Very slow build speeds for hello world

[kojix2]

I investigated where the Crystal compiler is spending time using classic print debugging.

module Crystal
  class Compiler
    def compile(source : Source | Array(Source), output_filename : String) : Result
      source = [source] unless source.is_a?(Array)
      # 1 new_program
      program = new_program(source)

      # 2 parse
      node = parse program, source

      # 3 semantic
      node = program.semantic node, cleanup: !no_cleanup?

      # 4 codegen
      units = codegen program, node, source, output_filename unless @no_codegen

      # 5 cleanup
      # ... omission ...
      Result.new program, node
    end
  end
end

I measured the time spent on each part of the Compiler#compile method.

1. new_program
2. parse
3. semantic
4. codegen
5. cleanup

I also timed the external LLVM functions.

• LibLLVM.run_passes
• LibLLVM.target_machine_emit_to_file

See the patch file for the actual code used.

After compiling and installing the Crystal compiler with changes to display elapsed time, I compiled the unmodified Crystal compiler.

Stage	Time (seconds)
new_program	0.000388207
parse	0.000065000
semantic	12.552620028
codegen	355.245409133
- LibLLVM.run_passes	252.340241198
- LibLLVM.target_machine_emit_to_file	93.280652845
cleanup	0.000013180
total	367.798495548

The results show that the two LLVM functions consume most of the time (93.97%).

This fact might be common knowledge for those who have been using Crystal for a long time or are part of the core team, but for someone like me who started using Crystal later, it is very interesting.