MAKE-EVENT-DETAILS

details on make-event expansion
Major Section:  MAKE-EVENT

The normal user of make-event can probably ignore this section, but we include it for completeness. We assume that the reader has read and understood the basic documentation for make-event (see make-event), but we begin below with a summary of expansion.

Introduction

Here is a summary of how we handle expansion involving make-event forms.

(make-event form :check-expansion nil)

This shows the :check-expansion default of nil, and is typical user input. We compute the expansion exp of form, which is the expansion of the original make-event expression and is evaluated in place of that expression.

(make-event form :check-expansion t)

The user presumably wants it checked that the expansion doesn't change in the future, in particular during include-book. If the expansion of form is exp, then we will evaluate exp to obtain the value as before, but this time we record that the expansion of the original make-event expression is (make-event form :check-expansion exp) rather than simply exp.

(make-event form :check-expansion exp) ; exp a cons

This is generated for the case that :check-expansion is t, as explained above. Evaluation is handled as described in that above case, except here we check that the expansion result is the given exp. (Actually, the user is also allowed supply such a form.) The original make-event expression does not undergo any expansion (intuitively, it expands to itself).

Now let us take a look at how we expand progn forms (encapsulate is handled similarly).

(progn ... (make-event form :check-expansion nil) ...)

The expansion is obtained by replacing the make-event form as follows. Let exp be the expansion of form, Then replace the above make-event form, which we denote as F, by (record-expansion F exp). Here, record-expansion is a macro that returns its second argument.

(progn ... (make-event form :check-expansion t) ...)

The expansion is of the form (record-expansion F exp) as in the nil case above, except that this time exp is (make-event form :check-expansion exp'), where exp' is the expansion of form.

(progn ... (make-event form :check-expansion exp) ...) ; exp a cons

No expansion takes place unless expansion takes place for at least one of the other subforms of the progn, in which case each such form F is replaced by (record-expansion F exp) where exp is the expansion of F.

Detailed semantics

In our explanation of the semantics of make-event, we assume familiarity with the notion of ``embedded event form'' (see embedded-event-form).

Let's say that the ``actual embedded event form'' corresponding to a given form is the underlying call of an ACL2 event: that is, LOCALs are dropped when ld-skip-proofsp is 'include-book, and macros are expanded away, thus leaving us with a progn, a make-event, or an event form (possibly encapsulate), any of which might have surrounding local, skip-proofs, or with-output calls.

Thus, such an actual embedded event form can be viewed as having the form (rebuild-expansion wrappers base-form) where base-form is a progn, a make-event, or an event form (possibly encapsulate), and wrappers are (as in ACL2 source function destructure-expansion) the result of successively removing the event form from the result of macroexpansion, leaving a sequence of (local), (skip-proofs), and (with-output ...) forms. In this case we say that the form ``destructures into'' the indicated wrappers and base-form, and that it can be ``rebuilt from'' those wrappers and base-form.

Elsewhere we define the notion of the ``expansion result'' from an evaluation (see make-event), and we mention that when expansion concludes, the ACL2 logical world and most of the state are restored to their pre-expansion values. Specifically, after evaluation of the argument of make-event (even if it is aborted), the ACL2 logical world is restored to its pre-evaluation value, as are all state global variables in the list *protected-system-state-globals*. Thus, assignments to user-defined state globals (see assign) do persist after expansion, since they are not in that list.

We recursively define the combination of evaluation and expansion of an embedded event form, as follows. We also simultaneously define the notion of ``expansion takes place,'' which is assumed to propagate upward (in a sense that will be obvious), such that if no expansion takes place, then the expansion of the given form is considered to be itself. It is useful to keep in mind a goal that we will consider later: Every make-event subterm of an expansion result has a :check-expansion field that is a consp, where for this purpose make-event is viewed as a macro that returns its :check-expansion field. (Implementation note: The latest expansion of a make-event, progn, progn!, or encapsulate is stored in state global 'last-make-event-expansion, except that if no expansion has taken place for that form then 'last-make-event-expansion has value nil.)

If the given form is not an embedded event form, then simply cause a soft error, (mv erp val state) where erp is not nil. Otherwise:

If the evaluation of the given form does not take place (presumably because local events are being skipped), then no expansion takes place. Otherwise:

Let x be the actual embedded event form corresponding to the given form, which destructures into wrappers W and base-form B. Then the original form is evaluated by evaluating x, and its expansion is as follows.

