Potential build performance regression in the latest versions (likely 2.26)

[chao2zhang]

My colleague monitoring our build pipeline recently notified me about a potential build-time regression of 30secs at 50percentile on CI. This was likely introduced by my change of dagger version bump, from 2.25.4 to 2.27.

I did some profiling using dagger versions 2.25.4, 2.26 and 2.27 and my results can be found here: https://docs.google.com/spreadsheets/d/1tLqwIUzluwqynKbtG3SCTB0Qr8Ne23-Z5wpSeM1EtUI/edit?usp=sharing
On average, for :app:assembleDebug,

• Dagger 2.25.4: 297.8s on avg
• Dagger 2.26: 307.2s on avg
• Dagger 2.27: 311.1s on avg

While I have yet read through every commit through the recent changes in dagger, I would like to call out a few characteristics of our codebase and build environment:

• I ran my benchmark on MacOS 10.15.4 with 32GB memory
• Our app heavily uses dagger with java annotation processor (not kapt) and has 80k+ lines in DaggerApplicationComponent.java
• Our app code has mixed Java and Kotlin code

From the benchmark result, that there is a performance regression introduced in Dagger 2.26, would you mind taking a look? I can definitely share more information if interested. Thanks in advance!