_*, or any configured project-local helper), raises, retries, or propagates:

  • propagation includes returning an error tuple, handing the error back under another name ({:error, r} = err -> ...; err), a trailing case/if/with whose branches propagate, and any returned value that still mentions what the error pattern bound ({count, [reason | errs]}). An accumulator sitting next to the error does not count — that is the shape this check exists for.
  • retries include calling the function the branch lives in, and the GenServer flavour: Process.send_after(self(), :connect, delay).
  • propagating_calls (empty by default) names the project's own doors out, such as a Phoenix send_error(conn, :unprocessable_entity, ...).

  Configurable log_calls, reporting_calls — which accept bare local names, so a project whose reporting goes through a private helper can name it — propagating_calls and skip_test_files.

  Known blind spot, documented in the moduledoc: a branch that accumulates the failure for the caller to report once per batch is the right pattern and cannot be seen one branch at a time. Mark those with # credo:disable-for-next-line, which turns the false positive into a reviewed statement of intent.

  Measured against a 226-file production codebase: 18 candidates, no floods.

Changed

• the check walks the AST itself instead of using Credo.Code.prewalk, so the enclosing function name travels down with it. That is what makes retry loops recognisable.

[0.2.1] - 2026-03-19

Added

• README guidance for function-level rescue support and custom acceptable_calls
• test coverage for project-specific error reporter call names

Changed

• improved Hex package shape by publishing only the relevant source and docs files

[0.2.0] - 2026-03-12

Fixed

• Function-level rescue detection: Now detects silent rescue in def/defp blocks (e.g., defp foo do ... rescue _ -> [] end), not just try/rescue. Previously, function-level rescues were invisible to the check.
• Accumulator bug in has_acceptable_call?: Macro.prewalk accumulator was reset to false on non-matching AST nodes, potentially missing acceptable calls found earlier in sibling branches.

Changed

• Expanded test suite from 5 to 20+ tests covering function-level rescue, accumulator correctness, mixed scenarios, and test file skipping.

[0.1.0] - 2025-01-07

Added

• Initial release
• CredoExceptionSwallow.Checks.Warning.SilentRescue check to detect rescue blocks that silently swallow exceptions
• Detects rescue blocks without proper logging, error reporting, or re-raising
• Configurable acceptable function calls (Logger, Sentry, ErrorReporter, reraise, raise)
• Option to skip test files (enabled by default)
• File exclusion support via Credo's standard files option