;;; FIXME: Doesn't this belong somewhere else, like early-c.lisp?
(declaim (special *constants* *free-vars* *component-being-compiled*
- *code-vector* *next-location* *result-fixups*
*free-funs* *source-paths*
*continuation-number* *continuation-numbers*
*number-continuations* *tn-id* *tn-ids* *id-tns*
;;; WARNINGS-P and FAILURE-P are as in CL:COMPILE or CL:COMPILE-FILE.
;;; This also wraps up WITH-IR1-NAMESPACE functionality.
(defmacro with-compilation-values (&body body)
- `(with-ir1-namespace
- (let ((*warnings-p* nil)
- (*failure-p* nil))
- (values (progn ,@body)
- *warnings-p*
- *failure-p*))))
+ ;; These bindings could just as well be in WITH-IR1-NAMESPACE, but
+ ;; since they're primarily debugging tools, it's nicer to have
+ ;; a wider unique scope by ID.
+ `(let ((*continuation-number* 0)
+ (*continuation-numbers* (make-hash-table :test 'eq))
+ (*number-continuations* (make-hash-table :test 'eql))
+ (*tn-id* 0)
+ (*tn-ids* (make-hash-table :test 'eq))
+ (*id-tns* (make-hash-table :test 'eql))
+ (*label-id* 0)
+ (*label-ids* (make-hash-table :test 'eq))
+ (*id-labels* (make-hash-table :test 'eql)))
+ (unwind-protect
+ (let ((*warnings-p* nil)
+ (*failure-p* nil))
+ (handler-bind ((compiler-error #'compiler-error-handler)
+ (style-warning #'compiler-style-warning-handler)
+ (warning #'compiler-warning-handler))
+ (values (progn ,@body)
+ *warnings-p*
+ *failure-p*)))
+ (clrhash *tn-ids*)
+ (clrhash *id-tns*)
+ (clrhash *continuation-numbers*)
+ (clrhash *number-continuations*)
+ (clrhash *label-ids*)
+ (clrhash *id-labels*))))
\f
;;;; component compilation
(defparameter *max-optimize-iterations* 3 ; ARB
#!+sb-doc
"The upper limit on the number of times that we will consecutively do IR1
- optimization that doesn't introduce any new code. A finite limit is
- necessary, since type inference may take arbitrarily long to converge.")
+optimization that doesn't introduce any new code. A finite limit is
+necessary, since type inference may take arbitrarily long to converge.")
(defevent ir1-optimize-until-done "IR1-OPTIMIZE-UNTIL-DONE called")
(defevent ir1-optimize-maxed-out "hit *MAX-OPTIMIZE-ITERATIONS* limit")
(return))
(incf loop-count)))
+ (when *check-consistency*
+ (do-blocks-backwards (block component)
+ (awhen (flush-dead-code block)
+ (let ((*compiler-error-context* it))
+ (compiler-warn "dead code detected at the end of ~S"
+ 'ir1-phases)))))
+
(ir1-finalize component)
(values))
(blast *free-funs*)
(blast *constants*))
(values))
-
-;;; Clear global variables used by the compiler.
-;;;
-;;; FIXME: It seems kinda nasty and unmaintainable to have to do this,
-;;; and it adds overhead even when people aren't using the compiler.
-;;; Perhaps we could make these global vars unbound except when
-;;; actually in use, so that this function could go away.
-(defun clear-stuff (&optional (debug-too t))
-
- (when debug-too
- (clrhash *continuation-numbers*)
- (clrhash *number-continuations*)
- (setq *continuation-number* 0)
- (clrhash *tn-ids*)
- (clrhash *id-tns*)
- (setq *tn-id* 0)
- (clrhash *label-ids*)
- (clrhash *id-labels*)
- (setq *label-id* 0))
-
- ;; (Note: The CMU CL code used to set CL::*GENSYM-COUNTER* to zero here.
- ;; Superficially, this seemed harmful -- the user could reasonably be
- ;; surprised if *GENSYM-COUNTER* turned back to zero when something was
- ;; compiled. A closer inspection showed that this actually turned out to be
- ;; harmless in practice, because CLEAR-STUFF was only called from within
- ;; forms which bound CL::*GENSYM-COUNTER* to zero. However, this means that
- ;; even though zeroing CL::*GENSYM-COUNTER* here turned out to be harmless in
- ;; practice, it was also useless in practice. So we don't do it any more.)
