guile/module/language/tree-il/analyze.scm

713 lines
30 KiB
Scheme
Raw Normal View History

;;; TREE-IL -> GLIL compiler
;; Copyright (C) 2001,2008,2009 Free Software Foundation, Inc.
;;;; This library is free software; you can redistribute it and/or
;;;; modify it under the terms of the GNU Lesser General Public
;;;; License as published by the Free Software Foundation; either
;;;; version 3 of the License, or (at your option) any later version.
;;;;
;;;; This library is distributed in the hope that it will be useful,
;;;; but WITHOUT ANY WARRANTY; without even the implied warranty of
;;;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
;;;; Lesser General Public License for more details.
;;;;
;;;; You should have received a copy of the GNU Lesser General Public
;;;; License along with this library; if not, write to the Free Software
;;;; Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;;; Code:
(define-module (language tree-il analyze)
#:use-module (srfi srfi-1)
#:use-module (srfi srfi-9)
#:use-module (system base syntax)
#:use-module (system base message)
#:use-module (language tree-il)
#:export (analyze-lexicals
report-unused-variables
report-possibly-unbound-variables))
;; Allocation is the process of assigning storage locations for lexical
;; variables. A lexical variable has a distinct "address", or storage
;; location, for each procedure in which it is referenced.
;;
;; A variable is "local", i.e., allocated on the stack, if it is
;; referenced from within the procedure that defined it. Otherwise it is
;; a "closure" variable. For example:
;;
;; (lambda (a) a) ; a will be local
;; `a' is local to the procedure.
;;
;; (lambda (a) (lambda () a))
;; `a' is local to the outer procedure, but a closure variable with
;; respect to the inner procedure.
;;
;; If a variable is ever assigned, it needs to be heap-allocated
;; ("boxed"). This is so that closures and continuations capture the
;; variable's identity, not just one of the values it may have over the
;; course of program execution. If the variable is never assigned, there
;; is no distinction between value and identity, so closing over its
;; identity (whether through closures or continuations) can make a copy
;; of its value instead.
;;
;; Local variables are stored on the stack within a procedure's call
;; frame. Their index into the stack is determined from their linear
;; postion within a procedure's binding path:
;; (let (0 1)
;; (let (2 3) ...)
;; (let (2) ...))
;; (let (2 3 4) ...))
;; etc.
;;
;; This algorithm has the problem that variables are only allocated
;; indices at the end of the binding path. If variables bound early in
;; the path are not used in later portions of the path, their indices
;; will not be recycled. This problem is particularly egregious in the
;; expansion of `or':
;;
;; (or x y z)
;; -> (let ((a x)) (if a a (let ((b y)) (if b b z))))
;;
;; As you can see, the `a' binding is only used in the ephemeral `then'
;; clause of the first `if', but its index would be reserved for the
;; whole of the `or' expansion. So we have a hack for this specific
;; case. A proper solution would be some sort of liveness analysis, and
;; not our linear allocation algorithm.
;;
;; Closure variables are captured when a closure is created, and stored
;; in a vector. Each closure variable has a unique index into that
;; vector.
;;
;; There is one more complication. Procedures bound by <fix> may, in
;; some cases, be rendered inline to their parent procedure. That is to
;; say,
;;
;; (letrec ((lp (lambda () (lp)))) (lp))
;; => (fix ((lp (lambda () (lp)))) (lp))
;; => goto FIX-BODY; LP: goto LP; FIX-BODY: goto LP;
;; ^ jump over the loop ^ the fixpoint lp ^ starting off the loop
;;
;; The upshot is that we don't have to allocate any space for the `lp'
;; closure at all, as it can be rendered inline as a loop. So there is
;; another kind of allocation, "label allocation", in which the
;; procedure is simply a label, placed at the start of the lambda body.
;; The label is the gensym under which the lambda expression is bound.
;;
;; The analyzer checks to see that the label is called with the correct
;; number of arguments. Calls to labels compile to rename + goto.
;; Lambda, the ultimate goto!
;;
;;
;; The return value of `analyze-lexicals' is a hash table, the
;; "allocation".
