(t call-cost))))
call-cost)))
-;;; Return some sort of guess for the cost of doing a test against TYPE.
-;;; The result need not be precise as long as it isn't way out in space. The
-;;; units are based on the costs specified for various templates in the VM
-;;; definition.
+;;; Return some sort of guess for the cost of doing a test against
+;;; TYPE. The result need not be precise as long as it isn't way out
+;;; in space. The units are based on the costs specified for various
+;;; templates in the VM definition.
(defun type-test-cost (type)
(declare (type ctype type))
(or (let ((check (type-check-template type)))
(+ 1
(if (numeric-type-low type) 1 0)
(if (numeric-type-high type) 1 0))))
+ (cons-type
+ (+ (type-test-cost (specifier-type 'cons))
+ (function-cost 'car)
+ (type-test-cost (cons-type-car-type type))
+ (function-cost 'cdr)
+ (type-test-cost (cons-type-cdr-type type))))
(t
(function-cost 'typep)))))
\f
;;;; checking strategy determination
-;;; Return the type we should test for when we really want to check for
-;;; Type. If speed, space or compilation speed is more important than safety,
-;;; then we return a weaker type if it is easier to check. First we try the
-;;; defined type weakenings, then look for any predicate that is cheaper.
+;;; Return the type we should test for when we really want to check
+;;; for TYPE. If speed, space or compilation speed is more important
+;;; than safety, then we return a weaker type if it is easier to
+;;; check. First we try the defined type weakenings, then look for any
+;;; predicate that is cheaper.
;;;
-;;; If the supertype is equal in cost to the type, we prefer the supertype.
-;;; This produces a closer approximation of the right thing in the presence of
-;;; poor cost info.
+;;; If the supertype is equal in cost to the type, we prefer the
+;;; supertype. This produces a closer approximation of the right thing
+;;; in the presence of poor cost info.
(defun maybe-weaken-check (type cont)
(declare (type ctype type) (type continuation cont))
(cond ((policy (continuation-dest cont)
(let ((stype-cost (type-test-cost stype)))
(when (or (< stype-cost min-cost)
(type= stype type))
- (setq found-super t)
- (setq min-type stype min-cost stype-cost))))))
+ (setq found-super t
+ min-type stype
+ min-cost stype-cost))))))
(if found-super
min-type
*universal-type*)))))
;;; Switch to disable check complementing, for evaluation.
(defvar *complement-type-checks* t)
-;;; Cont is a continuation we are doing a type check on and Types is a list
-;;; of types that we are checking its values against. If we have proven
-;;; that Cont generates a fixed number of values, then for each value, we check
-;;; whether it is cheaper to then difference between the proven type and
-;;; the corresponding type in Types. If so, we opt for a :HAIRY check with
-;;; that test negated. Otherwise, we try to do a simple test, and if that is
-;;; impossible, we do a hairy test with non-negated types. If true,
-;;; Force-Hairy forces a hairy type check.
+;;; CONT is a continuation we are doing a type check on and TYPES is a
+;;; list of types that we are checking its values against. If we have
+;;; proven that CONT generates a fixed number of values, then for each
+;;; value, we check whether it is cheaper to then difference between
+;;; the proven type and the corresponding type in TYPES. If so, we opt
+;;; for a :HAIRY check with that test negated. Otherwise, we try to do
+;;; a simple test, and if that is impossible, we do a hairy test with
+;;; non-negated types. If true, Force-Hairy forces a hairy type check.
;;;
-;;; When doing a non-negated check, we call MAYBE-WEAKEN-CHECK to weaken the
-;;; test to a convenient supertype (conditional on policy.) If debug-info is
-;;; not particularly important (debug <= 1) or speed is 3, then we allow
-;;; weakened checks to be simple, resulting in less informative error messages,
-;;; but saving space and possibly time.
+;;; When doing a non-negated check, we call MAYBE-WEAKEN-CHECK to
+;;; weaken the test to a convenient supertype (conditional on policy.)