-
- (values))
\f
;;;; trace output
;;; Read and compile the source file.
(defun sub-sub-compile-file (info)
(do-forms-from-info ((form current-index) info)
- (find-source-paths form current-index)
- (process-toplevel-form
- form `(original-source-start 0 ,current-index) nil)))
+ (with-source-paths
+ (find-source-paths form current-index)
+ (process-toplevel-form
+ form `(original-source-start 0 ,current-index) nil))))
;;; Return the INDEX'th source form read from INFO and the position
;;; where it was read.
(if (fopcompilable-p form)
(let ((*fopcompile-label-counter* 0))
(fopcompile form path nil))
- (let ((*lexenv* (make-lexenv
- :policy *policy*
- :handled-conditions *handled-conditions*
- :disabled-package-locks *disabled-package-locks*))
- (tll (ir1-toplevel form path nil)))
- (if (eq *block-compile* t)
- (push tll *toplevel-lambdas*)
- (compile-toplevel (list tll) nil))
- nil)))))
+ (with-ir1-namespace
+ (let ((*lexenv* (make-lexenv
+ :policy *policy*
+ :handled-conditions *handled-conditions*
+ :disabled-package-locks *disabled-package-locks*))
+ (tll (ir1-toplevel form path nil)))
+ (if (eq *block-compile* t)
+ (push tll *toplevel-lambdas*)
+ (compile-toplevel (list tll) nil))
+ nil))))))
;;; Macroexpand FORM in the current environment with an error handler.
;;; We only expand one level, so that we retain all the intervening
(declare (list path))
(multiple-value-bind (forms decls)
(parse-body body :doc-string-allowed nil :toplevel t)
- (let* ((*lexenv* (process-decls decls vars funs))
- ;; FIXME: VALUES declaration
- ;;
- ;; Binding *POLICY* is pretty much of a hack, since it
- ;; causes LOCALLY to "capture" enclosed proclamations. It
- ;; is necessary because CONVERT-AND-MAYBE-COMPILE uses the
- ;; value of *POLICY* as the policy. The need for this hack
- ;; is due to the quirk that there is no way to represent in
- ;; a POLICY that an optimize quality came from the default.
- ;;
- ;; FIXME: Ideally, something should be done so that DECLAIM
- ;; inside LOCALLY works OK. Failing that, at least we could
- ;; issue a warning instead of silently screwing up.
- (*policy* (lexenv-policy *lexenv*))
- ;; This is probably also a hack
- (*handled-conditions* (lexenv-handled-conditions *lexenv*))
- ;; ditto
- (*disabled-package-locks* (lexenv-disabled-package-locks *lexenv*)))
- (process-toplevel-progn forms path compile-time-too))))
+ (with-ir1-namespace
+ (let* ((*lexenv* (process-decls decls vars funs))
+ ;; FIXME: VALUES declaration
+ ;;
+ ;; Binding *POLICY* is pretty much of a hack, since it
+ ;; causes LOCALLY to "capture" enclosed proclamations. It
+ ;; is necessary because CONVERT-AND-MAYBE-COMPILE uses the
+ ;; value of *POLICY* as the policy. The need for this hack
+ ;; is due to the quirk that there is no way to represent in
+ ;; a POLICY that an optimize quality came from the default.
+ ;;
+ ;; FIXME: Ideally, something should be done so that DECLAIM
+ ;; inside LOCALLY works OK. Failing that, at least we could
+ ;; issue a warning instead of silently screwing up.
+ (*policy* (lexenv-policy *lexenv*))
+ ;; This is probably also a hack
+ (*handled-conditions* (lexenv-handled-conditions *lexenv*))
+ ;; ditto
+ (*disabled-package-locks* (lexenv-disabled-package-locks *lexenv*)))
+ (process-toplevel-progn forms path compile-time-too)))))
;;; Parse an EVAL-WHEN situations list, returning three flags,
;;; (VALUES COMPILE-TOPLEVEL LOAD-TOPLEVEL EXECUTE), indicating
'(original-source-start 0 0)))
(when name
(legal-fun-name-or-type-error name))
- (let* ((*lexenv* (make-lexenv
- :policy *policy*
- :handled-conditions *handled-conditions*
- :disabled-package-locks *disabled-package-locks*))
- (*compiler-sset-counter* 0)
- (fun (make-functional-from-toplevel-lambda lambda-expression
- :name name
- :path path)))
-
- ;; FIXME: The compile-it code from here on is sort of a
- ;; twisted version of the code in COMPILE-TOPLEVEL. It'd be
- ;; better to find a way to share the code there; or
- ;; alternatively, to use this code to replace the code there.
