Why kati is faster than GNU make?

A bunch of optimization techniques:

• There are some .mk files which are included many times. For example, build/core/base_rules.mk is included 10k times. Kati parses such files only once (2x slower if disabled).
• Kati has a GNU find emulator, which traverses the directory tree at first then returns cached results. $(shell find path/to/java/code -name '*.java') is super fast (1.6x slower if disabled).
• Kati uses interned strings everywhere for faster variable/target lookups.