Changelog

Copy Markdown View Source

All notable changes to mob_sms are documented here.

Format: Keep a Changelog. Versioning: SemVer.

[0.2.2] - 2026-10-01

Changed

  • Releases are signed with the mob first-party key. priv/mob_plugin.pub (the shared first-party public key, fingerprint ed25519:nc56w+1Kx0gIt/4EkHxnMZCKHMzp4+S5kS/HoSzEZkg=) is now committed, and the release workflow regenerates priv/mob_plugin.sig from the current manifest before every Hex publish, so mob_dev hosts verify the package instead of reporting it unsigned.

[0.2.1] - 2026-09-18

⚠ Consumer action required

If you copied the 0.2.0 OTP example from README.md or MobSms.OneTimeCode's @moduledoc, your <TextField> currently reads on_change={:code} (bare atom). Mob's renderer silently drops that form, so no handle_info({:change, :code, value}, _) clause ever fires — the field fills visually from iOS QuickType or Android SMS Retriever, but your assign doesn't run. Change it to on_change={{self(), :code}} (a {pid, tag} tuple).

Fixed

  • Demo screen on_change wiring: MobSms.DemoScreen's OTP <TextField> used the same wrong bare-atom form as the docs. Fixed to on_change={{self(), :code}}. Verified end-to-end on physical iPhone SE (iOS QuickType) and Moto G Power 5G 2024 (Android SMS Retriever) after the fix.
  • Regression test: source-level assertion in MobSms.DemoScreenTest walks demo_screen.ex and fails if any on_change= or on_tap= binding is a bare atom rather than a {pid, tag} tuple. This exact bug is what slipped through 0.2.0's publish; the test pins the form so it can't silently regress.

Docs

  • README + MobSms.OneTimeCode @moduledoc: on_change={:code} → on_change={{self(), :code}} in the usage examples so the pattern the docs teach is one that actually fires. Same correction applied to MobSms.DemoScreen's @moduledoc and inline comment.

[0.2.0] - 2026-09-18

Added

  • MobSms.OneTimeCode.arm(socket, on_receive: tag) — cross-platform SMS-delivered OTP autofill. Both platforms deliver the code as {:change, tag, code} (same shape a text field's on_change fires), so one handle_info clause covers iOS QuickType autofill AND Android's SMS Retriever. Pass the SAME atom as your text field's on_change tag.
    • iOS: no-op at the Elixir layer. Autofill is driven by the new text_content_type: :one_time_code prop on <TextField> (added in mob 0.9.1). iOS Messages parses the incoming SMS, surfaces the code as a QuickType suggestion above the keyboard, user taps to fill. Code flows in via the field's normal on_change. No entitlement, no framework, no server-side hash.
    • Android: registers a scoped BroadcastReceiver (only Google Play Services can broadcast to it, via the com.google.android.gms.auth.api.phone.permission.SEND gate) and calls SmsRetrieverClient.startSmsRetriever(). Delivery lands via a short-lived Elixir proxy process that translates the raw {:sms_otp, code} into {:change, tag, code} and forwards to the caller. No READ_SMS permission, no Play Store review gate.
    • Sad path: on Android, {:sms_otp, :timeout} fires on 5-min retriever timeout, GMS start failure, or eviction by a subsequent arm/2. Optional handler — a retype prompt is a fine default.
    • Server-side SMS format for Android: body + blank line + 11-char hash suffix. Full spec + AppSignatureHelper snippet for computing the hash in MobSms.OneTimeCode's @moduledoc.
  • Demo screen now exercises the OTP flow alongside the composer.

Changed

  • mob floor bumped to ~> 0.9.1. The OTP flow's iOS half needs the text_content_type prop, added in that release. Bump generated apps accordingly.
  • Plugin manifest declares com.google.android.gms:play-services-auth-api-phone:18.0.2 as a gradle_deps contribution. Present on 99%+ of Play-installed devices; aftermarket ROMs without Google Play Services fall straight to {:sms_otp, :timeout} from arm/2.