+;;; If SPEED is 3, or DEBUG-INFO is not particularly important (DEBUG
+;;; <= 1), then we allow weakened checks to be simple, resulting in
+;;; less informative error messages, but saving space and possibly
+;;; time.
+;;;
+;;; FIXME: I don't quite understand this, but it looks as though
+;;; that means type checks are weakened when SPEED=3 regardless of
+;;; the SAFETY level, which is not the right thing to do.
(defun maybe-negate-check (cont types force-hairy)
(declare (type continuation cont) (list types))
(multiple-value-bind (ptypes count)
(t
(values :hairy res)))))))
-;;; Determines whether Cont's assertion is:
-;;; -- Checkable by the back end (:SIMPLE), or
-;;; -- Not checkable by the back end, but checkable via an explicit test in
-;;; type check conversion (:HAIRY), or
+;;; Determines whether CONT's assertion is:
+;;; -- checkable by the back end (:SIMPLE), or
+;;; -- not checkable by the back end, but checkable via an explicit
+;;; test in type check conversion (:HAIRY), or
;;; -- not reasonably checkable at all (:TOO-HAIRY).
;;;
-;;; A type is checkable if it either represents a fixed number of values (as
-;;; determined by VALUES-TYPES), or it is the assertion for an MV-Bind. A type
-;;; is simply checkable if all the type assertions have a TYPE-CHECK-TEMPLATE.
-;;; In this :SIMPLE case, the second value is a list of the type restrictions
-;;; specified for the leading positional values.
+;;; A type is checkable if it either represents a fixed number of
+;;; values (as determined by VALUES-TYPES), or it is the assertion for
+;;; an MV-Bind. A type is simply checkable if all the type assertions
+;;; have a TYPE-CHECK-TEMPLATE. In this :SIMPLE case, the second value
+;;; is a list of the type restrictions specified for the leading
+;;; positional values.
;;;
-;;; We force a check to be hairy even when there are fixed values if we are in
-;;; a context where we may be forced to use the unknown values convention
-;;; anyway. This is because IR2tran can't generate type checks for unknown
-;;; values continuations but people could still be depending on the check being
-;;; done. We only care about EXIT and RETURN (not MV-COMBINATION) since these
-;;; are the only contexts where the ultimate values receiver
+;;; We force a check to be hairy even when there are fixed values if
+;;; we are in a context where we may be forced to use the unknown
+;;; values convention anyway. This is because IR2tran can't generate
+;;; type checks for unknown values continuations but people could
+;;; still be depending on the check being done. We only care about
+;;; EXIT and RETURN (not MV-COMBINATION) since these are the only
+;;; contexts where the ultimate values receiver
;;;
-;;; In the :HAIRY case, the second value is a list of triples of the form:
-;;; (Not-P Type Original-Type)
+;;; In the :HAIRY case, the second value is a list of triples of
+;;; the form:
+;;; (NOT-P TYPE ORIGINAL-TYPE)
;;;
-;;; If true, the Not-P flag indicates a test that the corresponding value is
-;;; *not* of the specified Type. Original-Type is the type asserted on this
-;;; value in the continuation, for use in error messages. When Not-P is true,
-;;; this will be different from Type.
+;;; If true, the NOT-P flag indicates a test that the corresponding
+;;; value is *not* of the specified TYPE. ORIGINAL-TYPE is the type
+;;; asserted on this value in the continuation, for use in error
+;;; messages. When NOT-P is true, this will be different from TYPE.
;;;
-;;; This allows us to take what has been proven about Cont's type into
-;;; consideration. If it is cheaper to test for the difference between the
-;;; derived type and the asserted type, then we check for the negation of this
-;;; type instead.
+;;; This allows us to take what has been proven about CONT's type into
+;;; consideration. If it is cheaper to test for the difference between
+;;; the derived type and the asserted type, then we check for the
+;;; negation of this type instead.
