NOTE-2-9-5

Changes in Version 3.0 since Version 2.9.4
Major Section:  RELEASE-NOTES

Fixed a bug in cw-gstack that was causing a hard error when attempting to report on a forced assumption. Thanks to Jared Davis for pointing this out and sending an example that helped us to determine a fix.

Added set-backchain-limit to the set of legal events that can be placed in encapsulate forms and books. Thanks to John Cowles for bringing this issue to our attention.

Fixed a bug that broke wet. Thanks to David Rager for bringing this bug to our attention.

Guard verification now evaluates ground subexpressions (those with no free variables) when computing the guard conjecture for the body of a function. Thanks to Jared Davis for useful conversations leading to this change. See verify-guards, in particular its ``Note on computation of guard conjectures and evaluation'' near the end of that topic, for more details.

Added a warning when a theory-invariant is redefined. Thanks to Jared Davis for suggesting a warning in this case and providing an informative example. Also, theory-invariants are now maintained more completely, as they are checked at the end of every event except for events executed on behalf of an include-book or the second pass of an encapsulate.

Fixed the handling of runic designators to match their specification (see theories), so that disabling the name of a defthm event disables all rules generated for that event.

(For those who do numerous builds using feature :acl2-mv-as-values, currently only OpenMCL and multi-threaded SBCL by default:) You can speed up builds by adding the following parameter to make, under conditions described in GNUmakefile: USE_ACL2_PROCLAIMS=:REUSE.

Arranged that traced functions (see trace$) are automatically untraced when events are undone (for example see ubt), at least for most underlying Common Lisp implementations.

The macro defun-sk now creates non-executable functions, which allows stobjs to be used where they had previously been prohibited. More generally, the user now has control over declare forms to be used by the underlying defun'd function; see defun-sk. Thanks to Sandip Ray for pointing out the need for such a modification.

:Definition rules are now treated, at least by default, as truly first-class definitions. In particular, :expand hints use the latest :definition rule by default. You may specify :install-body nil to get the previous behavior of :definition rules; See definition, and you may choose a previously-installed :definition rule to provide the current body; see set-body. Also see rule-classes for details of the :install-body field, and see hints to see a new :with directive for controlling expansion. The :with directive for :expand hints can even direct the use of a :rewrite rule for expansion! Thanks to various people, including Sandip Ray and Rob Sumners, for discussions on the issue of the applicability of :definition rules for :expand hints.

Constraints for functional instantiation now use the original definition rather than a simplified (``normalized'') version of it.

Fixed a bug that caused the prompt to stay the same when guard-checking is off (see set-guard-checking) and raw-mode is changed (see set-raw-mode).

Lemma names in directory books/ordinals have been changed by replacing /\ with & and replacing \/ with V. We made this change because backslash is an escape character and hence disappears unless it is itself escaped.

Fixed proof-tree output so that failed non-proof events do not cause the proof-tree to be re-printed. Thus for example, if you have already advanced the checkpoint marker, it will not be reset by subequent failed non-proof events. Thanks to Pete Manolios and Peter Dillinger for bringing this bug to our attention.

Fixed a bug that was preventing the printing of stobj fields as constants instead of numbers in certain cases. (Note that this bug only affected printing, not soundness.) Thanks to Eric Smith for bringing this problem to our attention and providing the following example (which now works fine).

(defstobj st fld1 fld2)
(in-theory (disable update-nth))
(defund run (st)
  (declare (xargs :stobjs (st))) ;adding this didn't seem to help..
  st)

;works great; *fld1* prints as  *fld1*
(thm (equal (update-nth *fld1* 'abc st)
            (car (cons x y))))

;*fld1* gets printed as 0, presumably because the call to run intervenes.
(thm (equal (update-nth *fld1* 'abc (run st))
            (car (cons x y))))

The macro progn now allows the use of macros defined within its bodies even when at the event level, as illustrated by the following example.

(progn (defmacro my-defun (&rest args)
         `(defun ,@args))
       (my-defun g (x) x))
Thanks to Anna Slobodova for bringing this issue to our attention. A related change is that all arguments of progn must now be embedded event forms (see embedded-event-form), so use er-progn instead if this is not the case.

The change to progn mentioned above also fixes a bug in handling local events inside a progn that is inside an encapsulate or in a book. For example, the following form formerly caused an error.

(encapsulate
 ()
 (defun foo (x) x)
 (progn (local (defun bar (x) x))
        (defun abc (x) x)))

We fixed two bugs in :puff and :puff*. The first, brought to our attention by Eric Smith (who we thank), caused a cryptic error message when puffing a command with no subsidiary stored events; try, for example, (encapsulate () (value-triple 3)). The second was due to a failure to restore the acl2-defaults-table. Suppose for example that we have certified the book foo.lisp, which contains (program) followed by some definitions and/or theorems. Now suppose we start ACL2 and execute the following.

