ACL2 Version 3.0.2 (December, 2006) Notes
Major Section:  RELEASE-NOTES

NOTE: This is an annotated copy of :DOC NOTE-3-0-2.

NOTE! New users can ignore these release notes, because the documentation has been updated to reflect all changes that are recorded here.

Fixed soundness bugs in the handling of primitive function pkg-witness, and improved its documentation. (The executable counterpart returned an incorrect default value, and the axiom symbol-package-name-pkg-witness-name needed pkg-name to be other than "" in order to avoid the default value of "ACL2".) As fallout, a new built-in :forward-chaining rule, symbol-package-name-of-symbol-is-not-empty-string, now asserts that the symbol-package-name of a symbol is never "". Thanks to Mike Gordon for bringing these soundness bugs to our attention by attempting to prove translations of ACL2 axioms in HOL4.

Pkg-witness bug allowed proof of nil. This shows the delicacy of *1* functions.

Fixed a soundness bug in linear arithmetic, due to incomplete tracking of forced assumptions while deriving inequalities. Thanks to Robert Krug for providing a fix and a proof of nil before the fix.

Tracking of forcing is subtle.

Fixed a soundness bug in the redundancy criterion for defun events, which has been modified; see redundant-events. This bug is illustrated below. Thanks to Peter Dillinger and Jared Davis for contributions to an email thread that led us to discover this bug. The solution is that for a definition to be redundant with an earlier definition, ACL2 no longer ignores :measure xargs except when skipping proofs (e.g., during include-book). However, a new ``measure'', (:? v1 ... vk), is supported, for specifying a measured subset of the set of formals, i.e., a set of formals that serves as the set of parameters for some valid measure.

(local (defun foo (x y) (declare (xargs :measure (acl2-count y))) (if (and (consp x) (consp y)) (foo (cons x x) (cdr y)) y)))

; So the following is redundant -- but it guesses a measure ; of (acl2-count x), which isn't right! (defun foo (x y) (if (and (consp x) (consp y)) (foo (cons x x) (cdr y)) y)))

; end of encapsulate

; Now we prove a non-theorem by exploiting the bug above, ; erroneously replacing formal y by a constant in the induction ; scheme hinted below. (This should not be allowed, as y should be ; labeled as a measured formal.)

(defthm bad (atom x) :rule-classes nil :hints (("Goal" :induct (foo x '(3)))))

; It's easy to get a contradiction by instantiating the ; non-theorem just above. (defthm contradiction nil :rule-classes nil :hints (("Goal" :use ((:instance bad (x '(7)))))))

We've previously talked about this one at some length.

Fixed a bug in :pl and the proof-checker's show-rewrites (sr) command that was causing a Lisp break. For :pl, also improved the display of unifying substitutions, modified output to take binding hypotheses (equal var term) into account properly, and arranged for inclusion of meta rules that modify the given term. Thanks to Eric Smith for bringing these issues to our attention.

Made nice :pl improvements.

Introduced new utilities for undoing commands, :ubu and :ubu!, which are analogous to :ubt and :ubt! (respectively) except that they only undo back up to, but not including, the indicated command.

:Ubu and :ubu! can be quite handy.

Fixed a performance bug, pointed out by Eric Smith, that was negating efforts made for the preceding release to avoid computation for disabled warnings.

Added time$ and value-triple to *acl2-exports*. Thanks to Bob Boyer and Erik Reeber (respectively) for bringing these issues to our attention.

Improved the automatic proclaiming of function types for GCL and OpenMCL, specifically to use an output format consistent with the Common Lisp spec. Thanks to Bob Boyer for bringing this issue to our attention.

Added books/misc/transfinite.lisp, which deals with transfinite induction in ACL2. Thanks to Eric Smith for contributing this book.

Illustrates how to do induction proofs by functional instantiation.

Added books/misc/process-book-readme.lisp to the distribution. Thanks to Sandip Ray for pointing out its omission.

Please contribute books! It's pretty simple now; look for link with "How to Contribute" on home page.

Added contributions books/concurrent-programs/bakery/ and books/concurrent-programs/german-protocol/. These contributions can be used as tutorials, especially by new ACL2 users, for learning how to model concurrent protocols in ACL2 and the steps involved in reasoning about their correctness. Thanks to Sandip Ray for these contributions. See the Readme.lsp files in these directories.

Perhaps Sandip has a word to say about these?

Theory invariants may now involve the variable ENS instead of the variable THEORY. The practical effect of this change is that any expression of the form (MEMBER-EQUAL rune THEORY) occurring in a theory-invariant expression should be replaced by (ACTIVE-RUNEP rune). See theory-invariant. Thanks to Eric Smith and Dave Greve for pointing out an inefficiency in the handling of theory invariants that led to this change, which can speed up their handling by orders of magnitude on large examples, and to Eric for testing this change and pointing out problems with an early implementation of it.

