You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Measure and reduce CLI startup latency and distribution size #25
User reports slow startup and a large EXE. Current publishing is self-contained single-file with native extraction.
Acceptance
Measure cold extraction and warm-cache startup, --version, dry-run and usable Home on Windows/Linux/macOS; record machine/build and repeated measurements.
Separate runtime/JIT, extraction, resources, settings/discovery and Home IPC costs; record executable/archive/cache sizes.
Remove avoidable work and publish before/after results plus a regression budget; preserve standalone installation, credentials, offline docs/licenses and scripting.
Implementation in #59. Nine paired local Linux launches: warm version 133.70 → 29.15 ms, Home 464.61 → 187.99 ms; state preview remains within noise. Remove the missing-host connection wait and use a minimal bare-version bootstrap. The 76.6 MB executable is unchanged in size. Compression was measured (40.6 MB executable) and rejected for now because startup worsened and gzip size slightly increased. No native extraction files were observed for these commands on this Linux .NET 10 build; this is not evidence about Windows/macOS or cold OS startup.
Reproducible benchmark, raw reports, limits and regression budget are in docs/performance/2026-10-26-startup.md. Native before/after benchmark workflow: https://github.com/Anheledir/SidePulse.NET/actions/runs/40317680557 . Keep #25 open for actual Windows console/ConPTY Home measurement and the remaining attribution/acceptance evidence. Size/AOT/trimming follow-up remains #26.
The executable size is effectively unchanged. Compression was measured and remains disabled because startup worsened and gzip download size did not improve. Keep #25 open for actual Windows Home/console measurement and remaining attribution/acceptance. #26 owns the AOT/trimming size decision.
Activity
Anheledir commentedon Oct 26, 2026
Implementation in #59. Nine paired local Linux launches: warm version 133.70 → 29.15 ms, Home 464.61 → 187.99 ms; state preview remains within noise. Remove the missing-host connection wait and use a minimal bare-version bootstrap. The 76.6 MB executable is unchanged in size. Compression was measured (40.6 MB executable) and rejected for now because startup worsened and gzip size slightly increased. No native extraction files were observed for these commands on this Linux .NET 10 build; this is not evidence about Windows/macOS or cold OS startup.
Reproducible benchmark, raw reports, limits and regression budget are in docs/performance/2026-10-26-startup.md. Native before/after benchmark workflow: https://github.com/Anheledir/SidePulse.NET/actions/runs/40317680557 . Keep #25 open for actual Windows console/ConPTY Home measurement and the remaining attribution/acceptance evidence. Size/AOT/trimming follow-up remains #26.
Anheledir commentedon Oct 26, 2026
Measured startup work merged in #59. Final benchmark run: https://github.com/Anheledir/SidePulse.NET/actions/runs/40318063431 . Nine paired launches per mode; these are warm bundle-cache medians, not cold OS/antivirus or physical-device timings.
The executable size is effectively unchanged. Compression was measured and remains disabled because startup worsened and gzip download size did not improve. Keep #25 open for actual Windows Home/console measurement and remaining attribution/acceptance. #26 owns the AOT/trimming size decision.
Reproduction, regression budget and local raw reports: https://github.com/Anheledir/SidePulse.NET/blob/main/docs/performance/2026-10-26-startup.md . Native functional tests include install/PATH/update/rollback and full CLI startup probes.