feat: add OTP logger handler bindings + fix latent bare-ok Result bug

[dbrattli]

What

Started as: add the missing OTP logger handler-management bindings so downstream apps don't fall back to raw emitErlExpr. While doing it we found — and fixed — a latent runtime bug in the existing Supervisor bindings, and documented the underlying gotcha in the bindings guide.

1. New logger handler bindings (src/otp/Logger.fs)

Faithful 1:1 bindings for the handler-management functions (OTP docs):

• add_handler/3, remove_handler/1, get_handler_config/1, set_handler_config/2, set_handler_config/3

Design:

• Typed returns, not obj. ok | {error, term()} → Result<unit, Dynamic>; get_handler_config → Result<BeamMap<Atom, obj>, Dynamic>. Matches the house convention (File/Supervisor).
• Typed config, not obj. Config params are BeamMap<Atom, obj> (the same typed map already used for log metadata in this file). value stays obj since handler-config values are genuinely heterogeneous.
• Fixed update_handler_config's latent bug: it returned unit, silently swallowing {error, _} (e.g. when OTP rejects changing a logger_std_h handler's type at runtime). Now Result<unit, Dynamic>.
• Thin and unopinionated — no logging-policy helpers.

2. Fix: latent bare-ok bug in Supervisor (src/otp/Supervisor.fs)

supervisor:terminate_child/2 and delete_child/2 were typed Result<unit, Atom> but bind to OTP functions that return a bare ok atom. Fable encodes Ok () as the tuple {ok, ok}, so a successful call matched neither Result branch — every success was silently wrong at runtime. (No test covered the ok path; the existing test only hit which_children on a missing supervisor.) Both now use an [<Emit>] case wrapper to bridge ok -> {ok, ok}, the same pattern File/Logger use.

3. Verification

• New test/test_basic_sup.erl — a one_for_one supervisor with a single transient child, wired into the just test pipeline (justfile).
• test/TestSupervisor.fs — 4 new tests covering both the Ok () and Error not_found paths of terminate_child and delete_child (the success path is the one that exposes a missing wrapper).
• test/TestLogger.fs — round-trips a handler through add_handler/3 + remove_handler/1, asserting Ok () (only passes if the bare-ok bridge is correct on the BEAM).

Confirmed against OTP directly that a temporary child's spec is auto-removed on termination — hence the transient child so delete_child can succeed after terminate_child.

just build clean (0 warnings) · just test → 356/356 passing.

Downstream apps can now write, with no emitErlExpr:

logger.remove_handler (Erlang.binaryToAtom "default") |> ignore
logger.add_handler (Erlang.binaryToAtom "default", Erlang.binaryToAtom "logger_std_h", someConfigMap) |> ignore

4. Docs (BINDINGS-GUIDE.md)

Documented the bare-ok trap as a new anti-pattern: a plain Result<unit, _> binding over a bare-ok OTP function compiles cleanly but is silently wrong at runtime, because Fable encodes Ok () as {ok, ok} and the error path masks the bug. Includes the fix, the carve-out (no wrapper needed when OTP already returns {ok, V} | {error, R}), and the rule: always test the success path of a bare-ok binding.

Commits

1. feat: add OTP logger handler-management bindings
2. refactor: type logger handler bindings with Result and BeamMap
3. fix: bridge bare ok in supervisor terminate_child/delete_child
4. docs: warn about Result<unit,_> over bare-ok OTP functions

🤖 Generated with Claude Code

[dbrattli]

Updated to address the obj-heavy typing feedback — replaced raw obj returns/params with the existing house conventions:

• Returns are now Result<_, Dynamic>. ok | {error, term()} → Result<unit, Dynamic> (matching Supervisor/File). add_handler/remove_handler/set_handler_config/update_handler_config use an [<Emit>] case wrapper to bridge OTP's bare ok into Fable's {ok, ok} Result representation — the same pattern Ets.fs/File.fs already use. (A plain binding can't, because Fable encodes Ok () as {ok, ok}, not bare ok.)
• get_handler_config → Result<BeamMap<Atom, obj>, Dynamic>. OTP already returns {ok, Config} | {error, _}, which is Fable's Result shape, so no wrapper is needed.
• config params are now BeamMap<Atom, obj> (the typed map already used for log metadata in this file) instead of obj. value stays obj since handler-config values are heterogeneous.

The round-trip test now asserts Ok () rather than comparing a boxed atom — which only passes if the bare-ok bridge is correct on the BEAM. Transpiled output confirms it:

case logger:add_handler(HandlerId, Modle, Config) of
    ok -> {ok, ok};
    {error, LoggerAddHandlerReason__} -> {error, LoggerAddHandlerReason__}
end

just build clean, just test 352/352.

[dbrattli]

Fixed supervisor:terminate_child/2 and delete_child/2 the same way (commit 9f58118).

Both return a bare ok atom in OTP, but were typed Result<unit, Atom> — which Fable encodes as {ok, ok}. So a successful terminate_child/delete_child never matched Ok () on the F# side. This was a latent bug: no test exercised the ok path (the only existing test hit which_children on a missing supervisor). Both now use the same [<Emit>] case-converter as File/Logger.

Added verification:

• test_basic_sup — a one_for_one supervisor callback module with a single transient counter child (transient, not temporary, so the spec survives terminate_child and can then be delete_child'd).
• 4 tests covering both branches of each binding:
  • terminate_child → Ok () for a running child; Error not_found for an unknown id
  • delete_child → Ok () after the child is terminated; Error not_found for an unknown id

While writing these I confirmed against OTP directly that a temporary child's spec is auto-removed on termination (so delete_child would return not_found) — hence the transient child. The binding was correct; only the test's restart-type assumption needed fixing.

just test → 356/356.