You'll need to change any theory-invariant expressions to use ENS instead of THEORY.

Theory invariants (see theory-invariant) are no longer checked on theories defined by deftheory events. After all, one can define a theory with deftheory that is not intended to be used as the current theory, but rather is intended to be combined with other theories (see theory-functions). Thanks to Eric Smith for bringing this issue to our attention.

Theory-invariant errors had been reported with very little detail when warnings were inhibited. This problem has been fixed; thanks to Eric Smith for bringing it to our attention and providing an example. We have also improved the handling of redundancy for theory-invariant events.

The macro defun-sk now has a new optional keyword, rewrite, that can be used to change the form of the :rewrite rule generated when the quantifier is forall. Thanks to Eric Smith and Sandip Ray for useful discussions on this topic. We have also slightly modified the hints for the defthm event underneath a defun-sk in order to make the proof more reliably efficient.

Users have wanted defun-sk to create nicer rewrite rules. This still isn't the default, but at least now it's easy to get this behavior.

A new event, reset-prehistory, allows setting of a barrier before which undoing is illegal. An argument to this macro allows the barrier to be made permanent; otherwise, it can be removed with :ubt-prehistory. Thanks to Peter Dillinger for useful conversations leading to the addition of reset-prehistory.

I could give a demo. This is nice if you have some preliminary stuff that you want to leave around.

A new query, (wormhole-p state), allows users to determine whether or not they are in a wormhole. Thanks to Peter Dillinger for providing this utility.

Value-triple no longer evaluates its form during include-book, and in raw Lisp its calls trivially macroexpand to nil, without any consideration of its argument. This change avoids errors and warnings when stobj names occur in the argument.

We fixed what could be considered a soundness hole that could occur by exploiting redefinition in a particular way. Thanks to Peter Dillinger for raising a question that led to discovery of this hole.

; The "what could be considered a soundness hole" could be exploited as
; follows.

;     (in-package "ACL2")
;     ; Portcullis commands:
;     #|
;     (set-ld-redefinition-action '(:warn! . :overwrite) state)
;     (encapsulate
;      ()
;      (defun foo () t)
;      (local (defun foo () nil))
;      (defthm foo-prop
;        (equal (foo) nil)
;        :rule-classes nil))
;     |# ; |
;     ; NOTE: After the above commands:
;     ; ACL2 !>(redefined-names state)
;     ; NIL
;     ; ACL2 !>
;     ; Now execute:
;     ; (certify-book "bad" 1)
;     (defthm contradiction
;       nil
;       :hints (("Goal" :use foo-prop))
;       :rule-classes nil)

; After certifying the book we can do this:

; (include-book "bad")
; (thm nil :hints (("Goal" :use contradiction)))

A bug has been fixed in handling of illegal theory expressions. Thanks to Eric Smith, who reported this problem and provided the example (in-theory '((:definition natp) (:rewrite doesntexist))) to show how a hard error could occur.

Improved error reporting by certify-book when the certification world contains inadmissible forms.

Modified defchoose to add two new keyword arguments. There is now a :doc keyword argument; previously, an optional documentation string (see doc-string) was to be placed just before the body, without a keyword. There is also a :strengthen argument that strengthens the axiom added, which allows for the definition of ``fixing'' functions for equivalence relations that choose canonical representatives of equivalence classes. See defchoose. Thanks for Dave Greve for useful discussions that led us to this :strengthen enhancement.

The new axiom allows the obtaining of a canonical representative.

Added books/misc/bash.lisp, which provides utilities for simplifying a goal into a list of subgoals (as documented at the top of that file). Thanks to Dave Greve for requesting this utility and suggesting refinements to its functionality, which have been incorporated.

Note also expander.lisp.

(For Emacs users only) The command meta-x new-shell provided by file emacs/emacs-acl2.el now puts you in shell-mode, which for example supports directory tracking. Thanks to Jared Davis for suggesting this change.

Fixed some mishandling of stobjs by make-event expansion.

Introduced a new event, defttag, that introduces a ``trust tag'' (``ttag'') allowing for extensions of ACL2 and for the use of generally unsafe ACL2 constructs. Thanks to Peter Dillinger, Sandip Ray, and Erik Reeber for useful discussions on defttag and the following related items.

This is a pretty sweeping change, since it lets you put almost anything into a book, even raw Lisp! I might discuss some of the sub-points just below.

A new event, remove-untouchable, can be used to give users access to system functions and data structures. We also fixed a bug in push-untouchable; and, it no longer is a no-op in :program mode. Thanks to Peter Dillinger for proposing remove-untouchable and suggesting that it and push-untouchable be functional in :program mode.

Raw-mode (see set-raw-mode) no longer disables certify-book. However, set-raw-mode is now disallowed unless there is an active ttag (see defttag). If you want to execute (set-raw-mode t) and there is no active ttag, consider executing (set-raw-mode-on!) instead.

