Build speed at the final commit: v0.27.8 against v0.27.9

The build-speed benchmark taken again on the last commit of the v0.27.9 work, after the runtime fixes (404 for an unknown path, the brotli boot loader, system tools by absolute path) and the falsifier-proof commits had landed. It replaces ../build-speed-final-20261015/ (compiler 94c57af01) as the source of the v0.27.9 release numbers.

The benchmark (after.json, bench-head.log)

scripts/bench-build.sh --reps 3 --sizes "50 100 200" --cases "app skyshop" --targets "web:app web". The arguments are the same as in ../build-speed-final-20261015/. The compiler is sky-out/sky of this tree, built by scripts/build.sh immediately before the run. Scenarios, units and first are as described there: wall and CPU in seconds (CPU is user + sys of sky and all its children), and peak RSS of the largest single process in MB. Each cell is the median of 3, except first, which is rep 1 of prime.

--target web:app

Nscenariowall beforewall afterCPU beforeCPU afterpeak RSS beforepeak RSS after
50first35.636.8857.0310.45610589
50warm3.131.393.933.07145127
50newdir33.743.9356.586.42664588
50edit28.9514.8538.3832.26604585
50cold53.4833.68161.12141.56756775
100first69.564.47157.6010.721027611
100warm17.201.3514.073.11275139
100newdir50.593.1179.115.81985586
100edit46.3512.9560.8432.91979587
100cold70.3828.71178.73138.62987767
200first139.3513.10189.0744.202781621
200warm64.941.6464.873.24773170
200newdir129.703.64183.236.752976594
200edit123.4512.88163.8544.382989614
200cold153.5231.80277.86136.533015805

--target web (Sky.Live)

Nscenariowall beforewall afterCPU beforeCPU afterpeak RSS beforepeak RSS after
50first7.322.9120.323.86655611
50warm0.620.511.141.068071
50newdir6.311.4420.392.46666601
50edit2.501.876.173.82594582
50cold20.8020.42101.72100.51781746
100first7.301.9120.773.88666581
100warm0.720.671.171.0910780
100newdir6.541.5020.772.28698590
100edit2.662.026.514.04598581
100cold21.3119.68102.9895.42789779
200first7.793.1422.064.87691587
200warm1.070.701.541.00186105
200newdir6.981.6522.062.63694591
200edit3.202.647.775.26603588
200cold20.2230.78102.89107.09764746

examples/13-skyshop (Go FFI, 76k symbols; medians of 3)

scenariowall beforewall afterCPU beforeCPU afterpeak RSS beforepeak RSS after
prime31.937.8786.386.1140411268
warm4.143.213.293.32313311
newdir26.724.7280.755.8439731270
edit10.306.8718.7413.8512971266
cold52.5149.88227.54214.5139724081

The genwhole shape (genwhole200-head.log, genwhole200-head.time)

The same recipe as ../build-speed-final-20261015/: genwhole.py examples/62-app-notes <dir> 200 (the script is committed there), the header text changed to a per-run token so the app packages are new to the Go cache, then one sky build --target web:app src/Main.sky with SKY_TIMINGS=1. One run. Scratch paths in the log are replaced by <scratch>.

genwhole N = 200, first buildv0.27.8v0.27.9 (9a665f22e)
wall1575.7 s29.7 s
CPU (user + sys)2171.7 s117.5 s
peak RSS (largest process)4,775 MB1,100 MB

The v0.27.8 column is the committed ../build-speed-final-20261015/genwhole200-before.time. v0.27.8 was not run again. After, the two legs run together (28.2 s for both): the backend leg takes 16.4 s, the frontend leg 8.2 s, and the in-process brotli of the new wasm 19.8 s beside the backend leg.

Reading it