Debuggers in Lispy Languages

This talk is about a unique way in which debuggers can be written in language which are lisp like.
With CIDER debug as an example, this talk takes you through how this approach works under the hood.
This talk was presented at Functional Conf 2019.

Transcript

1. Debuggers in Lisps
   Suvrat Apte
   (with Clojure)
   (Helpshift Inc)
   

   View Slide

2. Outline
   • Why we need debuggers
   • Demo: Clojure debugging
   • How GDB works
   • REPL middleware
   • How cider-debug works
   • Limitations
   • Questions
   

   View Slide

3. Debuggers
   • Examine state of programs as they run
   • Halt execution
   • Step through execution
   • Modify state at runtime
   • Change flow of execution
   

   View Slide

4. cider-nrepl
   Demo
   Clojure debugging
   

   View Slide

5. GDB - GNU Debugger
   • Debugger runs as a separate process
   • Spawns the target program as a child
   • Examines state with process tracing calls
   • Implementing breakpoints is a trick!
   • Puts invalid instructions on breakpoints
   • Processor traps OS
   • OS signals the process which caused it
   • It passes the signal to its parent (GDB)
   • GDB checks IP and breaks
   • Some processors support breakpoints
   GDB
   Target
   OS
   Processor
   

   View Slide

6. GDB - GNU Debugger
   • Dependence on OS for process tracing
   • Dependence on OS for signals
   • Dependence on processors for interrupts
   GDB
   Target
   OS
   Processor
   

   View Slide

7. Lisps
   • Lisp has minimal syntax
   • Code is data
   • Code itself is a valid AST
   • Debuggers can use code as data
   (map sq-root (filter even? numbers))
   map
   sq-root filter
   numbers
   even?
   

   View Slide

8. Clojure debugger
   f (x) = y
   Debugger (code) = debuggable code
   

   View Slide

9. What is a REPL?
   Read-Eval-Print-Loop
   (defn repl
   []
   (let [input (read)
   result (eval input)]
   (println result)
   (recur)))
   

   View Slide

10. Read Eval Print Loop (REPL)
    REPL Prompt:
    "=> (+ 1 2)
    "=> 3
    "=>
    REPL client REPL server
    Operation: “eval”
    Code: (+ 1 2)
    

    View Slide

11. Read Eval Print Loop (REPL)
    REPL client REPL server
    Operation: “eval”
    Result: 3
    REPL Prompt:
    "=> (+ 1 2)
    "=> 3
    "=>
    

    View Slide

12. nREPL Middleware
    nREPL
    CIDER
    nREPL
    CIDER
    Middleware
    

    View Slide

13. cider-nrepl
    •CIDER uses cider-nrepl middleware
    •A collection of nREPL middleware
    • complete
    • info
    • debug
    • More. . .
    

    View Slide

14. cider-nrepl
    Demo
    Client-server messaging
    

    View Slide

15. Form based coords
    Coor: (3 1 1 1)
    (defn math-operations
    [n1 n2]
    (let [addition (+ n1 n2)
    subtraction (- n1 n2)
    exp (power n1 n2)]
    {:addition addition
    :subtraction subtraction
    :exp exp}))
    

    View Slide

16. Coor: (3 1 1 1)
    (defn math-operations
    [n1 n2]
    (let [addition (+ n1 n2)
    subtraction (- n1 n2)
    exp (power n1 n2)]
    {:addition addition
    :subtraction subtraction
    :exp exp}))
    0 1
    2
    3
    Form based coords
    

    View Slide

17. Coor: (3 1 1 1)
    (defn math-operations
    [n1 n2]
    (let [addition (+ n1 n2)
    subtraction (- n1 n2)
    exp (power n1 n2)]
    {:addition addition
    :subtraction subtraction
    :exp exp}))
    0
    1
    2
    Form based coords
    

    View Slide

18. Coor: (3 1 1 1)
    (defn math-operations
    [n1 n2]
    (let [addition (+ n1 n2)
    subtraction (- n1 n2)
    exp (power n1 n2)]
    {:addition addition
    :subtraction subtraction
    :exp exp}))
    0 1
    2 3
    4 5
    Form based coords
    