;;
;; The allocation maps gensyms -- recall that each lexically bound
;; variable has a unique gensym -- to storage locations ("addresses").
;; Since one gensym may have many storage locations, if it is referenced
;; in many procedures, it is a two-level map.
;;
;; The allocation also stored information on how many local variables
;; need to be allocated for each procedure, lexicals that have been
;; translated into labels, and information on what free variables to
;; capture from its lexical parent procedure.
;;
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; In addition, we have a conflation: while we're traversing the code,
;; recording information to pass to the compiler, we take the
;; opportunity to generate labels for each lambda-case clause, so that
;; generated code can skip argument checks at runtime if they match at
;; compile-time.
;;
;; That is:
;;
;; sym -> {lambda -> address}
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; lambda -> (labels . free-locs)
;; lambda-case -> (gensym . nlocs)
;;
;; address ::= (local? boxed? . index)
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; labels ::= ((sym . lambda) ...)
;; free-locs ::= ((sym0 . address0) (sym1 . address1) ...)
;; free variable addresses are relative to parent proc.
(define (make-hashq k v)
(let ((res (make-hash-table)))
(hashq-set! res k v)
res))
(define (analyze-lexicals x)
;; bound-vars: lambda -> (sym ...)
;; all identifiers bound within a lambda
(define bound-vars (make-hash-table))
;; free-vars: lambda -> (sym ...)
;; all identifiers referenced in a lambda, but not bound
;; NB, this includes identifiers referenced by contained lambdas
(define free-vars (make-hash-table))
;; assigned: sym -> #t
;; variables that are assigned
(define assigned (make-hash-table))
;; refcounts: sym -> count
;; allows us to detect the or-expansion in O(1) time
(define refcounts (make-hash-table))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; labels: sym -> lambda
;; for determining if fixed-point procedures can be rendered as
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; labels.
(define labels (make-hash-table))
;; returns variables referenced in expr
(define (analyze! x proc labels-in-proc tail? tail-call-args)
(define (step y) (analyze! y proc labels-in-proc #f #f))
(define (step-tail y) (analyze! y proc labels-in-proc tail? #f))
(define (step-tail-call y args) (analyze! y proc labels-in-proc #f
(and tail? args)))
(define (recur/labels x new-proc labels)
(analyze! x new-proc (append labels labels-in-proc) #t #f))
(define (recur x new-proc) (analyze! x new-proc '() tail? #f))
(record-case x
((<application> proc args)
(apply lset-union eq? (step-tail-call proc args)
(map step args)))
((<conditional> test then else)
(lset-union eq? (step test) (step-tail then) (step-tail else)))
((<lexical-ref> gensym)
(hashq-set! refcounts gensym (1+ (hashq-ref refcounts gensym 0)))
(if (not (and tail-call-args
(memq gensym labels-in-proc)
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
(let ((p (hashq-ref labels gensym)))
(and p
(let lp ((c (lambda-body p)))
(and c (lambda-case? c)
(or
;; for now prohibit optional &
;; keyword arguments; can relax this
;; restriction later
(and (= (length (lambda-case-req c))
(length tail-call-args))
(not (lambda-case-opt c))
(not (lambda-case-kw c))
(not (lambda-case-rest c))
(not (lambda-case-predicate c)))
(lp (lambda-case-else c)))))))))
(hashq-set! labels gensym #f))
(list gensym))
((<lexical-set> gensym exp)
(hashq-set! assigned gensym #t)
(hashq-set! labels gensym #f)
(lset-adjoin eq? (step exp) gensym))
((<module-set> exp)
(step exp))
((<toplevel-set> exp)
(step exp))
((<toplevel-define> exp)
(step exp))
((<sequence> exps)
(let lp ((exps exps) (ret '()))
(cond ((null? exps) '())
((null? (cdr exps))
(lset-union eq? ret (step-tail (car exps))))
(else
(lp (cdr exps) (lset-union eq? ret (step (car exps))))))))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda> body)
;; order is important here
(hashq-set! bound-vars x '())
(let ((free (recur body x)))
(hashq-set! bound-vars x (reverse! (hashq-ref bound-vars x)))
(hashq-set! free-vars x free)
free))
((<lambda-case> vars predicate body else)
(hashq-set! bound-vars proc
(append (reverse vars) (hashq-ref bound-vars proc)))
(lset-union
eq?
