Major Section: BOOKS
Include-book
has a special provision for dealing with uncertified
books: If the file has no certificate or an invalid
certificate (i.e., one whose check sums describe files other
than the ones actually read), a warning is printed and the book is
otherwise processed as though it were certified and had an open
portcullis. (For details see books, see certificate,
and see portcullis.)
This can be handy, but it can have disastrous consequences.
The provision allowing uncertified books to be included can have disastrous consequences, ranging from hard lisp errors, to damaged memory, to quiet logical inconsistency.
It is possible for the inclusion of an uncertified book to render
the logic inconsistent. For example, one of its non-local
events
might be (defthm t-is-nil (equal t nil))
. It is also possible
for the inclusion of an uncertified book to cause hard errors or
breaks into raw Common Lisp. For example, if the file has been
edited since it was certified, it may contain too many open
parentheses, causing Lisp to read past ``end of file.'' Similarly,
it might contain non-ACL2 objects such as 3.1415
or ill-formed
event forms that cause ACL2 code to break.
Even if a book is perfectly well formed and could be certified (in a
suitable extension of ACL2's initial world), its uncertified
inclusion might cause Lisp errors or inconsistencies! For example,
it might mention packages that do not exist in the host world.
The portcullis of a certified book ensures that the correct
defpkg
s have been admitted, but if a book is read without
actually raising its portcullis, symbols in the file, e.g.,
acl2-arithmetic::fn
, could cause ``unknown package'' errors in
Common Lisp. Perhaps the most subtle disaster occurs if the host
world does have a defpkg
for each package used in the book
but the host defpkg
imports different symbols than those required
by the portcullis. In this case, it is possible that formulas
which were theorems in the certified book are non-theorems in the
host world, but those formulas can be read without error and
will then be quietly assumed.
In short, if you include an uncertified book, all bets are off regarding the validity of the future behavior of ACL2.
That said, it should be noted that ACL2 is pretty tough and if
errors don't occur, the chances are that deductions after the
inclusion of an uncertified book are probably justified in the
(possibly inconsistent) logical extension obtained by assuming the
admissibility and validity of the definitions and conjectures in the
book.