ElixirOffice 0.3.2

ElixirOffice desktop 0.3.2

Released 2026-10-23 (beta). Linux, Windows and macOS (Apple Silicon) packages; see the known limitations below.

Changed and fixed

• Release notices under your control. Check for updates is on by default and can be disabled in File > Options > Updates. The app makes at most one bounded background GitHub release-metadata attempt per day; prereleases are included. Release notes and Download open the matching release page in the system browser. Remind me later and Skip this version are available. Documents, file names, paths, recent files and user identity are not sent; GitHub receives ordinary HTTPS connection metadata, including the IP address. No asset download or installation runs inside the app. See release notices and privacy.
• Optional compact desktop title bar on Linux/X11. Options > Window > Use a custom title bar combines the title or tabs with the existing Quick Access actions and command search in one band. It applies to new windows and remains off by default. Windows, macOS, Wayland and unknown backends retain the native frame. A bounded native-frame recovery preserves live documents when compact startup or controls fail. See compact title bar.
• Aligned document icons. Native-frame header icons align with Save, Undo, Redo and the title in the desktop adapters for Word, Cell, Slide and PDF. Hosted editor styling is unchanged.
• Less repeated computer-font name work. Cache normalized font names during matching rather than repeatedly rebuilding the same normalized strings. This addresses the demonstrated large-font-list hot path behind the Playwrite investigation (issue31); improvement on the reporter's Windows machine remains unmeasured. See computer fonts.
• Reliable AppImage release inputs. Maintainer packaging uses a persistent, manifest-addressed raw tool store. Planning and execution check all pinned file sizes and hashes before reserving a build number. Packaging stages its own verified copies and disables download fallback, including if an input disappears after verification. Tool, runtime and builder-image pins are unchanged. See release preflight.

Validation and limits

• Linux/Windows build18 package checks passed; final macOS and physical-platform acceptance remain separately recorded. Source/UI tests and cross-reviews do not certify new Linux, Windows or macOS installers, physical printing, native browser actions, Gatekeeper, all display scales or all window managers.
• Compact-window source evidence covers an isolated Linux/X11 environment. Other platforms keep the native frame; no physical Windows/macOS or Wayland compact qualification is claimed. Actual Linux header geometry covers four editors in light/dark, native/compact configurations.
• Release-notice labels/privacy strings exist in 45 desktop locales; linguistic review is deferred to 0.3.3. Presence tests are not a linguistic verdict.
• The owner's earlier 30‚Äì60-second Windows opening freeze with high CPU on a Core i3 / 8 GiB remains unproven as resolved. New-window Wasm compilation and the existing deferred font discovery remain; no cross-editor compiled-module reuse is included. Windows tester timing and copied content-free opening diagnostics remain useful.
• The reported blank printing / Ctrl+P issue32 was not reproduced in the 0.3.1 investigation; no speculative printing fix is included here. Physical printers still require platform testing.
• Automatic installation is not included. The separate signed-update design and unshipped metadata helper do not enable an automatic updater in this release.
• Windows installers remain unsigned; macOS distribution remains Apple Silicon, ad-hoc signed and not notarized, subject to actual final packaging checks.
• Inherited conversion limits remain: the modern LibreOffice 26.8 encrypted ODS envelope is unsupported; strict ODF 1.3 and normative ODF 1.2 validity are not claimed. Desktop original-PDF text editing, OCR, PDF-to-DOCX, offline digital signatures and real-time coediting remain unavailable.

Mac validation provenance

The original macOS input archive omitted two test-only helper sources under tools/desktop-updates. They were supplied from the exact frozen product revision, with independently verified hashes. Mac validation also uses a reviewed test-only overlay: three compact-recovery fixtures explicitly select Linux, since compact frames are Linux-only. The reviewed Mac validation harness also routes the builder logger to stderr to avoid a reproduced Node 22 test-runner framing defect while retaining every diagnostic. One added logging regression brings that validation suite to 843 cases, with two workers; the original Linux-build and Windows-packaging suites remain 842 cases each. No existing assertion or timeout, application code, runtime pin, or build identity was changed. The final native Mac gate result remains separately recorded. The original input archive and private build metadata remain unchanged. Initial failures and subsequent validation receipts are retained in the audit evidence; the final result is recorded separately in BUILD-INFO.json. This does not qualify native GUI behavior, physical printing, or Gatekeeper.

Build qualification

Linux-build and Windows-packaging suites each passed 842/842 with zero failures/skips. Each packaged resource tree matched all 18776 selected entries and both runtime manifests. Clean DEB-only dependency installation passed a six-ELF audit as UID1000, with root-owned4755 sandbox helper and no manually provisioned runtime libraries. Default-worker, public ODS/CBC and static PDF browser/pixel gates passed. These checks do not certify a clean native start in the namespace-restricted container, physical printing, Gatekeeper, or final Windows/macOS GUI behavior.

Windows performance feedback

If opening or creating a document temporarily freezes the interface, report Windows and ElixirOffice version/build, CPU/RAM, approximate delay, first versus subsequent open, blank versus existing document, separate windows versus optional tabs, and CPU/memory observations. Include the app opening diagnostics if available, after reviewing them for private information. Do not share private document names, paths or contents. The original 30–60 second report on a Core i3 with 8 GiB RAM has not been remeasured; the font-name cache does not prove that startup delay fixed.