Fixed (0.2.0 review pass, before publish)

  • Kotlin bridge: re-arm evicts the previous receiver AND delivers {:sms_otp, :timeout} to the prior caller instead of leaving it hanging. Idempotency guard on onReceive (compare receiver identity, not just presence) prevents double-delivery on queued-post-unregister edge. startSmsRetriever wrapped in try/catch for the synchronous-throw path GMS-absent devices sometimes take.
  • Kotlin regex: extractCode now prefers the LAST 4–8-digit run in the SMS body proper (stripping the 11-char hash suffix first). Previous .find(body) behaviour returned the first digit run — "MyApp2024 code is 123456" would deliver "2024" instead of "123456".
  • Zig NIF: sendSmsOtp handles the enif_make_new_binary OOM path.

[0.1.2] - 2026-09-18

Fixed

  • iOS: use-after-free crash on overlapping composes (adversarial review of 0.1.1 caught this). MobSmsDelegate's release_active called removeAllObjects on the retention array — if two composes overlapped, the first one to finish dropped every strong ref, including the still-active second delegate. UIKit then delivered didFinishWithResult: on a dangling pointer. Now takes the finishing delegate and calls removeObject: so overlapping composes each survive to their own callback. Reproduction: fire MobSms.compose/2 twice from a screen without waiting for the first {:sms, _} message; the second sheet's Send/Cancel crashed the app.
  • iOS: top-VC walk now bails when the top VC is mid-transition. Presenting on a VC whose isBeingDismissed or isBeingPresented is set silently no-ops in UIKit and the delegate never fires, leaving the caller waiting forever for a {:sms, _}. The plugin now returns :not_available in that state.
  • Android: JVM-teardown and never-registered-bridge paths delivered no result to the caller. callBridgeCompose's get_jenv and null-bridge-class branches previously returned :error without sending {:sms, :not_available} — the caller waited forever. Both branches now honour the "one message per compose" contract.

Docs

  • Comment on sms_str_from_iolist (iOS NIF) explains the correct-but-double-negate !enif_inspect_binary && !enif_inspect_iolist_as_binary guard so a future reader doesn't "clean it up" into the wrong semantics.

Tests

  • MobSms.compose/2's argument-coercion path is now split into MobSms.normalize_opts/1, a testable helper. The previous tests wrapped the coercion inside assert_raise ErlangError, which fired on the missing NIF instead of on the coercion — deleting the |> to_string() calls did not fail the suite. The rewritten tests observe normalize_opts/1's output directly, so a regression to the coercion is caught. Also asserts Protocol.UndefinedError propagates for values with no String.Chars impl (rather than being swallowed).

[0.1.1] - 2026-09-18

Docs

  • Rework the "not silent-send" section of MobSms's @moduledoc and the same section in README.md so ex_doc's function-link resolver isn't asked to resolve a hypothetical future send/2 on this module. Removes a mix docs warning; the deferred-until-a-user-asks stance on silent send is unchanged.

[0.1.0] - 2026-09-18

Initial release. Cross-platform SMS composer plugin.

Added

  • MobSms.compose/2 — opens the system SMS composer pre-filled with the given :to and :body. Async result delivered as {:sms, atom} to the calling process.

  • iOS: MFMessageComposeViewController presented from the top VC, delegate reports :sent | :cancelled | :failed.

  • Android: Intent.ACTION_SENDTO with smsto: hands off to the user's default SMS app, delivers :composer_opened (Android cannot observe the send outcome).
  • Both platforms deliver :not_available when the device cannot send SMS (simulator on iOS; no SMS-capable app on Android).
  • MobSms.DemoScreen — a ready-to-run screen exercising the composer flow, auto-listed by the host's plugin home if it enumerates Mob.Plugins.screens/0.
  • Plugin manifest declares the MessageUI framework on iOS and no permissions on either platform.