Introduces CHANGELOG.md (Keep a Changelog style, Unreleased section) covering this batch, with F8's breaking change — remote plaintext RPC now refused by default — called out front and center along with the rpctls=1 / rpcallowplaintext=1 recovery steps. Also records the Security / Fixed / Added entries for F1-F7. Updates the tracking doc's status to reflect the completed release-notes + i18n back-fill and the remaining pre-release items (F1/F2 manual repros, CJK subset-font rebuild). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
3.1 KiB
3.1 KiB
Changelog
All notable user-facing changes to ObsidianDragon are documented here. The format loosely follows Keep a Changelog; the project uses Conventional Commits.
[Unreleased]
⚠️ Breaking changes
-
Remote RPC over plain HTTP is now refused by default. If your wallet is configured to reach a remote
rpchost/rpcconnectwithout TLS, it will no longer connect — it previously sent yourrpcuser/rpcpasswordin cleartext (capturable by anyone on the network path) after only a dismissible warning. To reconnect, either:- add
rpctls=1toDRAGONX.conf(preferred, if your daemon supports TLS), or - add
rpcallowplaintext=1toDRAGONX.confto explicitly accept the plaintext link.
Local and embedded daemons (
127.0.0.0/8,localhost,::1) are unaffected. - add
Security
- Refuse remote plaintext RPC credential transmission by default (see Breaking changes above).
- Tightened localhost detection: a hostname that merely starts with
127.(e.g.127.evil.com) is no longer mistaken for a loopback address, so it can no longer bypass the plaintext-RPC protection. - Sapling parameters are now integrity-checked (SHA-256) against pinned canonical digests
before use, instead of only checking that the files exist. A truncated or corrupt parameter
file is caught up front rather than surfacing later as a confusing shielded-operation failure.
(Cached via a
size:mtimemarker so it doesn't re-hash ~48 MB on every launch.)
Fixed
- Daemon crashes are no longer occasionally missed: a race between the UI thread and the process monitor could consume the daemon's exit status, hiding a crash and defeating the automatic-restart cap. The monitor is now the sole reaper.
- A daemon that fails to launch (missing execute permission, wrong architecture, corrupt binary) now reports a precise error immediately instead of briefly showing "running" and then a generic "exited unexpectedly (exit code 127)".
- A quick stop→start no longer triggers a restart storm: the wallet now waits briefly for a previous daemon to release the data-directory lock and shows a clear, non-crash message instead of exhausting the crash-restart budget.
- Failures while writing the daemon binaries or Sapling parameters (disk full, permission denied) are now surfaced clearly up front instead of failing opaquely when the daemon later can't start.
- Directory-creation failures on startup (read-only home, permission denied) now produce a clear "Cannot create " message instead of a confusing downstream "config missing" / "binary not found" error (or, in one path, an uncaught exception).
Added
- A "Taking longer than expected" notice now appears if the daemon is reachable but hasn't
finished initializing after ~45 s (configurable via
ui.toml), with guidance to restart the daemon or open the Console — instead of an indefinite silent spinner. It clears itself automatically once the daemon connects.
Engineering detail and the finding-by-finding rationale for this batch live in
docs/daemon-startup-hardening.md.