Major Section: EVENTS
WARNING! This event is intended for advanced users who, in essence, want to build extensions of ACL2. See see defttag, in particular, the ``WARNING'' there, and see the warning about stobjs at the end of this documentation topic.
Progn! can be used like progn, even in books. But unlike
progn, progn! does not require its constituent forms to be
events (see embedded-event-form), except that the first form cannot be
a symbol unless it is :state-global-bindings (advanced feature, described
below). However, see make-event for a ``Restriction to the Top Level'' that
still applies under a call of progn!.
Because progn! allows non-events, it differs from progn in another
important respect: progn! is illegal unless there is an active ttag;
see defttag.
See community book books/hacking/hacker.lisp for two macros,
with-raw-mode and with-redef-allowed, each defined in terms of
progn!, that allow arbitrary forms in contexts that would normally
require legal embedded event forms.
Given a form (progn! form1 form2 ... formk), ACL2 will evaluate each
formi in turn (for i from 1 to k). If a form returns more than one value
(see mv) where the first value returned is not nil, then no later form
will be evaluated and the result returned by the progn! call will be
(mv erp val state) for some non-nil value erp, signifying an
error (see ld-error-triples). Otherwise the evaluation is considered to
have succeeded, and will continue with later forms. The value returned by a
call of progn! with no such error is of the form (mv nil v state),
where v depends on the last form as follows. If the last form evaluates
to a single value, then v is that value, except if the value is a
stobj, say ST, then v is the symbol REPLACED-ST. Otherwise
the last form evaluates to some (mv nil x ...), and v is x unless
after the final form's evaluation we are in raw-mode (see set-raw-mode), in
which case the progn! call returns nil (so that ACL2 can at least
print the result -- imagine Lisp returning a pathname object from a load
call, for example).
The normal undoing mechanism does not generally apply to forms within a
progn! that are not legal ACL2 events (see embedded-event-form).
In particular, note that a non-local call of progn! in an
encapsulate event will generally be evaluated twice: once on each pass.
This fact is worth keeping in mind if you are using progn! to change the
state of the system; ask yourself if it is acceptable to apply that
state-changing operation more than once.
Please note that progn! may differ from progn in the following
sense: definitions within a call of progn! might not be compiled. For
example, consider the following book.
(in-package "ACL2") (defttag :test) (progn (defun f1 (x) x)) (progn! (defun f2 (x) x))If the underlying Lisp is GCL 2.6.7, then after including this certified book (where the default certification took place, creating a compiled file), then
f1 is a compiled function but f2 is not. For other Lisps supported
by ACL2, both f1 and f2 are compiled, though we are not sure that
every function under every call of progn! would similarly be compiled.We now describe, for system hackers only, a sophisticated extension of
progn! not mentioned above: support for keyword argument
:state-global-bindings. If the first argument of progn! is this
keyword, then the second argument is treated as a list of bindings as
expected by ACl2 system function state-global-let*. Thus, in the ACL2
loop,
(progn! :state-global-bindings bindings form1 form2 ... formk)is treated as follows:
(progn! (state-global-let* bindings (progn! form1 form2 ... formk)))However, in raw Lisp the former is just:
(progn form1 form2 ... formk)Thus, one should use the
:state-global-bindings argument with care, since
the behavior in the ACL2 loop can differ from that in Common Lisp. The
intention is that one bind only state global variables that are relevant
to evaluation of the forms within the ACL2 loop and are harmlessly ignored
for evaluation of those forms in raw Lisp. Here is a typical sort of
example, as state global ld-redefinition-action is not relevant to
the evaluation of defun in raw Lisp.
(progn! (remove-untouchable 'ld-redefinition-action nil)
(progn! :state-global-bindings
((ld-redefinition-action '(:doit . :overwrite)))
(defun foo (x)
(cons x x)))
(push-untouchable 'ld-redefinition-action nil))
Finally, we point out a pitfall of progn! related to stobjs. The
following book can cause a hard Lisp error, depending on the host Common
Lisp, when certified with a non-nil value for compile-flg
(see certify-book).
(in-package "ACL2") (defstobj st fld) (defttag :my-ttag) (progn! (update-fld 3 st))The problem is that the stobj variable
st is not known to raw Lisp.
The compilation problem disappears if the last form above is replaced with
the following two forms.
(include-book "hacking/hacker" :dir :system) (with-raw-mode (update-fld 3 *the-live-st*))
