Skip to content

CHANGELOG.md

Latest commit

 

History

History
53 lines (37 loc) · 7.58 KB

CHANGELOG.md

File metadata and controls

53 lines (37 loc) · 7.58 KB
Raw

Changelog

2.0.0

Breaking

  • HWID v2: the HWID is derived from the OS machine id. sha256("authforge-hwid-v2|" + app_id + "|" + machine_id) as lowercase hex, where machine_id is the trimmed, lowercased OS id: macOS IOPlatformUUID (from /usr/sbin/ioreg), Linux /etc/machine-id then /var/lib/dbus/machine-id, Windows HKLM\SOFTWARE\Microsoft\Cryptography\MachineGuid (64-bit registry view, via winreg). The previous MAC + CPU + disk serial fingerprint changed whenever the first network interface changed (macOS rotates private Wi-Fi addresses per network); it is kept only as a fallback, as legacy:<old fingerprint>, when no valid OS id can be read.
  • Every bound HWID changes once on upgrade. Online, a device rebinds into a free HWID slot, or the license needs an HWID reset. Offline .authforge files bound to a pre-2.0 HWID fail with hwid_mismatch and must be re-issued (collect the new HWID with create_activation_request()).
  • HWIDs are now identical across all AuthForge SDKs for the same app id and machine, and different between apps on the same machine. Containers without a unique /etc/machine-id should pass hwid_override.
  • hwid_override behaves as before.
  • login() fires on_failure once on a network failure. When /auth/validate is still unreachable after the network retry, login() now calls on_failure("login_failed", exc) once, with exc.code network_error or timeout, and returns False. Previously it called on_failure("network_error", exc) and then on_failure("login_failed", exc). Without a callback the process still ends with os._exit(1), now on login_failed. Matches check-ins (1.4.0), the C++ SDK and the C# SDK's Login(). Callers that react to the network_error reason should check exc.code on login_failed instead.
  • self_ban() reports a network failure once, by raising. When /auth/selfban is still unreachable after the network retry, self_ban() raises AuthForgeError with exc.code network_error or timeout and never calls on_failure or os._exit. Previously it called on_failure("network_error", exc) (or, without a callback, os._exit(1)) and then raised. Server rejections already only raised. Matches activate_offline(). The SDK no longer passes a network_error reason to on_failure anywhere.

Added

  • Module-level helpers hwid_from_machine_id, normalize_machine_id, parse_ioreg_platform_uuid, parse_reg_query_machine_guid.
  • Tests run the shared cross-SDK conformance vectors hwid_vectors.json (generated by authforge-node/generate_hwid_vectors.mjs), plus a macOS-only live check against ioreg.

1.5.0

Features

  • Self-serve offline activation: AuthForgeClient.activate_offline(license_key, path=None) -> bool calls POST /auth/offline/activate once and turns a license key into a lifetime .authforge file bound to this machine's HWID (the same HWID login() sends). The SDK verifies the file locally (public key(s), app id, HWID) before writing it, writes it byte-for-byte and atomically to path (creating parent directories), and authenticates the client exactly as login_from_file() does. get_offline_license()["replay"] is True when the server returned a file the machine already had. Needs the App Secret. Billing: 1 new offline activation = 2 credits for the app owner; re-activating the same machine is free.
  • On failure activate_offline() raises AuthForgeError, writes nothing (an existing file at path is left untouched), does not authenticate the client (an existing session is kept) and never calls on_failure or os._exit. An empty license key, then a missing App Secret, raise the usual ValueError before any request.
  • New server codes, all fatal and added to KNOWN_SERVER_ERRORS and DEFINITIVE_ERROR_CODES: offline_activation_disabled, offline_activation_requires_perpetual, offline_activation_requires_seats, offline_activation_limit_reached.
  • New SDK codes: offline_file_rejected (fatal, also listed in KNOWN_SERVER_ERRORS and DEFINITIVE_ERROR_CODES; the returned file failed local verification, the verifier's reason is in the message) and file_write_failed (transient; the verified file could not be written). A success response without a file raises unexpected_response (transient).
  • Tests run the shared self_serve_activation_vectors.json conformance vectors.

1.4.1

Fixes

  • Without an on_failure callback, a transient background check failure (network outage, timeout, rate_limited, system_error, no_credits, ...) no longer ends the process. The SDK prints one line to stderr, AuthForge: background check failed (<code>); retrying next interval, keeps the session and checks in again on the next interval. Previously a brief outage killed any app that enabled online check-ins without setting a callback.
  • Unchanged: without a callback, definitive failures (including the session_expired a transient failure becomes once the session TTL has passed) and failed login() calls still call os._exit(1).

Docs

  • The README and AGENTS.md examples no longer call os._exit(1) from on_failure. They set a threading.Event that the main thread checks so it can save work and exit, mention _thread.interrupt_main() for blocked main threads, and keep os._exit as a last resort after saving.

1.4.0

Behavior changes for callers

  • on_failure("heartbeat_failed", exc) now always receives an AuthForgeError (a ValueError subclass) with exc.code, exc.transient and exc.fatal. For server codes str(exc) equals the code. is_transient_error() and DEFINITIVE_ERROR_CODES are exported; TRANSIENT_ERROR_CODES lists the named transient codes for reference.
  • Classification: only revoked, expired, hwid_mismatch, blocked, session_expired, malformed_request, app_disabled, invalid_app and signature_mismatch are definitive. Everything else is transient, including no_credits, demo_quota_exceeded, app_burn_cap_reached, bad_request, invalid_key, every http_error_<status> and unknown codes.
  • Transient heartbeat failures are reported and background checks keep running on the next interval. Previously the loop stopped after any failure. After the session TTL passes, the next transient failure is reported as a definitive session_expired; grace-period expiry also reports session_expired.
  • Definitive heartbeat failures clear the stored session (as logout() does) and stop background checks before on_failure runs. Previously the session stayed authenticated.
  • New code unexpected_response (transient): a failed check-in whose body is not a JSON object with "status": "failed" and a non-empty error string. The message includes the raw status and error.
  • Request retries: only rate_limited, or HTTP 429 with no error code, is retried (after 2s, then 5s). no_credits, demo_quota_exceeded and app_burn_cap_reached on HTTP 429 are no longer retried. This also applies to login() and validate_license().
  • Heartbeat network failures no longer also fire on_failure("network_error", ...); they are reported once as heartbeat_failed with code network_error or timeout.
  • Server codes this SDK doesn't know are passed through as-is instead of becoming unknown_error (also in validate_license() results).
  • HTTP errors with a non-JSON body (http_error_<status>), network failures and unparseable responses now raise AuthForgeError instead of RuntimeError / plain ValueError.
  • on_failure runs with no SDK lock held, so it may call logout(), is_authenticated() or login(). A check-in still in flight during logout() or login() is discarded instead of restoring the old session.