- ;; (The second alternative might be pretty easy if we used
- ;; the :LOCALL-ONLY option to IR1-FOR-LAMBDA. Then maybe the
- ;; whole FUNCTIONAL-KIND=:TOPLEVEL case could go away..)
-
- (locall-analyze-clambdas-until-done (list fun))
-
- (let ((components-from-dfo (find-initial-dfo (list fun))))
- (dolist (component-from-dfo components-from-dfo)
- (compile-component component-from-dfo)
- (replace-toplevel-xeps component-from-dfo))
-
- (let ((entry-table (etypecase *compile-object*
- (fasl-output (fasl-output-entry-table
- *compile-object*))
- (core-object (core-object-entry-table
- *compile-object*)))))
- (multiple-value-bind (result found-p)
- (gethash (leaf-info fun) entry-table)
- (aver found-p)
- (prog1
- result
- ;; KLUDGE: This code duplicates some other code in this
- ;; file. In the great reorganzation, the flow of program
- ;; logic changed from the original CMUCL model, and that
- ;; path (as of sbcl-0.7.5 in SUB-COMPILE-FILE) was no
- ;; longer followed for CORE-OBJECTS, leading to BUG
- ;; 156. This place is transparently not the right one for
- ;; this code, but I don't have a clear enough overview of
- ;; the compiler to know how to rearrange it all so that
- ;; this operation fits in nicely, and it was blocking
- ;; reimplementation of (DECLAIM (INLINE FOO)) (MACROLET
- ;; ((..)) (DEFUN FOO ...))
- ;;
- ;; FIXME: This KLUDGE doesn't solve all the problem in an
- ;; ideal way, as (1) definitions typed in at the REPL
- ;; without an INLINE declaration will give a NULL
- ;; FUNCTION-LAMBDA-EXPRESSION (allowable, but not ideal)
- ;; and (2) INLINE declarations will yield a
- ;; FUNCTION-LAMBDA-EXPRESSION headed by
- ;; SB-C:LAMBDA-WITH-LEXENV, even for null LEXENV. -- CSR,
- ;; 2002-07-02
- ;;
- ;; (2) is probably fairly easy to fix -- it is, after all,
- ;; a matter of list manipulation (or possibly of teaching
- ;; CL:FUNCTION about SB-C:LAMBDA-WITH-LEXENV). (1) is
- ;; significantly harder, as the association between
- ;; function object and source is a tricky one.
- ;;
- ;; FUNCTION-LAMBDA-EXPRESSION "works" (i.e. returns a
- ;; non-NULL list) when the function in question has been
- ;; compiled by (COMPILE <x> '(LAMBDA ...)); it does not
- ;; work when it has been compiled as part of the top-level
- ;; EVAL strategy of compiling everything inside (LAMBDA ()
- ;; ...). -- CSR, 2002-11-02
- (when (core-object-p *compile-object*)
- (fix-core-source-info *source-info* *compile-object* result))
-
- (mapc #'clear-ir1-info components-from-dfo)
- (clear-stuff)))))))
+ (with-ir1-namespace
+ (let* ((*lexenv* (make-lexenv
+ :policy *policy*
+ :handled-conditions *handled-conditions*
+ :disabled-package-locks *disabled-package-locks*))
+ (*compiler-sset-counter* 0)
+ (fun (make-functional-from-toplevel-lambda lambda-expression
+ :name name
+ :path path)))
+
+ ;; FIXME: The compile-it code from here on is sort of a
+ ;; twisted version of the code in COMPILE-TOPLEVEL. It'd be
+ ;; better to find a way to share the code there; or
+ ;; alternatively, to use this code to replace the code there.
+ ;; (The second alternative might be pretty easy if we used
+ ;; the :LOCALL-ONLY option to IR1-FOR-LAMBDA. Then maybe the
+ ;; whole FUNCTIONAL-KIND=:TOPLEVEL case could go away..)