(lset-difference eq?
(lset-union eq? (if predicate (step predicate) '())
(step-tail body))
vars)
(if else (step-tail else) '())))
((<let> vars vals body)
(hashq-set! bound-vars proc
(append (reverse vars) (hashq-ref bound-vars proc)))
(lset-difference eq?
(apply lset-union eq? (step-tail body) (map step vals))
vars))
((<letrec> vars vals body)
(hashq-set! bound-vars proc
(append (reverse vars) (hashq-ref bound-vars proc)))
(for-each (lambda (sym) (hashq-set! assigned sym #t)) vars)
(lset-difference eq?
(apply lset-union eq? (step-tail body) (map step vals))
vars))
((<fix> vars vals body)
;; Try to allocate these procedures as labels.
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
(for-each (lambda (sym val) (hashq-set! labels sym val))
vars vals)
(hashq-set! bound-vars proc
(append (reverse vars) (hashq-ref bound-vars proc)))
;; Step into subexpressions.
(let* ((var-refs
(map
;; Since we're trying to label-allocate the lambda,
;; pretend it's not a closure, and just recurse into its
;; body directly. (Otherwise, recursing on a closure
;; that references one of the fix's bound vars would
;; prevent label allocation.)
(lambda (x)
(record-case x
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda> body)
;; just like the closure case, except here we use
;; recur/labels instead of recur
(hashq-set! bound-vars x '())
(let ((free (recur/labels body x vars)))
(hashq-set! bound-vars x (reverse! (hashq-ref bound-vars x)))
(hashq-set! free-vars x free)
free))))
vals))
(vars-with-refs (map cons vars var-refs))
(body-refs (recur/labels body proc vars)))
(define (delabel-dependents! sym)
(let ((refs (assq-ref vars-with-refs sym)))
(if refs
(for-each (lambda (sym)
(if (hashq-ref labels sym)
(begin
(hashq-set! labels sym #f)
(delabel-dependents! sym))))
refs))))
;; Stepping into the lambdas and the body might have made some
;; procedures not label-allocatable -- which might have
;; knock-on effects. For example:
;; (fix ((a (lambda () (b)))
;; (b (lambda () a)))
;; (a))
;; As far as `a' is concerned, both `a' and `b' are
;; label-allocatable. But `b' references `a' not in a proc-tail
;; position, which makes `a' not label-allocatable. The
;; knock-on effect is that, when back-propagating this
;; information to `a', `b' will also become not
;; label-allocatable, as it is referenced within `a', which is
;; allocated as a closure. This is a transitive relationship.
(for-each (lambda (sym)
(if (not (hashq-ref labels sym))
(delabel-dependents! sym)))
vars)
;; Now lift bound variables with label-allocated lambdas to the
;; parent procedure.
(for-each
(lambda (sym val)
(if (hashq-ref labels sym)
;; Remove traces of the label-bound lambda. The free
;; vars will propagate up via the return val.
(begin
(hashq-set! bound-vars proc
(append (hashq-ref bound-vars val)
(hashq-ref bound-vars proc)))
(hashq-remove! bound-vars val)
(hashq-remove! free-vars val))))
vars vals)
(lset-difference eq?
