Skip compilation when there's no code changes

[amomchilov]

Hey there,

It looks like the plugin does a full rebuild on every command, even when no code has changed. E.g. on the sample project, running swift package benchmark baseline list takes ~30 seconds, even on a repeated invocation.

$ time swift package benchmark baseline list
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.66s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.12s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.12s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.11s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.12s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.12s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkBoilerplateGenerator' complete! (0.12s)
Building for debugging...
[1/1] Write swift-version--58304C5D6DBC2206.txt
Build of product 'BenchmarkTool' complete! (1.83s)
Building benchmark targets in release mode for benchmark run...
Building Samples
Building Miscellaneous
Building Minimal
Building MemoryTwo
Building MemoryOne
Building Foundation-Benchmark
Building External-Benchmarks
Build complete!
swift package benchmark baseline list  45.83s user 4.07s system 160% cpu 31.024 total

I'd be happy to contribute to fixing this, if you could point me in the right directiont ot start.

[MahdiBM]

I wonder if a --skip-build flag can be of use, it can be directly passed to swift build by the benchmarking package.
I think It'll be valuable ... we can skip some CI time when we know that we have build the package in the previous step and there are no changes.

[hassila]

Yeah, this is frustrating - I believe this is a possible workaround as pointed out by @MahdiBM :

swiftlang/swift-package-manager#7210 (comment)

(that case is about this specific problem, there should be no rebuild really - if that SPM case is fixed, it should "just work" AFAICT)

Having a skip flag is problematic, as the discovery of the benchmarks are done using the SPM API (we get the path to the executables from there).

[MahdiBM]

for the record, explicitly using -c release makes the compiler not do double work for the most part and resolves this issue to a good extent.