racket/collects/stepper/private/model.ss
John Clements 05bdd7e0c5 support for begin0. Whew.
svn: r5522
2007-02-01 00:18:42 +00:00

323 lines
14 KiB
Scheme

;step collector state machine (not yet implemented):
;
; datatype held-type = NO-HELD-STEP | SKIPPED-STEP | HELD(args)
;
; states: global state of held
; global: held : held-type
; edge-names: first, skipped-first, second, skipped-second, double, late-let
;
;transitions (& actions):
;
; held = NO-HELD-STEP :
; first(x) : held := HELD(x)
; skipped-first : held := SKIPPED-STEP
; second(x) : trigger(NO-HELD-STEP, x), held := NO-HELD-STEP
; this happens when evaluating unannotated code
; skipped-second : held := NO-HELD-STEP
; I believe this can also arise in unannotated code
; double(x) : double-trigger(x), held := NO-HELD-STEP
; late-let(x) : late-let-trigger(x), held := NO-HELD-STEP
;
; held = SOME(SKIPPED-STEP) :
; first(x) : ERROR
; skipped-first : ERROR
; second(x) : held := NO-HELD-STEP
; this happens e.g. for evaluation of top-level var bound to a procedure
; skipped-second : held := NO-HELD-STEP
; double(x) : ERROR
; late-let(x) : ERROR
;
; held = SOME(HELD(args))
; first(x) : ERROR
; skipped-first : ERROR
; second(x) : trigger(HELD(args),x), held = NO-HELD-STEP
; skipped-second : held = NO-HELD-STEP
; double(x) : ERROR
; late-let(x) : ERROR
(module model mzscheme
(require (lib "contract.ss")
(lib "etc.ss")
(lib "match.ss")
(lib "class.ss")
(prefix a: "annotate.ss")
(prefix r: "reconstruct.ss")
"shared.ss"
"marks.ss"
"model-settings.ss"
"macro-unwind.ss"
"lifting.ss"
;; for breakpoint display
"display-break-stuff.ss")
(define program-expander-contract
((-> void?) ; init
((or/c eof-object? syntax? (cons/c string? any/c)) (-> void?)
. -> . void?) ; iter
. -> .
void?))
(provide/contract
[go (program-expander-contract ; program-expander
(step-result? . -> . void?) ; receive-result
(or/c render-settings? false/c) ; render-settings
boolean? ; track-inferred-names?
(or/c object? (symbols 'testing)) ;; FIXME: can do better: subclass of language% ; the language level
(procedure? . -> . void?) ; run-on-drscheme-side
. -> .
void?)])
; go starts a stepper instance
; see provide stmt for contract
(define (go program-expander receive-result render-settings
track-inferred-names? language-level run-on-drscheme-side)
;; finished-exps:
;; (listof (list/c syntax-object? (or/c number? false?)( -> any)))
;; because of mutation, these cannot be fixed renderings, but must be
;; re-rendered at each step.
(define finished-exps null)
(define/contract add-to-finished
((-> syntax?) (or/c (listof natural-number/c) false/c) (-> any)
. -> . void?)
(lambda (exp-thunk lifting-indices getter)
(set! finished-exps
(append finished-exps
(list (list exp-thunk lifting-indices getter))))))
;; the "held" variables are used to store the "before" step.
(define held-exp-list no-sexp)
(define held-step-was-app? #f)
(define held-finished-list null)
;; highlight-mutated-expressions :
;; ((listof (list/c syntax? syntax?)) (listof (list/c syntax? syntax?))
;; -> (list/c (listof syntax?) (listof syntax?)))
;; highlights changes occurring due to mutation. This function accepts the
;; left-hand-side expressions and the right-hand-side expressions, and
;; matches them against each other to see which ones have changed due to
;; mutation, and highlights these.
;; POSSIBLE RESEARCH POINT: if, say, (list 3 4) is mutated to (list 4 5),
;; should the 4 & 5 be highlighted individually or should the list as a
;; whole be highlighted. Is either one "wrong?" equivalences between
;; reduction semantics?
;;
;; 2005-11-14: punting. just highlight the whole damn thing if there are
;; any differences. In fact, just test for eq?-ness.
#;
(define (highlight-mutated-expressions lefts rights)
(if (or (null? lefts) (null? rights))
(list lefts rights)
(let ([left-car (car lefts)]
[right-car (car rights)])
(if (eq? (syntax-property left-car 'user-source)
(syntax-property right-car 'user-source))
(let ([highlights-added
(highlight-mutated-expression left-car right-car)]
[rest (highlight-mutated-expressions
(cdr lefts) (cdr rights))])
(cons (cons (car highlights-added) (car rest))
(cons (cadr highlights-added) (cadr rest))))))))
;; highlight-mutated-expression: syntax? syntax? -> syntax?
;; given two expressions, highlight 'em both if they differ at all.
;; notes: wanted to use simple "eq?" test... but this will fail when a
;; being-stepped definition (e.g. in a let) turns into a permanent one.
;; We pay a terrible price for the lifting thing. And, for the fact that
;; the highlighting follows from the reductions but can't obviously be
;; deduced from them.
#;
(define (highlight-mutated-expression left right)
(cond
;; if either one is already highlighted, leave them alone.
[(or (stepper-syntax-property left 'stepper-highlight)
(stepper-syntax-property right 'stepper-highlight))
(list left right)]
;; first pass: highlight if not eq?. Should be broken for local-bound
;; things as they pass into permanence.
