`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, `LOCAL`

s 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-state-globals-for-make-event*`

. 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`

, 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`

.)

Similarly to theIf 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.

`progn`

and `encapsulate`

cases above, book
certification causes a book to 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.

**A note on ttags**

See defttag for documentation of the notion of ``trust tag'' (``ttag'').
Here, we simply observe that if an event `(defttag tag-name)`

for
non-`nil`

`tag-name`

is admitted during the expansion phase of a
`make-event`

form, then although a ```TTAG NOTE`

'' will be printed to
standard output, and moreover `tag-name`

must be an allowed tag
(see defttag), nevertheless such expansion will not cause `tag-name`

to be
recorded once the expansion is complete. That is, there will be no lingering
effect of this `defttag`

form after the `make-event`

expansion is
complete; no certificate written will be affected (where we are certifying a
book), and the set of allowed ttags will not be affected. So for example, if
this `make-event`

form is in the top-level loop and subsequently we certify
or include a book, then `tag-name`

will not be associated with the
top-level loop by this `make-event`

form.