1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
|
<html>
<head><title>DEFUN.html -- ACL2 Version 3.1</title></head>
<body text=#000000 bgcolor="#FFFFFF">
<h2>DEFUN</h2>define a function symbol
<pre>Major Section: <a href="EVENTS.html">EVENTS</a>
</pre><p>
<pre>
Examples:
(defun app (x y)
(if (consp x)
(cons (car x) (app (cdr x) y))
y))<p>
(defun fact (n)
(declare (xargs :guard (and (integerp n)
(>= n 0))))
(if (zp n)
1
(* n (fact (1- n)))))
<p>
General Form:
(defun fn (var1 ... varn) doc-string dcl ... dcl body),
</pre>
where <code>fn</code> is the symbol you wish to define and is a new symbolic
name (see <a href="NAME.html">name</a>), <code>(var1 ... varn)</code> is its list of formal
parameters (see <a href="NAME.html">name</a>), and <code>body</code> is its body. The definitional
axiom is logically admissible provided certain restrictions are met.
These are sketched below.<p>
Note that ACL2 does not support the use of <code>lambda-list</code> keywords
(such as <code>&optional</code>) in the formals list of functions. We do support
some such keywords in macros and often you can achieve the desired
syntax by defining a macro in addition to the general version of
your function. See <a href="DEFMACRO.html">defmacro</a>. The <a href="DOCUMENTATION.html">documentation</a> string,
<code><a href="DOC-STRING.html">doc-string</a></code>, is optional; for a description of its form,
see <a href="DOC-STRING.html">doc-string</a>.<p>
The <em>declarations</em> (see <a href="DECLARE.html">declare</a>), <code>dcl</code>, are also optional.
If more than one <code>dcl</code> form appears, they are effectively grouped together
as one. Perhaps the most commonly used ACL2 specific declaration is
of the form <code>(declare (xargs :guard g :measure m))</code>. This declaration
in the <code>defun</code> of some function <code>fn</code> has the effect of making the
``<a href="GUARD.html">guard</a>'' for <code>fn</code> be the term <code>g</code> and the ``measure'' be the term <code>m</code>.
The notion of ``measure'' is crucial to ACL2's definitional
principle. The notion of ``guard'' is not, and is discussed elsewhere;
see <a href="VERIFY-GUARDS.html">verify-guards</a> and see <a href="SET-VERIFY-GUARDS-EAGERNESS.html">set-verify-guards-eagerness</a>. Note that the
<code>:measure</code> is ignored in <code>:</code><code><a href="PROGRAM.html">program</a></code> mode; see <a href="DEFUN-MODE.html">defun-mode</a>.<p>
We now briefly discuss the ACL2 definitional principle, using the
following definition form which is offered as a more or less generic
example.
<pre>
(defun fn (x y)
(declare (xargs :guard (g x y)
:measure (m x y)))
(if (test x y)
(stop x y)
(step (fn (d x) y))))
</pre>
Note that in our generic example, <code>fn</code> has just two arguments, <code>x</code> and
<code>y</code>, the <a href="GUARD.html">guard</a> and measure terms involve both of them, and the body is
a simple case split on <code>(test x y)</code> leading to a ``non-recursive''
branch, <code>(stop x y)</code>, and a ``recursive'' branch. In the recursive
branch, <code>fn</code> is called after ``decrementing'' <code>x</code> to <code>(d x)</code> and some step
function is applied to the result. Of course, this generic example
is quite specific in form but is intended to illustrate the more
general case.<p>
Provided this definition is admissible under the logic, as outlined
below, it adds the following axiom to the logic.
<pre>
Defining Axiom:
(fn x y)
=
(if (test x y)
(stop x y)
(step (fn (d x) y)))
</pre>
Note that the <a href="GUARD.html">guard</a> of <code>fn</code> has no bearing on this logical axiom.<p>
This defining axiom is actually implemented in the ACL2 system by a
<code>:</code><code><a href="DEFINITION.html">definition</a></code> rule, namely
<pre>
(equal (fn x y)
(if (test a b)
(stop a b)
(step (fn (d a) b)))).
