bug#6415: [PATCH] fix edebug instrumentation of dotted pairs in macros

[Steve Yegge]

** Description

There were two separate problems conspiring to create this bug.

First, the edebug spec for `destructuring-bind' is incorrect.

Its definition has three violations of the CL hyperspec for

destructuring lambda lists:

  - it should not support the &environment keyword

  - it should support the &whole keyword

  - it should support dotted forms

It so happens that the `cl-macro-list1' edebug-spec does all three

of these things properly.

The second problem is in edebug.  The unification algorithm has

improper or missing handling for dotted pairs in specs.  I chose

to add the handling to `edebug-match-specs' since it seemed to be

the cleanest place to insert it.

** ChangeLog

2011-09-26  Steve Yegge  <stevey@google.com>

* emacs-lisp/cl-specs.el: Fixed edebug-spec for

`destructuring-bind' to allow dotted pairs in the

destructuring lambda list.  (Bug#6415)

* emacs-lisp/edebug.el: Fixed edebug instrumentation of

dotted pairs in edebug specifications for macros.  (Bug#6415)

** The patch itself

=== modified file 'lisp/emacs-lisp/cl-specs.el'

--- lisp/emacs-lisp/cl-specs.el 2011-02-11 03:54:12 +0000

+++ lisp/emacs-lisp/cl-specs.el 2011-09-26 16:37:19 +0000

@@ -90,7 +90,7 @@

   ((&rest (symbol sexp)) cl-declarations body))

 

 (def-edebug-spec destructuring-bind

-  (&define cl-macro-list def-form cl-declarations def-body))

+  (&define cl-macro-list1 def-form cl-declarations def-body))

 

 ;; Setf

 

=== modified file 'lisp/emacs-lisp/edebug.el'

--- lisp/emacs-lisp/edebug.el 2011-08-21 17:43:31 +0000

+++ lisp/emacs-lisp/edebug.el 2011-09-26 16:44:39 +0000

@@ -1567,8 +1567,28 @@

       (let ((edebug-dotted-spec t));; Containing spec list was dotted.

  (edebug-match-specs cursor (list specs) remainder-handler)))

 

-     ;; Is the form dotted?

-     ((not (listp (edebug-cursor-expressions cursor)));; allow nil

+     ;; Special handling for the tail of a dotted form.

+     ((and

+       ;; Is the cursor on the tail of a dotted form?

+       (not (listp (edebug-cursor-expressions cursor)));; allow nil

+       ;; When matching a dotted form such as (a b . c) against a

+       ;; spec list that looks like

+       ;;     ([&rest ...] [&optional ...]+ . [&or arg nil])

+       ;; ,e.g., the `cl-macro-list1' edebug-spec, then the &rest spec

+       ;; will consume everything up to the dotted tail (`c' in this

+       ;; example).  At that point the spec list will look like so:

+       ;;     ([&optional ...]+ . [&or arg nil])

+       ;; We need to be able to consume zero or more [&optional ...]

+       ;; spec(s) without moving the cursor or signaling an error.

+       ;; The current continuation provides no state that tells us

+       ;; about the upcoming &optional specs, so we use lookahead:

+

+       ;; Recurse normally if we're about to process an optional spec.

+       (not (eq (car specs) '&optional))

+       ;; Recurse normally if the spec is a dotted list.

+       (not (and (listp specs)

+                 (not (listp (cdr (last specs)))))))

+      ;; Otherwise we need to be on the tail of a dotted spec.

       (if (not edebug-dotted-spec)

   (edebug-no-match cursor "Dotted spec required."))

       ;; Cancel dotted spec and dotted form.