Redefinition of system functions is disallowed unless there is an active ttag. However, redef! now introduces (defttag :redef!) in order to allow redefinition of system functions.

A new event, progn!, is a legal embedded event form that can go in books and both encapsulate and progn forms (see embedded-event-form), and is similar to progn except that it allows arbitrary forms. Thus, a progn! form is potentially dangerous and can only be evaluated if there is an active ttag.

See ttags-seen for information about how to find the ttags known in the current ACL2 world, and for related caveats.

A new book created with Peter Dillinger, books/misc/hacker.lisp, uses progn! to define utiliities with-raw-mode and with-redef-allowed, which respectively allow raw Lisp evaluation and redefinition to take place within a certifiable book (!).

Macro with-output is no longer allowed in function bodies because it does not have (and has never had) any effect in raw Lisp. See with-output for a workaround.

Fixed a bug in redundancy of defstobj in raw Lisp, which caused an error when certifying a book with a redundant defstobj event whose stobj had already been modified. Here is an example:
(defstobj st fld)
(update-fld 3 st)
(certify-book "foo" 1) ; where foo.lisp contains (defstobj st fld)

New books illustrating make-event have been contributed in directory books/make-event/: dotimes.lisp (David Rager), stobj-test.lisp, and logical-tangent.lisp (Peter Dillinger).

Modified print-object$ (see io) so that it no longer prints an extra space at the end.

Replaced the ``draconian restriction to avoid capture'' that had prevented some :functional-instance hints from being legal. The corresponding check now only requires that no variable free in the functional substitution is captured by a let or mv-let (or lambda) binding. See lemma-instance.

Wow, this had been awaiting a fix for a long time.

Added new extended metafunction, mfc-rw+, which is equivalent to mfc-rw except that it takes an alist argument, which may be useful for efficiency. See extended-metafunctions. Thanks to Robert Krug for suggesting this more efficient variant of mfc-rw.

Added support for the ignorable declare form.

This one is particularly useful for automatically-generated code when you don't want to use set-ignore-ok.

We now cause an error on a call of open-input-channel (see io) with an argument string whose first character is the | character. Thanks to Bob Boyer for providing an example (several months ago) showing the danger of such calls, namely that the following command would log you out and kill all of your processes when running on top of GCL in Linux:

(open-input-channel "|kill -9 -1" :object state)

Restricted the use of make-event to contexts in which it can be tracked properly, under legal events (see embedded-event-form). Thanks to Peter Dillinger for bringing an example to our attention that led to this fix.

Fixed a bug that was avoiding guard-checking for the functions compress1 and compress2. Thanks to David Rager for bringing this bug to our attention.

Added an error message when a defun or mutual-recursion event fails, to clarify whether failure is for the measure conjecture or for the guard conjecture. Thanks to David Rager for requesting clarification for such failures.

Fixed a bug in reporting of guard violations (hard Lisp error) when certain macros (for example, cond) are used in the guard. Thanks to Jared Davis for bringing this problem to our attention and providing assistance with the solution, in particular by providing a helpful example.

Grant Passmore has contributed a resolution/paramodulation prover written in ACL2, in directory books/deduction/passmore/. Thanks, Grant.

I'm sure Grant would like feedback.

Improved the error message when illegal theories are encountered.

Improved the suppression of output for inhibit-output arguments of routines in the book books/misc/expander.lisp. Thanks to Qiang Zhang for pointing out the possibility for improvement here.

Added a new directory books/arithmetic-3/extra/ that extends books/arithmetic-3 with additional rules, contributed by Alex Spiridonov with guidance from Robert Krug. WARNING: This directory is under development. It may undergo large changes in future releases, so please consider it experimental and subject to change. Feedback is welcomed.

Alex and Robert would probably like feedback.

As part of the work mentioned just above, Robert Krug and Alex Spiridonov contributed improvements to books/arithmetic-3/:

o A new rule |(* (/ x) (/ (expt x n)))| in bind-free/collect.lisp, which is important for reducing collect-* expressions though it slowed down one proof (see comment above this rule in bind-free/collect.lisp).

o Slight improvements of rules integerp-mod and rationalp-mod in floor-mod/floor-mod.lisp.

o To avoid conflict with books/rtl/rel6/arithmetic/, renamed rule mod-minus to mod-neg in floor-mod/floor-mod.lisp, and renamed integerp-+-reduce-leading-constant to integerp-+-reduce-leading-rational-constant in bind-free/integerp.lisp.

(GCL on Windows only) Made a low-level change to avoid multiplying stacks for GCL on Windows, since GCL 2.6.6 broke while doing this.

Fixed bugs in linear arithmetic (rarely evidenced, it seems) involving using < to compare complex rational constants. Thanks to Robert Krug for helping with the fixes.

Added a new event, assert-event, for checking that forms evaluate to non-nil values. Thanks to Peter Dillinger for suggesting and collaborating on this addition.

Could be useful.