SkillPedia Skills Encyclopedia
中文 EN

技能 2026-09-23 05:02 UTC

godot-skillv0.5.0

版本【v0.5.0】由提交人 haxqer 提交,主要更新类型为:问题缺陷修复与稳定性提升。提交说明:ci: Explain an unshared checkout and run the framebuffer gate locally too。解决了已知运行时异常与边界错误,提升系统稳定性。

haxqer 037c03e haxqer/godot-skill

Commit提交信息原文

ci: Explain an unshared checkout and run the framebuffer gate locally too

Verifying main from fresh clones (macOS, and Linux in the CI image) found two
ways scripts/run_tests_linux.sh could mislead, both in the same direction as
the failures this repo exists to catch: quietly.

- A checkout on a path the Docker VM does not share bind-mounts as an EMPTY
  directory with no error (colima shares only the home directory by default),
  so the runner copied nothing and died with a bare "run_tests.sh: not found",
  exit 127 — indistinguishable from a broken checkout. It now probes the mount
  in the same image before anything depends on it and says what to do.
- The runner claimed to run "exactly the way CI runs it" but skipped CI's
  "Confirm the framebuffer renders" step, so an image with broken GL would
  have reported a green suite in which every real-render test skipped itself.
  The gate is now one script, scripts/confirm_framebuffer.sh, that both the
  workflow and the runner invoke under the same xvfb-run line, before the full
  suite and before a named subset alike. Extracting it also fixed the inline
  step itself: under `set -euo pipefail` the scenario runner's non-zero exit
  ended the step before its own "no usable frame" message could print.

Proven both ways: the gate passes on a Mac (real display) and in the image
under Xvfb (64x64, 102 colours), and exits 1 with the message in the image
without Xvfb. Fresh-clone results for the tree this builds on: macOS 35/35,
Linux arm64 35/35; x86_64 was only reachable through QEMU user emulation here,
where the emulator itself segfaults, so native x86_64 evidence remains the
GitHub runner's job.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>