(apply lset-union eq? body-refs var-refs)
vars)))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<let-values> exp body)
(lset-union eq? (step exp) (step body)))
(else '())))
;; allocation: sym -> {lambda -> address}
;; lambda -> (nlocs labels . free-locs)
(define allocation (make-hash-table))
(define (allocate! x proc n)
(define (recur y) (allocate! y proc n))
(record-case x
((<application> proc args)
(apply max (recur proc) (map recur args)))
((<conditional> test then else)
(max (recur test) (recur then) (recur else)))
((<lexical-set> exp)
(recur exp))
((<module-set> exp)
(recur exp))
((<toplevel-set> exp)
(recur exp))
((<toplevel-define> exp)
(recur exp))
((<sequence> exps)
(apply max (map recur exps)))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda> body)
;; allocate closure vars in order
(let lp ((c (hashq-ref free-vars x)) (n 0))
(if (pair? c)
(begin
(hashq-set! (hashq-ref allocation (car c))
x
`(#f ,(hashq-ref assigned (car c)) . ,n))
(lp (cdr c) (1+ n)))))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
(let ((nlocs (allocate! body x 0))
(free-addresses
(map (lambda (v)
(hashq-ref (hashq-ref allocation v) proc))
(hashq-ref free-vars x)))
(labels (filter cdr
(map (lambda (sym)
(cons sym (hashq-ref labels sym)))
(hashq-ref bound-vars x)))))
;; set procedure allocations
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
(hashq-set! allocation x (cons labels free-addresses)))
n)
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda-case> vars predicate body else)
(max
(let lp ((vars vars) (n n))
(if (null? vars)
(let ((nlocs (max (if predicate (allocate! predicate body n) n)
(allocate! body proc n))))
;; label and nlocs for the case
(hashq-set! allocation x (cons (gensym ":LCASE") nlocs))
nlocs)
(begin
(hashq-set! allocation (car vars)
(make-hashq
proc `(#t ,(hashq-ref assigned (car vars)) . ,n)))
(lp (cdr vars) (1+ n)))))
(if else (allocate! else proc n) n)))
((<let> vars vals body)
(let ((nmax (apply max (map recur vals))))
(cond
;; the `or' hack
((and (conditional? body)
(= (length vars) 1)
(let ((v (car vars)))
(and (not (hashq-ref assigned v))
(= (hashq-ref refcounts v 0) 2)
(lexical-ref? (conditional-test body))
(eq? (lexical-ref-gensym (conditional-test body)) v)
(lexical-ref? (conditional-then body))
(eq? (lexical-ref-gensym (conditional-then body)) v))))
(hashq-set! allocation (car vars)
(make-hashq proc `(#t #f . ,n)))
;; the 1+ for this var
(max nmax (1+ n) (allocate! (conditional-else body) proc n)))
(else
(let lp ((vars vars) (n n))
(if (null? vars)
(max nmax (allocate! body proc n))
(let ((v (car vars)))
(hashq-set!
allocation v
(make-hashq proc
`(#t ,(hashq-ref assigned v) . ,n)))
(lp (cdr vars) (1+ n)))))))))
((<letrec> vars vals body)
(let lp ((vars vars) (n n))
(if (null? vars)
(let ((nmax (apply max
(map (lambda (x)
(allocate! x proc n))
vals))))
(max nmax (allocate! body proc n)))
(let ((v (car vars)))
(hashq-set!
allocation v
(make-hashq proc
`(#t ,(hashq-ref assigned v) . ,n)))
(lp (cdr vars) (1+ n))))))
((<fix> vars vals body)
(let lp ((in vars) (n n))
(if (null? in)
(let lp ((vars vars) (vals vals) (nmax n))
(cond
((null? vars)
(max nmax (allocate! body proc n)))
((hashq-ref labels (car vars))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; allocate lambda body inline to proc
(lp (cdr vars)
(cdr vals)
(record-case (car vals)
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda> body)
(max nmax (allocate! body proc n))))))
(else
;; allocate closure
(lp (cdr vars)
(cdr vals)
(max nmax (allocate! (car vals) proc n))))))
(let ((v (car in)))
(cond
((hashq-ref assigned v)
(error "fixpoint procedures may not be assigned" x))
((hashq-ref labels v)
;; no binding, it's a label
(lp (cdr in) n))
(else
;; allocate closure binding
(hashq-set! allocation v (make-hashq proc `(#t #f . ,n)))
(lp (cdr in) (1+ n))))))))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<let-values> exp body)
(max (recur exp) (recur body)))
(else n)))
(analyze! x #f '() #t #f)
(allocate! x #f 0)
allocation)
;;;
;;; Unused variable analysis.
