Skip to content

VALIDATION.md

Latest commit

 

History

History
104 lines (86 loc) · 5.83 KB

VALIDATION.md

File metadata and controls

104 lines (86 loc) · 5.83 KB
Raw

Validation record

Project: Keypour 0.1.0. Toolchain: V 0.5.2 (release commit 7647ce1). Validation performed on 2026-10-20 in a Linux x86_64 workspace.

Results established here

Component Verification Result
V application Linux GCC production compilation, CLI help/version/config checks Passed
V core policy Hotkey parsing, config defaults/types/ranges/unknown fields, CLI overrides, literal Unicode text, CRLF/CR handling, controls/invalid UTF-8 rejection, size limits, deadlines/cancellation/focus transitions Passed
Linux adapter GCC C11/gnu11 with -Wall -Wextra -Werror -pedantic Passed
Linux application Ten scenarios against a separate X11 receiver on a private Xvfb server Passed
X11 clipboard protocol Forced INCR chunks, limits, stale-reply isolation, exact encodings; ten deterministic checks Passed
Windows adapter MinGW-w64 x86_64 C11 with -Wall -Wextra -Werror -pedantic Passed
Windows native behavior Separate Win32 EDIT receiver under Wine 9 + private Xvfb Passed, zero test failures
Complete Windows application V cross-compilation with default garbage collector; PE32+ architecture and system-only DLL imports Passed
Windows CLI --version, --help, --check-config under Wine Passed
Linux AppImage Native dependency bundling and extract-and-run CLI smoke test Passed
GitHub workflow YAML parsing and actionlint 1.7.12 Passed
Shell/Python helpers Shell syntax checks and Python compilation Passed
macOS adapter Implementation/API review and arm64 build job supplied Native build/runtime not verified here

The ten X11 application scenarios exercise:

  1. Exact punctuation, upper/lower case, Tab, CRLF and CR normalization, and a trailing Enter; no typing before the selection click.
  2. A custom hotkey, immutable snapshot after the clipboard changes, and optional removal of trailing line breaks.
  3. Consumed Escape cancellation.
  4. Cancellation by pressing the hotkey a second time.
  5. Rejection of unsupported controls before typing any prefix.
  6. Whole-text preflight when a later character is unavailable in the layout.
  7. Caps Lock handling without toggling the user's lock state.
  8. Focus-change cancellation partway through a long paste, with no text delivered to the second test window in that test.
  9. Armed selection timeout.
  10. Large clipboard capture followed by cancellation.

A separate X11 protocol fixture forces INCR even for short text and checks split UTF-8, exact byte limits, oversized headers, understated streaming overflow, late SelectionNotify/partial-INCR timeouts and next-request isolation, wrong encodings, and Latin-1 fallback conversion.

The Windows native suite checks UTF-16/UTF-8 clipboard preservation, Cyrillic, emoji, size bounds, malformed UTF-16, delayed clipboard-owner timeout/recovery, hotkey delivery/repeats, physical versus injected click/Escape handling, exact physical and Unicode text in a separate process, focus checks, and released modifiers. This is Wine integration testing, not proof of behavior on a physical Windows 10 machine.

What still requires the target environment

  • Run the GitHub workflow in the destination repository to establish that the native Windows PowerShell build, macOS arm64 build, and Ubuntu 22.04 package jobs pass on those hosted runners. No GitHub repository/release was published from this workspace.
  • Run macOS permission and event acceptance checks in tests/MANUAL_MACOS.md. This workspace has no Mac SDK or desktop and did not produce a macOS executable.
  • Confirm the keyboard layout and a short punctuation sample in the actual VNC/IPMI viewer/firmware. Keypour cannot acknowledge delivery inside a remote machine. Some viewers reserve hotkeys or reject synthetic/Unicode events.
  • Focus checking is performed before each character; checking focus and injecting the next event are separate OS operations. It does not guarantee atomic delivery across a simultaneous focus switch. It also cannot detect a field change inside the same top-level window unless accompanied by a click.
  • Linux arbitrary Unicode keymap remapping remains experimental. The default physical-key path does not change the system keymap.

Package baseline and signatures

The locally exercised AppImage is built on Ubuntu 24.04 and imports GLIBC_2.38 symbols. It therefore has a glibc 2.38+ baseline. The workflow intentionally builds official Linux release packages on Ubuntu 22.04, whose glibc baseline is 2.35; those CI artifacts have not been produced in this workspace.

Windows packages are unsigned. The macOS packaging script applies an ad-hoc signature, with no Developer ID signature or notarization. SHA256SUMS verifies file bytes and does not establish a publisher identity.

Linux Mint shortcut regression (2026-10-20)

A follow-up run on Ubuntu 24.04 arm64 with a private Xvfb server reproduced Clipboard capture cancelled by reserving Ctrl+Alt+Escape in another X11 client, as Cinnamon does. That arming check passes with the fix. A conflict on plain Escape now retains its actual diagnostic through clipboard capture.

All 12 clipboard/arming checks and 13 end-to-end X11 scenarios passed, including consumed Escape with Caps Lock, Num Lock, and both enabled. The application's C code was generated with pinned V 0.5.2 and compiled with GCC inside the VM. The V core tests also passed with that compiler on macOS arm64. This follow-up did not exercise a physical Linux Mint desktop or rebuild the x86_64 AppImage.

Re-run

./scripts/bootstrap-v.sh
./scripts/build.sh
./scripts/integration-linux.sh
./scripts/build.sh --package

See BUILDING.md for native development/test dependencies and Windows/macOS commands. All desktop integration tests belong on an isolated test desktop; they generate synthetic keystrokes and replace that desktop's test clipboard.