How to Use

For automated use (e.g. in CI), use main.py. See its help with main.py --help. Note that metrics collection relies on printproto and jq tools being on $PATH.

The most basic invocation, e.g. ./incremental_build.py libc, is logically equivalent to

1. running m --skip-soong-tests libc and then
2. parsing $OUTDIR/soong_metrics and $OUTDIR/bp2build_metrics.pb files
3. Adding timing-related metrics from those files into out/timing_logs/summary.csv

There are a number of CUJs set up in cuj_catalog.py and they are run sequentially, such that each row in summary.csv are the timings of various “events” during an incremental build.

You may also add rows to summary.csv after a manual run, using perf_metrics.py script. This is particularly useful when you don't want to modify cuj_catalog.py for one-off tests.

Currently:

1. run a build (conceptually, m droid)
2. printproto to parse metrics related pb files
3. use jq to filter data 4 collate data into a csv file
4. goto 1 until various CUJs are exhausted

For CI, we should:

1. run a build with some identifiable tag (not sure what mechanisms are available)
2. goto 1 until various CUJs are exhausted
3. rely on plx to collate data from all builds and provide a filtering mechanism based on that tag from step 1