Handle irb monkey-patching in StackProf::Report

[parkr]

Hey!

I generated a StackProf::Report in production for a slow process. I found a method that appeared to be quite expensive and refactored it. I applied the monkey-patch in my irb session and regenerated the stackprof. Unfortunately, when I went to call Report#print_method, I got an error saying that stackprof couldn't open file irb.

Could we add support for reading source from irb? Report#source_display currently calls File.readlines(file), but when the source location of a method is in irb, then it errors.

[parkr]

I took a stab at this today. The irb module took some getting used to, but I found a solution.

module StackProf
  class Report
    def source_file_lines(file)
      if file.eql?("(irb)")
        IRB.CurrentContext.io.instance_variable_get(:@line)
      else
        File.readlines(file)
      end
    end

    def source_display(f, file, lines, range=nil)
      source_file_lines(file).each_with_index do |code, i|
        next unless range.nil? || range.include?(i)
        if lines and lineinfo = lines[i+1]
          total_samples, samples = lineinfo
          if version == 1.0
            samples = total_samples
            f.printf "% 5d % 7s  | % 5d  | %s", samples, "(%2.1f%%)" % (100.0*samples/overall_samples), i+1, code
          elsif samples > 0
            f.printf "% 5d  % 8s / % 5d  % 7s  | % 5d  | %s", total_samples, "(%2.1f%%)" % (100.0*total_samples/overall_samples), samples, "(%2.1f%%)" % (100.0*samples/overall_samples), i+1, code
          else
            f.printf "% 5d  % 8s                   | % 5d  | %s", total_samples, "(%3.1f%%)" % (100.0*total_samples/overall_samples), i+1, code
          end
        else
          if version == 1.0
            f.printf "               | % 5d  | %s", i+1, code
          else
            f.printf "                                  | % 5d  | %s", i+1, code
          end
        end
      end
    rescue SystemCallError
      f.puts "        SOURCE UNAVAILABLE"
    end

  end
end

The IRB.CurrentContext.io returns some form of "input method." These input methods all define a @line instance variable which holds all the IRB history of the current session.

Here's sample output:

GitHub[production] (main):161:0> report.print_method("Object#foobar")
Object#foobar ((irb):1)
  samples:  9823 self (88.0%)  /   9823 total (88.0%)
  callers:
    9823  (  100.0%)  Object#irb_binding
  code:
                                  |     1  |  9823   (88.0%) /  9823  (88.0%)  |     2  | def foobar
                                  |     3  | end

[parkr]

If you add a newline to the end of each line of code in the IRB history, then you get this:

GitHub[production] (main):185:0> report.print_method("Object#foobar")
Object#foobar ((irb):1)
  samples:  9823 self (88.0%)  /   9823 total (88.0%)
  callers:
    9823  (  100.0%)  Object#irb_binding
  code:
                                  |     1  |
 9823   (88.0%) /  9823  (88.0%)  |     2  | def foobar

                                  |     3  | end

All you need to do is change #source_file_lines:

module StackProf
  class Report
    def source_file_lines(file)
      if file.eql?("(irb)")
        IRB.CurrentContext.io.instance_variable_get(:@line).map { |line| line.to_s+"\n" }
      else
        File.readlines(file)
      end
    end
  end
end

[NickLaMuro]

@parkr You probably could do:

IRB.CurrentContext.io.instance_variable_get(:@line).map(&:to_s).join("\n")

If you wanted something a little more terse. Though, does cause one extra loop iteration.

[parkr]

@NickLaMuro That would create just one string for all of IRB history, I think. I’m looking to mimic File.readlines, which returns an array of lines, each with terminating new lines.

[NickLaMuro]

Ah , you are correct. My bad

[parkr]

No problem! Thanks for the optimization help anyway!