If B is (make-event form :check-expansion val), then expansion takes place if and only if val is not a consp and no error occurs, as now described. Let R be the expansion result from protected evaluation of form, if there is no error. R must be an embedded event form, or it is an error. Then evaluate/expand R, where if val is not nil then state global 'ld-skip-proofsp is initialized to nil. (This initialization is important so that subsequent expansions are checked in a corresponding environment, i.e., where proofs are turned on in both the original and subsquent environments.) It is an error if this evaluation causes an error. Otherwise, the evaluation yields a value, which is the result of evaluation of the original make-event expression, as well as an expansion, E_R. Let E be rebuilt from W and E_R. The expansion of the original form is E if val is nil, and otherwise is the result of replacing the original form's :check-expansion field with E, with the added requirement that if val is not t (thus, a consp) then E must equal val or else we cause an error.

If B is either (progn form1 form2 ...) or (encapsulate sigs form1 form2 ...), then after evaluating B, the expansion of the original form is the result of rebuilding from B, with wrappers W, after replacing each formi in B for which expansion takes place by (record-expansion formi formi'), where formi' is the expansion of formi. Note that these expansions are determined as the formi are evaluated in sequence (where in the case of encapsulate, this determination occurs only during the first pass). Except, if no expansion takes place for any formi, then the expansion of the original form is itself.

Otherwise, the expansion of the original form is itself.

Similarly to the progn and encapsulate cases above, book certification causes a book to be replaced by its so-called ``book expansion.'' There, each event ev for which expansion took place during the proof pass of certification -- say, producing ev' -- is replaced by (record-expansion ev ev').

Implementation Note. The book expansion is actually implemented by way of the :expansion-alist field of its certificate, which associates 0-based positions of top-level forms in the book (not including the initial in-package form) with their expansions. Thus, the book's source file is not overwritten; rather, the certificate's expansion-alist is applied when the book is included or compiled. End of Implementation Note.

It is straightforward by computational induction to see that for any expansion of an embedded event form, every make-event sub-event has a consp :check-expansion field. Here, by ``sub-event'' we mean to expand macros; and we also mean to traverse progn and encapsulate forms as well as :check-expansion fields of make-event forms. Thus, we will only see make-event forms with consp :check-expansion fields in the course of include-book forms, the second pass of encapsulate forms, and raw Lisp. This fact guarantees that an event form will always be treated as its original expansion.

Notes on ttags

See defttag for documentation of the notion of ``trust tag'' (``ttag''). We note here that even if an event (defttag tag-name) for non-nil tag-name is admitted only during the expansion phase of a make-event form, then such expansion will nevertheless still cause tag-name to be recorded in the logical world (assuming that the make-event form is admitted). So in order to certify such a book, a suitable :ttags argument must be supplied; see certify-book.

ACL2 does provide a way to avoid the need for :ttags arguments in such cases. The idea is to certify a book twice, where the results of make-event expansion are saved from the first call of certify-book and provided to the second call. see set-write-acl2x.

Finally, we discuss a very unusual case where certification does not involve trust tags but a subsequent include-book does involve trust tags: a make-event call specifying :check-expansion t, whose expansion generates a defttag event during include-book but not certify-book. Consider the following book.

(in-package "ACL2")
(make-event
 (er-progn
  (if (@ skip-notify-on-defttag) ; non-nil when including a certified book
      (pprogn
       (fms "Value of (@ skip-notify-on-defttag): ~x0~|"
            (list (cons #0 (@ skip-notify-on-defttag)))
            *standard-co* state nil)
       (encapsulate
        ()
        (defttag :foo)
        (value-triple "Imagine something bad here!")))
    (value nil))
  (value '(value-triple :some-value)))
 :check-expansion t)
This book certifies successfully without the need for a :ttags argument for certify-book. Indeed, the above book's certificate does not specify :foo as a trust tag associated with the book, because no defttag event was executed during book certification. Unfortunately, if we try to include this book without specifying a value of :ttags that allows :foo, book inclusion will cause executing of the above defttag. If that inclusion happens in the context of certifying some superior book and the appropriate :ttags arguments have not been provided, that certification will fail.