    View Slide

19. Coor: (3 1 1 1)
    (defn math-operations
    [n1 n2]
    (let [addition (+ n1 n2)
    subtraction (- n1 n2)
    exp (power n1 n2)]
    {:addition addition
    :subtraction subtraction
    :exp exp}))
    0 1 2
    Form based coords
    

    View Slide

20. That’s why you should use Emacs!
    

    View Slide

21. How does it work?
    1. Walk through code
    1. Mark forms/symbols with coordinates
    2. Expand macros
    3. Walk through code
    1. Wrap forms/symbols with pre-breakpoint macro
    4. Pre-breakpoint macro expansion
    1. Put a breakpoint if necessary
    When a function is marked for debugging
    

    View Slide

22. How does it work?
    1. Necessary information is sent to client
    1. Coordinates
    2. Return value at current position
    3. Values of local vars
    4. Possible actions
    2. Waits for response from client. . .
    When a breakpoint is hit
    

    View Slide

23. How does it work?
    1. Walk through code
    1. Mark forms/symbols with coordinates
    2. Expand macros
    3. Walk through code
    1. Wrap forms/symbols with pre-breakpoint macro
    4. Pre-breakpoint macro expansion
    1. Put a breakpoint if necessary
    When a function is marked for debugging
    

    View Slide

24. How does it work?
    • Simple eval
    • Reader macro: #dbg
    • cider.nrepl.middleware.util.instrument/walk-indexed
    • Assigns coordinates to forms recursively
    

    View Slide

25. How does it work?
    1. Walk through code
    1. Mark forms/symbols with coordinates
    2. Expand macros
    3. Walk through code
    1. Wrap forms/symbols with pre-breakpoint macro
    4. Pre-breakpoint macro expansion
    1. Put a breakpoint if necessary
    When a function is marked for debugging
    

    View Slide

26. How does it work?
    1. Walk through code
    1. Mark forms/symbols with coordinates
    2. Expand macros
    3. Walk through code
    1. Wrap forms/symbols with pre-breakpoint macro
    4. Pre-breakpoint macro expansion
    1. Put a breakpoint if necessary
    When a function is marked for debugging
    

    View Slide

27. How does it work?
    1. Walk through code
    1. Mark forms/symbols with coordinates
    2. Expand macros
    3. Walk through code
    1. Wrap forms/symbols with pre-breakpoint macro
    4. Pre-breakpoint macro expansion
    1. Put a breakpoint if necessary
    When a function is marked for debugging
    

    View Slide

28. How does it work?
    1. Necessary information is sent to client
    1. Coordinates
    2. Return value at current position
    3. Values of local vars
    4. Possible actions
    2. Waits for response from client. . .
    When a breakpoint is hit
    

    View Slide

29. How does it work?
    1. Necessary information is sent to client
    1. Coordinates
    2. Return value at current position
    3. Values of local vars
    4. Possible actions
    2. Waits for response from client. . .
    When a breakpoint is hit
    

    View Slide

30. Clojure debugger
    f (x) = y
    Debugger (code) = debuggable code
    

    View Slide

31. Clojure debugger
    f (x) = y
    Debugger (code) = debuggable code
    Clojure Clojure
    

    View Slide

32. GDB - GNU Debugger
    f (p, q, r) = y
    Debugger (code, os, processor) = debuggable code
    C Assembly
    

    View Slide

33. Limitations
    • Cannot instrument bytecode
    • Cannot go back in time
    • Cannot replay
    

    View Slide

34. cider-debug
    • Debug.clj: 655 lines
    • Instrument.clj: 372 lines
    • The code is very readable!
    

    View Slide

35. cider-debug
    • Bozhidar Batsov
    • Artur Malabara
    • Vitalie Spinu
    • Chris Perkins
    • Neil Okamoto
    • Michael Griffiths
    • Lars Anderson
    • Dominic Monroe
    Thank you!
    

    View Slide

36. Lisp is similar to Chess.
    It will take only a day to learn the rules.
    But years to master !
    - Harold Abelson
    

    View Slide