(include-book "foo")
(defthm test-thm
  (equal x x)
  :rule-classes nil)
If we now execute :puff 1, ACL2 will roll back the world to before the include-book; then ``puff'' the include-book, which will leave us in :program mode; and finally skip re-execution of the defthm because such events are skipped in :program mode. The fix is to re-install the acl2-defaults-table immediately after the include-book to its pre-include-book value.

A new event, make-event, provides something like macros that take state. For example, one can use it to put tests into certified books, do proof search, and generate new function names. Many examples appear in directory books/make-event/. See make-event. Thanks to Bob Boyer and Jared Davis for useful feedback and to Warren Hunt, David Rager, and Sandip Ray for helpful discussions leading to some of the examples in directory books/make-event/.

In support of make-event, which is described in the preceding paragraph, certify-book has a new keyword argument, :save-expansion, that controls whether the result of expanding make-event forms is written out to a file. See certify-book; and for a discussion of book expansion files, see make-event.

We fixed a soundness bug that did not correctly detect local events. For example, the following event was admitted.

(encapsulate
 ()
 (local
  (encapsulate
   ()
   (local (progn (program))) ; or, (local (with-output :off summary (program)))
   (set-irrelevant-formals-ok t)
   (defun foo (x)
     (declare (xargs :measure (acl2-count x)))
     (1+ (foo x)))))
 (defthm inconsistent
   nil
   :hints (("Goal" :use foo))
   :rule-classes nil))

A new value for guard checking, :none, is now allowed. If you execute :set-guard-checking :none, then no guard checking will take place (but raw Lisp code will not be executed in this case). As a result, you should never see a guard violation, even for calls of :program mode functions. We thank Pete Manolios, who has long wanted this feature, and also Peter Dillinger, for asking for it. New documentation explains the interaction between the defun-mode and the value supplied to :set-guard-checking. See guard-evaluation-table, see guard-evaluation-examples-script, and see guard-evaluation-examples-log.

In the course of adding the guard-checking value :none described in the paragraph above, we eliminated an optimization that eliminated guard checking for some recursive calls of :logic mode mutually-recursive functions that have not had their guards verified. But we doubt that this change will be noticed by any users!)

The ACL2 hyper-card has been enhanced, thanks to David Rager, with a listing of ``Useful EMACS Commands'' to match comments in emacs/emacs-acl2.el.

Users contributed books following the Readme.lsp methodology: data-structures/memories and unicode (Jared Davis), proofstyles (Sandip Ray and J Moore).

Made some improvements to books/Makefile-generic (a file discussed elsewhere; see book-makefiles). In particular, improved handling of .acl2 files for dependencies target.

(Only OpenMCL and, with feature :acl2-mv-as-values, GCL) Fixed a bug that was causing proclaiming to fail when definitions are submitted interactively.

The default stack size has been increased for several lisps.

(Very technical) A restriction has been weakened on the use of local stobjs under a call of an ACL2 evaluator (trans-eval or simple-translate-and-eval). Now, the error can only take place for stobj names that occur in the term being evaluated. Thanks to Erik Reeber for bringing this issue to our attention.

The notion of ``ancestor'' has been changed slightly. This notion is used by extended metafunctions and break-rewrite (see extended-metafunctions and see brr@), and also with backchain limits (see backchain-limit and see set-backchain-limit). Basically, each time a hypothesis is encountered during application of a rewrite rule, that hypothesis is pushed (after instantiating and negating) onto the current list of ancestors before it is rewritten. However, hypotheses of the form (equal var term), where var is free (see free-variables), had not been included in the ancestors (similarly for (equiv var (double-rewrite term)) where equiv is a known equivalence relation). Now such ``binding hypotheses'' are included in a special way in ancestors data structures. In particular, (null (mfc-ancestors mfc)) will now be true if and only if the term being rewritten is part of the current goal as opposed to a hypothesis from a rule encountered during backchaining, even if that hypothesis is a binding hypothesis. Thanks to Dave Greve for bringing this issue to our attention.

Termination and induction analysis now continue through both arguments of prog2$, not just the second. (Normally, the gathering up of if tests stops at function calls; but it continued through the second argument of prog2$, and now it will continue through both arguments.) Thanks to Sol Swords for discussion leading to this change.

The ACL2 distribution is now kept on the http server rather than the ftp server (but the home page has not been moved). Thanks to Robert Krug for letting us know that some ACL2 users have found it inconvenient to fetch ACL2 using ftp.

The file books/README.html has been renamed to books/Readme.html, since some browsers don't show the former in the directory listing.