(defun continuation-check-types (cont)
(declare (type continuation cont))
(let ((type (continuation-asserted-type cont))
(t
(values :too-hairy nil))))))
-;;; Return true if Cont is a continuation whose type the back end is likely
-;;; to want to check. Since we don't know what template the back end is going
-;;; to choose to implement the continuation's DEST, we use a heuristic. We
-;;; always return T unless:
-;;; -- Nobody uses the value, or
-;;; -- Safety is totally unimportant, or
+;;; Return true if CONT is a continuation whose type the back end is
+;;; likely to want to check. Since we don't know what template the
+;;; back end is going to choose to implement the continuation's DEST,
+;;; we use a heuristic. We always return T unless:
+;;; -- nobody uses the value, or
+;;; -- safety is totally unimportant, or
;;; -- the continuation is an argument to an unknown function, or
-;;; -- the continuation is an argument to a known function that has no
-;;; IR2-Convert method or :fast-safe templates that are compatible with the
-;;; call's type.
+;;; -- the continuation is an argument to a known function that has
+;;; no IR2-Convert method or :FAST-SAFE templates that are
+;;; compatible with the call's type.
;;;
-;;; We must only return nil when it is *certain* that a check will not be done,
-;;; since if we pass up this chance to do the check, it will be too late. The
-;;; penalty for being too conservative is duplicated type checks.
+;;; We must only return NIL when it is *certain* that a check will not
+;;; be done, since if we pass up this chance to do the check, it will
+;;; be too late. The penalty for being too conservative is duplicated
+;;; type checks.
;;;
-;;; If there is a compile-time type error, then we always return true unless
-;;; the DEST is a full call. With a full call, the theory is that the type
-;;; error is probably from a declaration in (or on) the callee, so the callee
-;;; should be able to do the check. We want to let the callee do the check,
-;;; because it is possible that the error is really in the callee, not the
-;;; caller. We don't want to make people recompile all calls to a function
-;;; when they were originally compiled with a bad declaration (or an old type
-;;; assertion derived from a definition appearing after the call.)
+;;; If there is a compile-time type error, then we always return true
+;;; unless the DEST is a full call. With a full call, the theory is
+;;; that the type error is probably from a declaration in (or on) the
+;;; callee, so the callee should be able to do the check. We want to
+;;; let the callee do the check, because it is possible that the error
+;;; is really in the callee, not the caller. We don't want to make
+;;; people recompile all calls to a function when they were originally
+;;; compiled with a bad declaration (or an old type assertion derived
+;;; from a definition appearing after the call.)
(defun probable-type-check-p (cont)
(declare (type continuation cont))
(let ((dest (continuation-dest cont)))
(t t))))
;;; Return a form that we can convert to do a hairy type check of the
-;;; specified Types. Types is a list of the format returned by
-;;; Continuation-Check-Types in the :HAIRY case. In place of the actual
-;;; value(s) we are to check, we use 'DUMMY. This constant reference is later
-;;; replaced with the actual values continuation.
+;;; specified TYPES. TYPES is a list of the format returned by
+;;; CONTINUATION-CHECK-TYPES in the :HAIRY case. In place of the
+;;; actual value(s) we are to check, we use 'DUMMY. This constant
+;;; reference is later replaced with the actual values continuation.
;;;
-;;; Note that we don't attempt to check for required values being unsupplied.
-;;; Such checking is impossible to efficiently do at the source level because
-;;; our fixed-values conventions are optimized for the common MV-Bind case.
+;;; Note that we don't attempt to check for required values being
+;;; unsupplied. Such checking is impossible to efficiently do at the
+;;; source level because our fixed-values conventions are optimized
+;;; for the common MV-BIND case.
;;;
-;;; We can always use Multiple-Value-Bind, since the macro is clever about
-;;; binding a single variable.
+;;; We can always use MULTIPLE-VALUE-BIND, since the macro is clever
+;;; about binding a single variable.
