repl/debugger integration

[Andy Wingo]

Hi all,

A quick note to mention that I've pushed some patches that integrate the
debugger and the repl. Now if you have an error, you will be dropped
into a recursive repl:

    scheme@(guile-user)> (eval '(let a 10) (current-module))
    Throw to key `syntax-error':
    ERROR: In procedure macroexpand:
    ERROR: let: bad let in (let a 10)
    Entering a new prompt. Type `,bt' for a backtrace or `,q' to return to the 
old prompt.
    scheme@(guile-user) [1]> 

You see the [1] in the prompt, showing that you aren't at the bottom
level any more.

    scheme@(guile-user) [1]> ,help debug
    Debug Commands [abbrev]:

     ,backtrace [COUNT] [#:width W] [#:full? F]
                                 [,bt] - Print a backtrace.
     ,up [COUNT]                       - Select a calling stack frame.
     ,down [COUNT]                     - Select a called stack frame.
     ,frame [IDX]                [,fr] - Show a frame.
     ,procedure                [,proc] - Print the procedure for the selected 
frame.
     ,locals                           - Show local variables.

The commands that were part of the debugger are now meta-commands in the
repl:

    scheme@(guile-user) [1]> ,bt
    In ice-9/boot-9.scm:
     1056:  5 (#<procedure 3030ab0 ()>)
       61:  4 (call-with-prompt prompt0 #<procedure 2ee45e0 at syst...> #)
    In unknown file:
            3 (eval (let a 10) #<directory (guile-user) 2647510>)
    In ice-9/eval.scm:
      457:  2 (#<procedure 26537e0 at ice-9/eval.scm:451:4 (exp)> (# ...))
    In ice-9/psyntax.scm:
     1223:  1 (chi-top (let a 10) () ((top)) e (eval) (hygiene #))
    In unknown file:
            0 (scm-error syntax-error macroexpand "~a: ~a in ~a" (...) #f)

    scheme@(guile-user) [1]> ,frame
            0 (scm-error syntax-error macroexpand "~a: ~a in ~a" (...) #f)

    scheme@(guile-user) [1]> ,locals
      No local variables.

    scheme@(guile-user) [1]> ,up
    In ice-9/psyntax.scm:
     1223:  1 (chi-top (let a 10) () ((top)) e (eval) (hygiene #))

    scheme@(guile-user) [1]> ,locals
      Local variables:
      $1 = e = (let a 10)
      $2 = r = ()
      $3 = w = ((top))
      $4 = m = e
      $5 = esew = (eval)
      $6 = mod = (hygiene guile-user)
      $7 = type = core-form
      $8 = value = #<procedure 2614400 at ice-9/psyntax.scm:2078:19 (e r...>
      $9 = e = (let a 10)
      $10 = w = ((top))
      $11 = s = #f
      $12 = mod = (hygiene guile-user)
      $13 = x = #<undefined>

The module is whatever was current when the repl began. I don't think
the behavior of the "repl" command in the old debugger was very useful,
you often didn't have the bindings you expect -- though of course you
can switch modules while in the recursive repl. (Hm, should save the
module excursion there, I guess).

If you run debugging commands at the base repl you'll just get an error
message:

    scheme@(guile-user) [1]> ,q
    ERROR: In procedure macroexpand:
    ERROR: let: bad let in (let a 10)
    scheme@(guile-user)> ,bt
    Nothing to debug.

In the future we should be able to customize this behavior using repl
options (for example, a user should be able to say "always give me a
backtrace, but don't bother dropping me into a recursive
repl"). Feedback appreciated regarding what the default option should
be, of course. Or any other feedback folk might have is welcome.

Regards,

Andy
-- 
http://wingolog.org/