[2.x] perf: Regression of no-op compilation #8360
Description
Originally posted by @mrdziuban in #7658
Steps
My build at work contains 28 projects, which is likely larger than most builds, but a lot smaller than the 84 projects in lila etc.
clean; Test/compile
Test/compile
Test/compile
Test/compile
Test/compile
Test/compile
Problem
With sbt 2.0.0-RC6, the same test showed much worse performance:
[success] elapsed time: 19 s, cache 99%, 961 disk cache hits, 9 onsite tasks
[success] elapsed time: 17 s, cache 99%, 964 disk cache hits, 6 onsite tasks
[success] elapsed time: 14 s, cache 99%, 964 disk cache hits, 6 onsite tasks
[success] elapsed time: 14 s, cache 99%, 964 disk cache hits, 6 onsite tasks
[success] elapsed time: 13 s, cache 99%, 964 disk cache hits, 6 onsite tasks
Note
With sbt 1.11.7:
[success] Total time: 3 s, completed Oct 24, 2025, 1:31:01 PM
[success] Total time: 2 s, completed Oct 24, 2025, 1:31:03 PM
[success] Total time: 1 s, completed Oct 24, 2025, 1:31:04 PM
[success] Total time: 1 s, completed Oct 24, 2025, 1:31:05 PM
[success] Total time: 1 s, completed Oct 24, 2025, 1:31:05 PM