(defun make-type-check-form (types)
(let ((temps (make-gensym-list (length types))))
`(multiple-value-bind ,temps 'dummy
types)
(values ,@temps))))
-;;; Splice in explicit type check code immediately before the node which is
-;;; Cont's Dest. This code receives the value(s) that were being passed to
-;;; Cont, checks the type(s) of the value(s), then passes them on to Cont.
+;;; Splice in explicit type check code immediately before the node
+;;; which is CONT's DEST. This code receives the value(s) that were
+;;; being passed to CONT, checks the type(s) of the value(s), then
+;;; passes them on to CONT.
(defun convert-type-check (cont types)
(declare (type continuation cont) (type list types))
(with-ir1-environment (continuation-dest cont)
(continuation-starts-block new-start)
(substitute-continuation-uses new-start cont)
- ;; Setting TYPE-CHECK in CONT to :DELETED indicates that the check has
- ;; been done.
+ ;; Setting TYPE-CHECK in CONT to :DELETED indicates that the
+ ;; check has been done.
(setf (continuation-%type-check cont) :deleted)
- ;; Make the DEST node start its block so that we can splice in the
- ;; type check code.
+ ;; Make the DEST node start its block so that we can splice in
+ ;; the type check code.
(when (continuation-use prev)
(node-ends-block (continuation-use prev)))
(new-block (continuation-block new-start))
(dummy (make-continuation)))
- ;; Splice in the new block before DEST, giving the new block all of
- ;; DEST's predecessors.
+ ;; Splice in the new block before DEST, giving the new block
+ ;; all of DEST's predecessors.
(dolist (block (block-pred prev-block))
(change-block-successor block prev-block new-block))
- ;; Convert the check form, using the new block start as START and a
- ;; dummy continuation as CONT.
+ ;; Convert the check form, using the new block start as START
+ ;; and a dummy continuation as CONT.
(ir1-convert new-start dummy (make-type-check-form types))
;; TO DO: Why should this be true? -- WHN 19990601
(assert (eq (continuation-block dummy) new-block))
- ;; KLUDGE: Comments at the head of this function in CMU CL said that
- ;; somewhere in here we
+ ;; KLUDGE: Comments at the head of this function in CMU CL
+ ;; said that somewhere in here we
;; Set the new block's start and end cleanups to the *start*
;; cleanup of PREV's block. This overrides the incorrect
;; default from WITH-IR1-ENVIRONMENT.
(let ((node (continuation-use dummy)))
(setf (block-last new-block) node)
- ;; Change the use to a use of CONT. (We need to use the dummy
- ;; continuation to get the control transfer right, because we want to
- ;; go to PREV's block, not CONT's.)
+ ;; Change the use to a use of CONT. (We need to use the
+ ;; dummy continuation to get the control transfer right,
+ ;; because we want to go to PREV's block, not CONT's.)
(delete-continuation-use node)
(add-continuation-use node cont))
;; Link the new block to PREV's block.
(link-blocks new-block prev-block))
- ;; MAKE-TYPE-CHECK-FORM generated a form which checked the type of
- ;; 'DUMMY, not a real form. At this point we convert to the real form by
- ;; finding 'DUMMY and overwriting it with the new continuation. (We can
- ;; find 'DUMMY because no LET conversion has been done yet.) The
- ;; [mv-]combination code from the mv-bind in the check form will be the
- ;; use of the new check continuation. We substitute for the first
- ;; argument of this node.
+ ;; MAKE-TYPE-CHECK-FORM generated a form which checked the type
+ ;; of 'DUMMY, not a real form. At this point we convert to the
+ ;; real form by finding 'DUMMY and overwriting it with the new
+ ;; continuation. (We can find 'DUMMY because no LET conversion
+ ;; has been done yet.) The [mv-]combination code from the
+ ;; mv-bind in the check form will be the use of the new check
+ ;; continuation. We substitute for the first argument of this
+ ;; node.
(let* ((node (continuation-use cont))
(args (basic-combination-args node))
(victim (first args)))
(values))
-;;; Emit a type warning for Node. If the value of node is being used for a
-;;; variable binding, we figure out which one for source context. If the value
-;;; is a constant, we print it specially. We ignore nodes whose type is NIL,
-;;; since they are supposed to never return.
