Skip to content

HackDigest 0.5.0

Latest

Choose a tag to compare

@github-actions github-actions released this 10 Nov 02:45
· 26 commits to main since this release
8f4d926

Maintenance release: a data-robustness fix, the glaze dependency caught
up to its first tagged release, and the product site moves onto the
family design system. No new features; the app's settings schema and
routes are unchanged.

Fixed

  • A UTF-8 BOM in settings.json (produced by BOM-emitting editors such as PowerShell 5.1's Set-Content -Encoding UTF8 or legacy Notepad saves) made read-json throw, and the reader's error handler silently turned the whole configuration into an empty map — the provider config vanished with no diagnostics and the startup key migration saw no key. The data-file reader now strips a leading BOM once, for settings.json and library.json alike, and notes it on stderr (#47, regression-tested).
  • Windows WebView2 controller creation failing with E_ABORT when several app instances or windows raced the machine-wide glaze-webview2 user-data folder: absorbed by the glaze v0.18.0 pin (one data folder per process, GLAZE_WEBVIEW2_DATA_DIR for a stable cache path) together with automatic retries for transient controller aborts.

Changed

  • The glaze pin flips from the 0.4-era main head (671ea4d) to the v0.18.0 release across CI and release workflows. Runtime deltas absorbed: per-process WebView2 data folders on Windows, atomic directory replacement in the updater, AppImage payload resolution through $APPDIR, Applications drop target and unique volume names for plain dmg builds, and a native Windows ARM64 webview backend. Schema compatibility verified (zero breaking changes); all four packaged-native CI legs selftest the v0.18.0 build.
  • The product site homepage is rebuilt on the family site-kit (v1.0.3): family design system (terracotta accent, Newsreader display), browser-locale language detection with a persisted override, and a unified download resolver over the real release assets.

Release infrastructure

  • A keep-alive regression test locks in the #40 finding (the reported persistent-connection misframe was a mock artifact — llm.rkt opens a fresh connection per request and Racket sends Connection: close by default): 30 sequential titles batches against a real keep-alive server with server-side framing/parse assertions.