;;;
;; <binding-info> records are used during tree traversals in
;; `report-unused-variables'. They contain a list of the local vars
;; currently in scope, a list of locals vars that have been referenced, and a
;; "location stack" (the stack of `tree-il-src' values for each parent tree).
(define-record-type <binding-info>
(make-binding-info vars refs locs)
binding-info?
(vars binding-info-vars) ;; ((GENSYM NAME LOCATION) ...)
(refs binding-info-refs) ;; (GENSYM ...)
(locs binding-info-locs)) ;; (LOCATION ...)
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
;; FIXME!!
(define (report-unused-variables tree env)
"Report about unused variables in TREE. Return TREE."
(tree-il-fold (lambda (x info)
;; X is a leaf: extend INFO's refs accordingly.
(let ((refs (binding-info-refs info))
(vars (binding-info-vars info))
(locs (binding-info-locs info)))
(record-case x
((<lexical-ref> gensym)
(make-binding-info vars (cons gensym refs) locs))
(else info))))
(lambda (x info)
;; Going down into X: extend INFO's variable list
;; accordingly.
(let ((refs (binding-info-refs info))
(vars (binding-info-vars info))
(locs (binding-info-locs info))
(src (tree-il-src x)))
(define (extend inner-vars inner-names)
(append (map (lambda (var name)
(list var name src))
inner-vars
inner-names)
vars))
(record-case x
((<lexical-set> gensym)
(make-binding-info vars (cons gensym refs)
(cons src locs)))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda-case> req opt rest kw vars)
;; FIXME keywords.
(let ((names `(,@req ,@(or opt '()) . ,(or rest '()))))
(make-binding-info (extend vars names) refs
(cons src locs))))
((<let> vars names)
(make-binding-info (extend vars names) refs
(cons src locs)))
((<letrec> vars names)
(make-binding-info (extend vars names) refs
(cons src locs)))
((<fix> vars names)
(make-binding-info (extend vars names) refs
(cons src locs)))
(else info))))
(lambda (x info)
;; Leaving X's scope: shrink INFO's variable list
;; accordingly and reported unused nested variables.
(let ((refs (binding-info-refs info))
(vars (binding-info-vars info))
(locs (binding-info-locs info)))
(define (shrink inner-vars refs)
(for-each (lambda (var)
(let ((gensym (car var)))
;; Don't report lambda parameters as
;; unused.
(if (and (not (memq gensym refs))
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
(not (and (lambda-case? x)
(memq gensym
inner-vars))))
(let ((name (cadr var))
;; We can get approximate
;; source location by going up
;; the LOCS location stack.
(loc (or (caddr var)
(find pair? locs))))
(warning 'unused-variable loc name)))))
(filter (lambda (var)
(memq (car var) inner-vars))
vars))
(fold alist-delete vars inner-vars))
;; For simplicity, we leave REFS untouched, i.e., with
;; names of variables that are now going out of scope.
;; It doesn't hurt as these are unique names, it just
;; makes REFS unnecessarily fat.
(record-case x
tree-il support for case-lambda * module/language/tree-il.scm (<lambda>, <lambda-case>): Split lambda into the lambda itself, denoting the procedure, and lambda-case, denoting a particular arity case. Lambda-case is fairly featureful, and has not yet been fully tested. (<let-values>): Use a <lambda-case> as the binding expression. Seems to suit the purpose well. Adapt parsers, unparsers, traversal operators, etc. Sometimes in this first version we assume there are no optional args, rest args, or a predicate. * module/language/tree-il/analyze.scm (analyze-lexicals): Adapt for the new case-lambda regime. Fairly well commented. It actually simplifies things. (report-unused-variables): Update for new tree-il. * module/language/tree-il/compile-glil.scm: Adapt for the new tree-il. There are some first stabs here at proper case-lambda compilation, but they are untested as of yet. * module/language/tree-il/inline.scm (inline!): Rework so we can recurse on a single node; though these transformations are strictly reductive, so they should complete in bounded time. Simplify accordingly, and adapt to case-lambda. Oh, and we handle lambda->let in not just the nullary case. * module/ice-9/psyntax.scm (build-simple-lambda, build-case-lambda) (build-lambda-case): New constructors. The idea is that after syntax expansion, we shouldn't have to deal with improper lists any more. Build-simple-lambda is a shortcut for the common case. The others are not fully exercised yet. Adapt callers. (syntax): Add some debugging in the lambda case. I don't fully understand this, but in practice we don't seem to see rest args here. (lambda): Inline chi-lambda-clause, and adapt for build-simple-lambda. * module/ice-9/psyntax-pp.scm: Regenerated. * test-suite/tests/tree-il.test: Update tests for new tree-il lambda format, and to expect post-prelude labels for all glil programs.