[(eq? left right)
(list left right)]
[else (list (stepper-syntax-property left 'stepper-highlight)
(stepper-syntax-property right 'stepper-highlight))]))
;; mutated on receipt of a break, used in displaying breakpoint stuff.
(define steps-received 0)
(define break
(opt-lambda (mark-set break-kind [returned-value-list #f])
(set! steps-received (+ steps-received 1))
;; have to be careful else this won't be looked up right away:
(when (getenv "PLTSTEPPERUNSAFE")
(let ([steps-received/current steps-received])
(run-on-drscheme-side
(lambda ()
(display-break-stuff
steps-received/current
mark-set break-kind returned-value-list)))))
(let* ([mark-list (and mark-set (extract-mark-list mark-set))])
(define (reconstruct-all-completed)
(map (match-lambda
[`(,source-thunk ,lifting-indices ,getter)
(match (r:reconstruct-completed
(source-thunk) lifting-indices
getter render-settings)
[#(exp #f) (unwind exp render-settings)]
[#(exp #t) exp])])
finished-exps))
;; (printf "break called with break-kind: ~a ..." break-kind)
(if (r:skip-step? break-kind mark-list render-settings)
(begin
;; (printf " but it was skipped!\n")
(when (or (eq? break-kind 'normal-break)
;; not sure about this...
(eq? break-kind 'nomal-break/values))
(set! held-exp-list skipped-step)))
(begin
;; (printf "and it wasn't skipped.\n")
(case break-kind
[(normal-break normal-break/values)
(begin
(when (and (eq? break-kind 'normal-break)
returned-value-list)
(error 'break
"broken invariant: normal-break can't have returned values"))
(set! held-finished-list (reconstruct-all-completed))
(set! held-exp-list
(map (lambda (exp)
(unwind exp render-settings))
(maybe-lift
(r:reconstruct-left-side
mark-list returned-value-list render-settings)
#f)))
(set! held-step-was-app? (r:step-was-app? mark-list)))]
[(result-exp-break result-value-break)
(if (eq? held-exp-list skipped-step)
;; don't render if before step was a skipped-step
(set! held-exp-list no-sexp)
(let* ([new-finished-list (reconstruct-all-completed)]
[reconstructed
(map (lambda (exp)
(unwind exp render-settings))
(maybe-lift
(r:reconstruct-right-side
mark-list returned-value-list render-settings)
#f))]
[result
(if (eq? held-exp-list no-sexp)
;; in this case, there was no "before" step, due
;; to unannotated code. In this case, we make the
;; optimistic guess that none of the finished
;; expressions were mutated. It would be somewhat
;; painful to do a better job, and the stepper
;; makes no guarantees in this case.
(make-before-after-result
;; NB: this (... ...) IS UNRELATED TO
;; THE MACRO IDIOM OF THE SAME NAME
(list #`(... ...))
(append new-finished-list reconstructed)
'normal)
(let*-values
([(step-kind)
(if (and held-step-was-app?
(eq? break-kind 'result-exp-break))
'user-application
'normal)]
[(left-exps right-exps)
;; write this later:
;; (identify-changed
;; (append held-finished-list held-exps)
;; (append new-finished-list reconstructed))
(values (append held-finished-list
held-exp-list)
(append new-finished-list
reconstructed))])
(make-before-after-result
left-exps right-exps step-kind)))])
(set! held-exp-list no-sexp)
(receive-result result)))]
[(double-break)
;; a double-break occurs at the beginning of a let's
;; evaluation.
(when (not (eq? held-exp-list no-sexp))
(error
'break-reconstruction
"held-exp-list not empty when a double-break occurred"))
(let* ([new-finished-list (reconstruct-all-completed)]
[reconstruct-result
(r:reconstruct-double-break mark-list render-settings)]
[left-side (map (lambda (exp) (unwind exp render-settings))
(maybe-lift (car reconstruct-result) #f))]
[right-side (map (lambda (exp) (unwind exp render-settings))
(maybe-lift (cadr reconstruct-result) #t))])
;; add highlighting code as for other cases...
(receive-result
(make-before-after-result
(append new-finished-list left-side)
(append new-finished-list right-side)
'normal)))]
[(expr-finished-break)
(unless (not mark-list)
(error 'break
"expected no mark-list with expr-finished-break"))
;; in an expr-finished-break, the returned-vals hold (listof
;; (list/c source lifting-index getter)) this will now include
;; define-struct breaks, for which the source is the source
;; and the getter causes an error.
(for-each (lambda (source/index/getter)
(apply add-to-finished source/index/getter))
returned-value-list)]
[else (error 'break "unknown label on break")]))))))
(define maybe-lift
(if (render-settings-lifting? render-settings)
lift
;; ... oh dear; model.ss should disable the double-break & late-let break when lifting is off.
(lambda (stx dont-care) (list stx))))
(define (step-through-expression expanded expand-next-expression)
(let* ([annotated (a:annotate expanded break track-inferred-names?
language-level)])
(eval-syntax annotated)
(expand-next-expression)))
(define (err-display-handler message exn)
(if (not (eq? held-exp-list no-sexp))
(begin
(receive-result
(make-before-error-result (append held-finished-list held-exp-list)
message))
(set! held-exp-list no-sexp))
(receive-result (make-error-result message))))
(program-expander
(lambda ()
;; swap these to allow errors to escape (e.g., when debugging)
(error-display-handler err-display-handler)
#;(void)
)
(lambda (expanded continue-thunk) ; iter
(r:reset-special-values)
(if (eof-object? expanded)
(begin
(receive-result (make-finished-stepping)))
(step-through-expression expanded continue-thunk))))))