+
+ (locall-analyze-clambdas-until-done (list fun))
+
+ (let ((components-from-dfo (find-initial-dfo (list fun))))
+ (dolist (component-from-dfo components-from-dfo)
+ (compile-component component-from-dfo)
+ (replace-toplevel-xeps component-from-dfo))
+
+ (let ((entry-table (etypecase *compile-object*
+ (fasl-output (fasl-output-entry-table
+ *compile-object*))
+ (core-object (core-object-entry-table
+ *compile-object*)))))
+ (multiple-value-bind (result found-p)
+ (gethash (leaf-info fun) entry-table)
+ (aver found-p)
+ (prog1
+ result
+ ;; KLUDGE: This code duplicates some other code in this
+ ;; file. In the great reorganzation, the flow of program
+ ;; logic changed from the original CMUCL model, and that
+ ;; path (as of sbcl-0.7.5 in SUB-COMPILE-FILE) was no
+ ;; longer followed for CORE-OBJECTS, leading to BUG
+ ;; 156. This place is transparently not the right one for
+ ;; this code, but I don't have a clear enough overview of
+ ;; the compiler to know how to rearrange it all so that
+ ;; this operation fits in nicely, and it was blocking
+ ;; reimplementation of (DECLAIM (INLINE FOO)) (MACROLET
+ ;; ((..)) (DEFUN FOO ...))
+ ;;
+ ;; FIXME: This KLUDGE doesn't solve all the problem in an
+ ;; ideal way, as (1) definitions typed in at the REPL
+ ;; without an INLINE declaration will give a NULL
+ ;; FUNCTION-LAMBDA-EXPRESSION (allowable, but not ideal)
+ ;; and (2) INLINE declarations will yield a
+ ;; FUNCTION-LAMBDA-EXPRESSION headed by
+ ;; SB-C:LAMBDA-WITH-LEXENV, even for null LEXENV. -- CSR,
+ ;; 2002-07-02
+ ;;
+ ;; (2) is probably fairly easy to fix -- it is, after all,
+ ;; a matter of list manipulation (or possibly of teaching
+ ;; CL:FUNCTION about SB-C:LAMBDA-WITH-LEXENV). (1) is
+ ;; significantly harder, as the association between
+ ;; function object and source is a tricky one.
+ ;;
+ ;; FUNCTION-LAMBDA-EXPRESSION "works" (i.e. returns a
+ ;; non-NULL list) when the function in question has been
+ ;; compiled by (COMPILE <x> '(LAMBDA ...)); it does not
+ ;; work when it has been compiled as part of the top-level
+ ;; EVAL strategy of compiling everything inside (LAMBDA ()
+ ;; ...). -- CSR, 2002-11-02
+ (when (core-object-p *compile-object*)
+ (fix-core-source-info *source-info* *compile-object* result))
+
+ (mapc #'clear-ir1-info components-from-dfo))))))))
(defun process-toplevel-cold-fset (name lambda-expression path)
(unless (producing-fasl-file)
(compile-load-time-value-lambda lambdas)
(compile-toplevel-lambdas lambdas))
- (mapc #'clear-ir1-info components)
- (clear-stuff)))
+ (mapc #'clear-ir1-info components)))
(values))
;;; Actually compile any stuff that has been queued up for block
(with-compilation-values
(sb!xc:with-compilation-unit ()
(with-world-lock ()
- (clear-stuff)
(sub-sub-compile-file info)
(unless (zerop (hash-table-count *code-coverage-records*))
;; Dump the code coverage records into the fasl.
- (fopcompile `(record-code-coverage
- ',(namestring *compile-file-pathname*)
- ',(let (list)
- (maphash (lambda (k v)
- (declare (ignore k))
- (push v list))
- *code-coverage-records*)
- list))
- nil
- nil))
+ (with-source-paths
+ (fopcompile `(record-code-coverage
+ ',(namestring *compile-file-pathname*)
+ ',(let (list)
+ (maphash (lambda (k v)
+ (declare (ignore k))
+ (push v list))
+ *code-coverage-records*)
+ list))
+ nil
+ nil)))
(finish-block-compilation)
(let ((object *compile-object*))
(etypecase object