2009-10-14 00:08:35 +02:00
((<lambda-case> vars)
(make-binding-info (shrink vars refs) refs
(cdr locs)))
((<let> vars)
(make-binding-info (shrink vars refs) refs
(cdr locs)))
((<letrec> vars)
(make-binding-info (shrink vars refs) refs
(cdr locs)))
((<fix> vars)
(make-binding-info (shrink vars refs) refs
(cdr locs)))
(else info))))
(make-binding-info '() '() '())
tree)
tree)
;;;
;;; Unbound variable analysis.
;;;
;; <toplevel-info> records are used during tree traversal in search of
;; possibly unbound variable. They contain a list of references to
;; potentially unbound top-level variables, a list of the top-level defines
;; that have been encountered, and a "location stack" (see above).
(define-record-type <toplevel-info>
(make-toplevel-info refs defs locs)
toplevel-info?
(refs toplevel-info-refs) ;; ((VARIABLE-NAME . LOCATION) ...)
(defs toplevel-info-defs) ;; (VARIABLE-NAME ...)
(locs toplevel-info-locs)) ;; (LOCATION ...)
(define (goops-toplevel-definition proc args env)
;; If application of PROC to ARGS is a GOOPS top-level definition, return
;; the name of the variable being defined; otherwise return #f. This
;; assumes knowledge of the current implementation of `define-class' et al.
(define (toplevel-define-arg args)
(and (pair? args) (pair? (cdr args)) (null? (cddr args))
(record-case (car args)
((<const> exp)
(and (symbol? exp) exp))
(else #f))))
(record-case proc
((<module-ref> mod public? name)
(and (equal? mod '(oop goops))
(not public?)
(eq? name 'toplevel-define!)
(toplevel-define-arg args)))
((<toplevel-ref> name)
;; This may be the result of expanding one of the GOOPS macros within
;; `oop/goops.scm'.
(and (eq? name 'toplevel-define!)
(eq? env (resolve-module '(oop goops)))
(toplevel-define-arg args)))
(else #f)))
;; TODO: Combine with `report-unused-variables' so we don't traverse the tree
;; once for each warning type.
(define (report-possibly-unbound-variables tree env)
"Return possibly unbound variables in TREE. Return TREE."
(define toplevel
compilation enviroments are always modules; simplifications & refactorings * module/ice-9/boot-9.scm (make-fresh-user-module): New public function, makes an anonymous beautified module. * module/language/objcode/spec.scm: We used to have some things in here that allowed lexical variable names and values to be a part of the environment, but no more. Now an environment is just a module. If you want to "inject" free variables into code, just use lambda. * module/language/scheme/compile-tree-il.scm (compile-tree-il): Same here. Also, rely on the fact that an environment *will* be a module -- because (system base compile) guarantees that for us. * module/language/scheme/spec.scm (scheme): In the reader, rely on the environment being a module. Define a #:make-default-environment handler, which returns a beautified module, augmented with a fresh definition for current-reader, so that side effects to current-reader are restricted to the compilation unit. * module/language/tree-il/analyze.scm (report-possibly-unbound-variables): * module/language/tree-il/compile-glil.scm (compile-glil): * module/language/tree-il/optimize.scm (optimize!): The environment will be a module. * module/system/base/language.scm (<language>): New field, `make-default-environment'. Defaults to `make-fresh-user-module'. (default-environment): New accessor, returns a default environment for a language. * module/system/repl/common.scm (repl-compile): Always compile relative to the current module, because a module is always acceptable as an environment. * module/system/base/compile.scm (compile-file, compile-and-load): Both of these have a new keyword argument, #:env. For `compile-file', it defaults to the default environment of the source language, and for `compile-and-load', to the current module. (read-and-compile): If there are no expressions read, pass the joiner its default environment (via `default-environment joint').