</pre>
See <a href="DEFINITION.html">definition</a> for a discussion of how definition rules are
applied. Roughly speaking, the rule causes certain instances of
<code>(fn x y)</code> to be replaced by the corresponding instances of the
body above. This is called ``opening up'' <code>(fn x y)</code>. The
instances of <code>(fn x y)</code> opened are chosen primarily by heuristics
which determine that the recursive calls of <code>fn</code> in the opened body
(after simplification) are more desirable than the unopened call of
<code>fn</code>.<p>
This discussion has assumed that the definition of <code>fn</code> was
admissible. Exactly what does that mean? First, <code>fn</code> must be a
previously unaxiomatized function symbol (however,
see <a href="LD-REDEFINITION-ACTION.html">ld-redefinition-action</a>). Second, the formal parameters
must be distinct variable names. Third, the <a href="GUARD.html">guard</a>, measure, and
body should all be terms and should mention no free variables except
the formal parameters. Thus, for example, body may not contain
references to ``global'' or ``special'' variables; ACL2 constants or
additional formals should be used instead.<p>
The final conditions on admissibility concern the termination of the
recursion. Roughly put, all applications of <code>fn</code> must terminate.
In particular, there must exist a binary relation, <code>rel</code>, and some
unary predicate <code>mp</code> such that <code>rel</code> is well-founded on objects
satisfying <code>mp</code>, the measure term <code>m</code> must always produce
something satisfying <code>mp</code>, and the measure term must decrease
according to <code>rel</code> in each recursive call, under the hypothesis
that all the tests governing the call are satisfied. By the meaning
of well-foundedness, we know there are no infinitely descending
chains of successively <code>rel</code>-smaller <code>mp</code>-objects. Thus, the
recursion must terminate.<p>
The only primitive well-founded relation in ACL2 is <code><a href="O_lt_.html">o<</a></code>
(see <a href="O_lt_.html">o<</a>), which is known to be well-founded on the
<code><a href="O-P.html">o-p</a></code>s (see <a href="O-P.html">o-p</a>). For the proof of
well-foundedness, see <a href="PROOF-OF-WELL-FOUNDEDNESS.html">proof-of-well-foundedness</a>. However it is
possible to add new well-founded relations. For details,
see <a href="WELL-FOUNDED-RELATION.html">well-founded-relation</a>. We discuss later how to specify
which well-founded relation is selected by <code>defun</code> and in the
present discussion we assume, without loss of generality, that it is
<code><a href="O_lt_.html">o<</a></code> on the <code><a href="O-P.html">o-p</a></code>s.<p>
For example, for our generic definition of <code>fn</code> above, with measure
term <code>(m x y)</code>, two theorems must be proved. The first establishes
that <code>m</code> produces an ordinal:
<pre>
(o-p (m x y)).
</pre>
The second shows that <code>m</code> decreases in the (only) recursive call of
<code>fn</code>:
<pre>
(implies (not (test x y))
(o< (m (d x) y) (m x y))).
</pre>
Observe that in the latter formula we must show that the
``<code>m</code>-size'' of <code>(d x)</code> and <code>y</code> is ``smaller than'' the <code>m</code>-size of <code>x</code> and <code>y</code>,
provided the test, <code>(test x y)</code>, in the body fails, thus leading to
the recursive call <code>(fn (d x) y)</code>.<p>
See <a href="O_lt_.html">o<</a> for a discussion of this notion of ``smaller
than.'' It should be noted that the most commonly used ordinals are
the natural numbers and that on natural numbers, <code><a href="O_lt_.html">o<</a></code> is just
the familiar ``less than'' relation (<code><a href="_lt_.html"><</a></code>). Thus, it is very common
to use a measure <code>m</code> that returns a nonnegative integer, for then
<code>(o-p (m x y))</code> becomes a simple conjecture about the type of
<code>m</code> and the second formula above becomes a conjecture about the less-than
relationship of nonnegative integer arithmetic.<p>
The most commonly used measure function is <code><a href="ACL2-COUNT.html">acl2-count</a></code>, which
computes a nonnegative integer size for all ACL2 objects.
