Repository navigation
MCP 0.5.1 ships alongside: plugdev_tail_logs accepts problems: true.
[added]
plug server logs --problems(and MCPplugdev_tail_logsproblems: true) groups WARN/ERROR entries with their exception and first plugin stack frame and hides PlugDev's own offline-mode/template warnings, so a test loop checks a reload in ~100 bytes of JSON instead of the raw log- The ready banner and JSON
server_readyevent report where startup time went:Server ready in 17.2s (build 2.6s · boot 13.2s) scripts/benchmark-loop.mjsmeasures real boot, safe-reload and restart times on the Paper fixture (--deps,--fresh,BENCH_CLIfor A/B)
[changed]
- New projects and Minecraft versions reuse Paperclip's downloaded and patched server files from
~/.plugdev; a fresh project's first boot dropped from 139–205 s to ~35 s in the loop benchmark - Paper JARs verified within 24 hours start without a Fill API request or a 50 MB re-hash; an unreachable API falls back to the cached JAR
- Test deps resolve in parallel and remember "latest" lookups for 24 hours; a failed lookup uses the last working JAR instead of aborting
plug run - Java 19+ dev servers use a dynamic AppCDS archive in the run dir (~20–30% faster Paper boots in the loop benchmark; opt out with
jvm.args: ["-Xshare:off"]) - Java discovery reads each JDK's
releasefile instead of startingjava -versionper candidate (5.3 s → 0.25 s with five JDKs installed) plug runloads the Ink panel only when.panel/.configopens (CLI startup 2.0 s → 0.9 s)- Restarts build while the old server keeps running: a compile error leaves it up, downtime is only stop + start, and
watch.reloadJava: restartno longer builds twice
[fixed]
- A restart (plugin.yml edit,
watch.reloadJava: restart, Folia, live-config restart) no longer ends the wholeplug runsession and orphans the new server - Reload confirmation compares byte offsets, so earlier "—"/"…" log lines can no longer replay an old reload marker as a false "Reload complete"
- Server console output is decoded as UTF-8 on Windows instead of showing "�" for non-ASCII text
- The published skill sync test ignores CRLF line endings from Windows checkouts