2009-10-16 15:27:10 +02:00
(tree-il-fold (lambda (x info)
;; X is a leaf: extend INFO's refs accordingly.
(let ((refs (toplevel-info-refs info))
(defs (toplevel-info-defs info))
(locs (toplevel-info-locs info)))
(define (bound? name)
(or (and (module? env)
(module-variable env name))
(memq name defs)))
(record-case x
((<toplevel-ref> name src)
(if (bound? name)
info
(let ((src (or src (find pair? locs))))
(make-toplevel-info (alist-cons name src refs)
defs
locs))))
(else info))))
(lambda (x info)
;; Going down into X.
(let* ((refs (toplevel-info-refs info))
(defs (toplevel-info-defs info))
(src (tree-il-src x))
(locs (cons src (toplevel-info-locs info))))
(define (bound? name)
(or (and (module? env)
(module-variable env name))
(memq name defs)))
(record-case x
((<toplevel-set> name src)
(if (bound? name)
(make-toplevel-info refs defs locs)
(let ((src (find pair? locs)))
(make-toplevel-info (alist-cons name src refs)
defs
locs))))
((<toplevel-define> name)
(make-toplevel-info (alist-delete name refs eq?)
(cons name defs)
locs))
((<application> proc args)
;; Check for a dynamic top-level definition, as is
;; done by code expanded from GOOPS macros.
(let ((name (goops-toplevel-definition proc args
env)))
(if (symbol? name)
(make-toplevel-info (alist-delete name refs
eq?)
(cons name defs)
locs)
(make-toplevel-info refs defs locs))))
compilation enviroments are always modules; simplifications & refactorings * module/ice-9/boot-9.scm (make-fresh-user-module): New public function, makes an anonymous beautified module. * module/language/objcode/spec.scm: We used to have some things in here that allowed lexical variable names and values to be a part of the environment, but no more. Now an environment is just a module. If you want to "inject" free variables into code, just use lambda. * module/language/scheme/compile-tree-il.scm (compile-tree-il): Same here. Also, rely on the fact that an environment *will* be a module -- because (system base compile) guarantees that for us. * module/language/scheme/spec.scm (scheme): In the reader, rely on the environment being a module. Define a #:make-default-environment handler, which returns a beautified module, augmented with a fresh definition for current-reader, so that side effects to current-reader are restricted to the compilation unit. * module/language/tree-il/analyze.scm (report-possibly-unbound-variables): * module/language/tree-il/compile-glil.scm (compile-glil): * module/language/tree-il/optimize.scm (optimize!): The environment will be a module. * module/system/base/language.scm (<language>): New field, `make-default-environment'. Defaults to `make-fresh-user-module'. (default-environment): New accessor, returns a default environment for a language. * module/system/repl/common.scm (repl-compile): Always compile relative to the current module, because a module is always acceptable as an environment. * module/system/base/compile.scm (compile-file, compile-and-load): Both of these have a new keyword argument, #:env. For `compile-file', it defaults to the default environment of the source language, and for `compile-and-load', to the current module. (read-and-compile): If there are no expressions read, pass the joiner its default environment (via `default-environment joint').
2009-10-16 15:27:10 +02:00
(else
(make-toplevel-info refs defs locs)))))
(lambda (x info)
;; Leaving X's scope.
(let ((refs (toplevel-info-refs info))
(defs (toplevel-info-defs info))
(locs (toplevel-info-locs info)))
(make-toplevel-info refs defs (cdr locs))))
(make-toplevel-info '() '() '())
tree))
(for-each (lambda (name+loc)
(let ((name (car name+loc))
(loc (cdr name+loc)))
(warning 'unbound-variable loc name)))
(reverse (toplevel-info-refs toplevel)))
tree)