See <a href="ACL2-COUNT.html">acl2-count</a>.<p>
Probably the most common recursive scheme in Lisp <a href="PROGRAMMING.html">programming</a> is
when some formal is supposed to be a list and in the recursive call
it is replaced by its <code><a href="CDR.html">cdr</a></code>. For example, <code>(test x y)</code> might be simply
<code>(atom x)</code> and <code>(d x)</code> might be <code>(cdr x)</code>. In that case, <code>(acl2-count x)</code>
is a suitable measure because the <code><a href="ACL2-COUNT.html">acl2-count</a></code> of a <code><a href="CONS.html">cons</a></code> is strictly
larger than the <code><a href="ACL2-COUNT.html">acl2-count</a></code>s of its <code><a href="CAR.html">car</a></code> and <code><a href="CDR.html">cdr</a></code>. Thus, ``recursion
by <code><a href="CAR.html">car</a></code>'' and ``recursion by <code><a href="CDR.html">cdr</a></code>'' are trivially admitted if
<code><a href="ACL2-COUNT.html">acl2-count</a></code> is used as the measure and the definition protects every
recursive call by a test insuring that the decremented argument is a
<code><a href="CONSP.html">consp</a></code>. Similarly, ``recursion by <code><a href="1-.html">1-</a></code>'' in which a positive integer
formal is decremented by one in recursion, is also trivially
admissible. See <a href="BUILT-IN-CLAUSES.html">built-in-clauses</a> to extend the class of
trivially admissible recursive schemes.<p>
We now turn to the question of which well-founded relation <code>defun</code>
uses. It should first be observed that <code>defun</code> must actually select
both a relation (e.g., <code><a href="O_lt_.html">o<</a></code>) and a domain predicate (e.g.,
<code><a href="O-P.html">o-p</a></code>) on which that relation is known to be well-founded.
But, as noted elsewhere (see <a href="WELL-FOUNDED-RELATION.html">well-founded-relation</a>), every
known well-founded relation has a unique domain predicate associated
with it and so it suffices to identify simply the relation here.<p>
The <code><a href="XARGS.html">xargs</a></code> field of a <code><a href="DECLARE.html">declare</a></code> permits the explicit specification of
any known well-founded relation with the keyword
<code>:</code><code><a href="WELL-FOUNDED-RELATION.html">well-founded-relation</a></code>. An example is given below. If the <code><a href="XARGS.html">xargs</a></code>
for a <code>defun</code> specifies a well-founded relation, that relation and its
associated domain predicate are used in generating the termination
conditions for the definition.<p>
If no <code>:</code><code><a href="WELL-FOUNDED-RELATION.html">well-founded-relation</a></code> is specified, <code>defun</code> uses the
<code>:</code><code><a href="WELL-FOUNDED-RELATION.html">well-founded-relation</a></code> specified in the <code><a href="ACL2-DEFAULTS-TABLE.html">acl2-defaults-table</a></code>.
See <a href="SET-WELL-FOUNDED-RELATION.html">set-well-founded-relation</a> to see how to set the default
well-founded relation (and, implicitly, its domain predicate). The
initial default well-founded relation is <code><a href="O_lt_.html">o<</a></code> (with domain
predicate <code><a href="O-P.html">o-p</a></code>).<p>
This completes the brief sketch of the ACL2 definitional principle.<p>
On very rare occasions ACL2 will seem to "hang" when processing a
definition, especially if there are many subexpressions of the body
whose function symbol is <code><a href="IF.html">if</a></code> (or which macroexpand to such an
expression). In those cases you may wish to supply the following to
<code><a href="XARGS.html">xargs</a></code>: <code>:normalize nil</code>. This is an advanced feature that turns
off ACL2's usual propagation upward of <code>if</code> tests.<p>
The following example illustrates all of the available declarations,
but is completely nonsensical. For documentation, see <a href="XARGS.html">xargs</a> and
see <a href="HINTS.html">hints</a>.
<pre>
(defun example (x y z a b c i j)
(declare (ignore a b c)
(type integer i j)
(xargs :guard (symbolp x)
:measure (- i j)
:well-founded-relation my-wfr
:hints (("Goal"
:do-not-induct t
:do-not '(generalize fertilize)
:expand ((assoc x a) (member y z))
:restrict ((<-trans ((x x) (y (foo x)))))
:hands-off (length binary-append)
:in-theory (set-difference-theories
(current-theory :here)
'(assoc))
:induct (and (nth n a) (nth n b))
:non-executable t
:use ((:instance assoc-of-append
(x a) (y b) (z c))
(:functional-instance
(:instance p-f (x a) (y b))
(p consp)
(f assoc)))))
:guard-hints (("Subgoal *1/3'"
:use ((:instance assoc-of-append
(x a) (y b) (z c)))))
:mode :logic
:normalize nil
:otf-flg t))
(example-body x y z i j))
</pre>
<br><br><br><a href="acl2-doc.html"><img src="llogo.gif"></a> <a href="acl2-doc-index.html"><img src="index.gif"></a>
</body>
</html>
|