Major Section: RELEASE-NOTES
(GCL only) A bug in symbol-package-name has been fixed that could be
exploited to prove nil, and hence is a soundness bug. Thanks to Dave
Greve for sending us an example of a problem with defcong (see below)
that led us to this discovery.
ACL2 now warns when defcong specifies equal as the first
equivalence relation, e.g., (defcong equal iff (member x y) 2). The
warning says that the rule has no effect because equal already refines
all other equivalence relations. Formerly, this caused an error unless
:event-name was supplied (see defcong), and in fact the error was a
nasty raw Lisp error on GCL platforms due to some mishandling of packages by
ACL2 that has been fixed (see the paragraph about symbol-package-name
above). Thanks to Dave Greve for sending a helpful example in his report of
this problem.
(GCL only) The build process was broken for GCL 2.6.0 (and perhaps some earlier versions), and has been fixed. Thanks to Jose Luis Ruiz-Reyna for bringing this problem to our attention.
(GCL only) We have increased the hole size to at least 20% of max-pages, which may eliminate some garbage collection at the expense of larger virtual memory (not larger resident memory or larger image). Thanks to Camm Maguire for helpful explanations on this topic.
We have clarified the guard warning message that is printed during evaluation of recursively-defined functions whose guards have not been verified, for example:
ACL2 Warning [Guards] in TOP-LEVEL: Guard-checking may be inhibited
on some recursive calls of executable counterparts (i.e., in the ACL2
logic), including perhaps EVENLP. To check guards on all recursive
calls:
(set-guard-checking :all)
To leave behavior unchanged except for inhibiting this message:
(set-guard-checking :nowarn)
And, ACL2 no longer prints that message when the guard was
unspecified for the function or was specified as T. Thanks to Serita
Nelesen for bringing the latter issue to our attention. Finally, ACL2 now
prints such a warning at most once during the evaluation of any top-level
form; thanks to Bill Young for pointing out this issue.
The function verbose-pstack has been enhanced to allow specified prover
functions not to be traced. See verbose-pstack.
Added lp, wet, and set-non-linearp to *acl2-exports*,
and hence to the "ACL2-USER" package.
The distributed book
books/arithmetic-3/bind-free/integerp.lisp has been modified in order to prevent
potential looping; specifically, the definition of function
reduce-integerp-+-fn-1. Thanks to Robert Krug for providing this change.
A small improvement was made in the wet failure message when the error
occurs during translation to internal form. Thanks to Jared Davis for
pointing out the obscurity of some wet error messages.
We have improved ACL2's evaluation mechanism for the function bad-atom<=,
which now is specified to return nil if neither argument is a so-called
``bad atom'' (as recognized by function bad-atom). The following events
had caused a hard error, for example. (We're sorry that bad-atom and
bad-atom<= are not documented, but we also consider it unlikely that
anyone needs such documentation; otherwise, please contact the implementors.)
(defun foo (x y) (declare (xargs :guard t)) (bad-atom<= x y)) (defun bar (x y) (declare (xargs :guard t)) (foo x y)) (thm (equal (bar 3 4) 7))We have also changed the guard on
alphorder to require both arguments
to be atoms.
For forms (local x) that are skipped during include-book, or during
the second pass of certify-book or encapsulate, ACL2 had
nevertheless checked that x is a legal event form. This is no longer the
case.
The proof-checker now does non-linear arithmetic when appropriate. It
had formerly ignored set-non-linearp executed in the ACL2 command loop.
Incremental releases are now supported. See version and see set-tainted-okp. Thanks to Hanbing Liu for discovering a flaw in our original design.
The pattern-matching algorithm for :rewrite rules has been made
slightly more restrictive, thanks to a suggestion and examples from Robert
Krug. For example, previously one could get an infinite loop as follows.
(defstub foo (x) t)
(defaxiom foo-axiom
(equal (foo (+ 1 x))
(foo x)))
(thm (foo 0)) ; or replace 0 by any integer!
That is because the term (foo 0) was considered to match against the
pattern (foo (+ 1 x)), with x bound to -1. While such matching
is sound, it leads to an infinite loop since it allows foo-axiom to
rewrite (foo 0) to (foo -1), and then (foo -1) to (foo -2),
and so on. The fix is to insist that the new value, in this case -1, is
no larger in size according to acl2-count than the old value, in this
case 0. Since that test fails, the match is considered to fail and the
loop no longer occurs. An analogous fix has been made for multiplication,
where now we only match when the new term is still a non-zero integer. That
change avoids a loop here.
(defstub foo (x) t)
(defaxiom foo-axiom
(equal (foo (* 2 x))
(foo x)))
(thm (foo 0)) ; or try (thm (foo 4))
Added macro find-lemmas in books/misc/find-lemmas.lisp (see
brief documentation there) for finding all lemmas that mention all function
symbols in a given list.
:Restrict hints now work for :definition rules, though
they continue to be ignored by the preprocessor and hence you may want to use
:do-not '(preprocess) with any restrict hints. Thanks to John Matthews
for pointing out the lack of support for :definition rules in
:restrict hints.
Some books have been updated. In particular, there is a new directory
books/workshops/2004/ in workshops distribution, for the 2004 ACL2
workshop. There is also a new version of Jared Davis's ordered sets library,
formerly in books/finite-set-theory/osets-0.81/ but now in
books/finite-set-theory/osets/.
Fixed a bug in the (under-the-hood) raw Lisp definition of defchoose,
which had been causing a warning in CMU Common Lisp.
[Technical improvements related to the use of ``make dependencies'' for
certifying distributed books:]
File books/Makefile-generic now does a
better job with ``make dependencies,'' specifically with respect to
handling *.acl2 files and handling include-book commands with
:dir :system. Regarding the latter, suppose for example that book
basic.lisp contains the line:
(include-book "arithmetic/top-with-meta" :dir :system)Then
make dependencies would generate the following line:
basic.cert: $(ACL2_SRC_BOOKS)/arithmetic/top-with-meta.certThus, if
:dir :system is used with include-book, the corresponding
Makefile should define the variable ACL2_SRC_BOOKS. A standard
Makefile header for a books directory could thus be as follows.
# The following variable should represent the ACL2 source directory. It is the # only variable in this Makefile that may need to be edited. ACL2_SRC = ../../../../../..Finally, the ``ACL2_SRC_BOOKS = $(ACL2_SRC)/books include $(ACL2_SRC_BOOKS)/Makefile-generic ACL2 = $(ACL2_SRC)/saved_acl2
-s'' flag may now be omitted when running
``make dependencies.''