+;;; Emit a type warning for NODE. If the value of NODE is being used
+;;; for a variable binding, we figure out which one for source
+;;; context. If the value is a constant, we print it specially. We
+;;; ignore nodes whose type is NIL, since they are supposed to never
+;;; return.
(defun do-type-warning (node)
(declare (type node node))
(let* ((*compiler-error-context* node)
what (type-specifier dtype) atype-spec))))
(values))
-;;; Mark Cont as being a continuation with a manifest type error. We set
-;;; the kind to :ERROR, and clear any FUNCTION-INFO if the continuation is an
-;;; argument to a known call. The last is done so that the back end doesn't
-;;; have to worry about type errors in arguments to known functions. This
-;;; clearing is inhibited for things with IR2-CONVERT methods, since we can't
-;;; do a full call to funny functions.
+;;; Mark CONT as being a continuation with a manifest type error. We
+;;; set the kind to :ERROR, and clear any FUNCTION-INFO if the
+;;; continuation is an argument to a known call. The last is done so
+;;; that the back end doesn't have to worry about type errors in
+;;; arguments to known functions. This clearing is inhibited for
+;;; things with IR2-CONVERT methods, since we can't do a full call to
+;;; funny functions.
(defun mark-error-continuation (cont)
(declare (type continuation cont))
(setf (continuation-%type-check cont) :error)
(setf (basic-combination-kind dest) :error)))
(values))
-;;; Loop over all blocks in Component that have TYPE-CHECK set, looking for
-;;; continuations with TYPE-CHECK T. We do two mostly unrelated things: detect
-;;; compile-time type errors and determine if and how to do run-time type
-;;; checks.
+;;; Loop over all blocks in Component that have TYPE-CHECK set,
+;;; looking for continuations with TYPE-CHECK T. We do two mostly
+;;; unrelated things: detect compile-time type errors and determine if
+;;; and how to do run-time type checks.
;;;
-;;; If there is a compile-time type error, then we mark the continuation and
-;;; emit a warning if appropriate. This part loops over all the uses of the
-;;; continuation, since after we convert the check, the :DELETED kind will
-;;; inhibit warnings about the types of other uses.
+;;; If there is a compile-time type error, then we mark the
+;;; continuation and emit a warning if appropriate. This part loops
+;;; over all the uses of the continuation, since after we convert the
+;;; check, the :DELETED kind will inhibit warnings about the types of
+;;; other uses.
;;;
-;;; If a continuation is too complex to be checked by the back end, or is
-;;; better checked with explicit code, then convert to an explicit test.
-;;; Assertions that can checked by the back end are passed through. Assertions
-;;; that can't be tested are flamed about and marked as not needing to be
-;;; checked.
+;;; If a continuation is too complex to be checked by the back end, or
+;;; is better checked with explicit code, then convert to an explicit
+;;; test. Assertions that can checked by the back end are passed
+;;; through. Assertions that can't be tested are flamed about and
+;;; marked as not needing to be checked.
;;;
-;;; If we determine that a type check won't be done, then we set TYPE-CHECK
-;;; to :NO-CHECK. In the non-hairy cases, this is just to prevent us from
-;;; wasting time coming to the same conclusion again on a later iteration. In
-;;; the hairy case, we must indicate to LTN that it must choose a safe
-;;; implementation, since IR2 conversion will choke on the check.
+;;; If we determine that a type check won't be done, then we set
+;;; TYPE-CHECK to :NO-CHECK. In the non-hairy cases, this is just to
+;;; prevent us from wasting time coming to the same conclusion again
+;;; on a later iteration. In the hairy case, we must indicate to LTN
+;;; that it must choose a safe implementation, since IR2 conversion
+;;; will choke on the check.
;;;
;;; The generation of the type checks is delayed until all the type
;;; check decisions have been made because the generation of the type
projects / sbcl.git / blobdiff