Optimize read-only flag codegen While the @Assume annotations optimize read-only flag queries in optimized targets, this annotation is unused in targets that are built with D8. Instead of using the FeatureFlagsImpl lookup for such flags reads, simply return the read-only flag value directly from the static Flags flag accessor entrypoint. Guard this behind a new build flag. This reduces read-only flag query costs in the framework and other unoptimized targets by ~80%. A follow-up change will further optimize residual FeatureFlagsImpl usage for read-only cases, while a separate set of changes will optimize the read-write cases. Bug: 447177427 Flag: build.RELEASE_ACONFIG_OPTIMIZE_READ_ONLY_JAVA Test: atest AconfigPerfTests:android.os.flagging.AconfigFlagsPerfTest Change-Id: Ica74c2e091134ca6f283c64559f5b32fc1e91302
This is the Makefile-based portion of the Android Build System.
For documentation on how to run a build, see Usage.txt
For a list of behavioral changes useful for Android.mk writers see Changes.md
For an outdated reference on Android.mk files, see build-system.html. Our Android.mk files look similar, but are entirely different from the Android.mk files used by the NDK build system. When searching for documentation elsewhere, ensure that it is for the platform build system -- most are not.
This Makefile-based system is in the process of being replaced with Soong, a new build system written in Go. During the transition, all of these makefiles are read by Kati, and generate a ninja file instead of being executed directly. That's combined with a ninja file read by Soong so that the build graph of the two systems can be combined and run as one.