X-Git-Url: https://jxself.org/git/?a=blobdiff_plain;f=loopy.scm;h=2f733c01d4fe32afa21641bc456c37efc4f0f171;hb=35a90099f27585e853a8b8aee498c5228d8f9697;hp=e03cab6ae59d03ec34b72cd84f511f3ec6ca8c00;hpb=d73acd37f6940e5f9adcf449f814279fb5c7de68;p=8sync.git diff --git a/loopy.scm b/loopy.scm index e03cab6..2f733c0 100644 --- a/loopy.scm +++ b/loopy.scm @@ -1,35 +1,322 @@ -(use-modules (srfi srfi-9) - (srfi srfi-9 gnu) - (ice-9 q) - (ice-9 match)) +(define-module (loopy agenda) + #:use-module (srfi srfi-9) + #:use-module (srfi srfi-9 gnu) + #:use-module (ice-9 q) + #:use-module (ice-9 match) + #:use-module (ice-9 receive) + #:export ( + make-agenda agenda? + agenda-queue agenda-prompt-tag + agenda-port-pmapping agenda-schedule + + make-async-prompt-tag + + + make-time-segment time-segment? + time-segment-time time-segment-queue + + time-< time-= time-<= time-+ + + + make-time-delta time-delta? + time-delta-sec time-delta-usec + + + make-schedule schedule? + schedule-add! schedule-empty? + schedule-segments + + schedule-segments-split schedule-extract-until! + add-segments-contents-to-queue! + + make-port-mapping + port-mapping-set! port-mapping-remove! + port-mapping-empty? port-mapping-non-empty? + + + make-run-request run-request? + run-request-proc run-request-when + + run wrap run-wrap run-wrap-at + + %current-agenda + start-agenda agenda-run-once)) ;; @@: Using immutable agendas here, so wouldn't it make sense to -;; replace this +;; replace this queue stuff with using pfds based immutable queues? + + +;;; Agenda definition +;;; ================= + +;;; The agenda consists of: +;;; - a queue of immediate items to handle +;;; - sheduled future events to be added to a future queue +;;; - a tag by which running processes can escape for some asynchronous +;;; operation (from which they can be returned later) +;;; - a mapping of ports to various handler procedures +;;; +;;; The goal, eventually, is for this all to be immutable and functional. +;;; However, we aren't there yet. Some tricky things: +;;; - The schedule needs to be immutable, yet reasonably efficient. +;;; - Need to use immutable queues (ijp's pfds library?) +;;; - Modeling reading from ports as something repeatable, +;;; and with reasonable separation from functional components? (define-immutable-record-type - (make-agenda-intern queue prompt-tag) + (make-agenda-intern queue prompt-tag port-mapping schedule time) agenda? (queue agenda-queue) - (prompt-tag agenda-prompt-tag)) + (prompt-tag agenda-prompt-tag) + (port-mapping agenda-port-mapping) + (schedule agenda-schedule) + (time agenda-time)) (define (make-async-prompt-tag) (make-prompt-tag "prompt")) (define* (make-agenda #:key (queue (make-q)) - (prompt (make-prompt-tag))) - (make-agenda-intern queue prompt)) + (prompt (make-prompt-tag)) + (port-mapping (make-port-mapping)) + (schedule (make-schedule)) + (time (gettimeofday))) + (make-agenda-intern queue prompt port-mapping schedule time)) + + + +;;; Schedule +;;; ======== + +;;; This is where we handle timed events for the future + +;; This section totally borrows from the ideas in SICP +;; <3 <3 <3 + +;; NOTE: time is a cons of (seconds . microseconds) + +(define-record-type + (make-time-segment-intern time queue) + time-segment? + (time time-segment-time) + (queue time-segment-queue)) + +(define (time-segment-right-format time) + (match time + ;; time is already a cons of second and microsecnd + (((? integer? s) . (? integer? u)) time) + ;; time was just an integer (just the second) + ((? integer? _) (cons time 0)) + (_ (throw 'invalid-time "Invalid time" time)))) + +(define* (make-time-segment time #:optional (queue (make-q))) + (make-time-segment-intern time queue)) + +(define (time-< time1 time2) + (cond ((< (car time1) + (car time2)) + #t) + ((and (= (car time1) + (car time2)) + (< (cdr time1) + (cdr time2))) + #t) + (else #f))) + +(define (time-= time1 time2) + (and (= (car time1) (car time2)) + (= (cdr time1) (cdr time2)))) + +(define (time-<= time1 time2) + (or (time-< time1 time2) + (time-= time1 time2))) + + +(define-record-type + (make-time-delta-intern sec usec) + time-delta? + (sec time-delta-sec) + (usec time-delta-usec)) + +(define* (make-time-delta sec #:optional usec) + (make-time-delta-intern sec (or usec 0))) + +(define (time-+ time time-delta) + (cons (+ (car time) (time-delta-sec time-delta)) + (+ (cdr time) (time-delta-usec time-delta)))) + + +(define-record-type + (make-schedule-intern segments) + schedule? + (segments schedule-segments set-schedule-segments!)) + +(define* (make-schedule #:optional segments) + (make-schedule-intern (or segments '()))) + +;; TODO: This code is reasonably easy to read but it +;; mutates AND is worst case of O(n) in both space and time :( +;; but at least it'll be reasonably easy to refactor to +;; a more functional setup? +(define (schedule-add! schedule time proc) + (let ((time (time-segment-right-format time))) + (define (new-time-segment) + (let ((new-segment + (make-time-segment time))) + (enq! (time-segment-queue new-segment) proc) + new-segment)) + (define (loop segments) + (define (segment-equals-time? segment) + (time-= time (time-segment-time segment))) + + (define (segment-more-than-time? segment) + (time-< time (time-segment-time segment))) + + ;; We could switch this out to be more mutate'y + ;; and avoid the O(n) of space... is that over-optimizing? + (match segments + ;; If we're at the end of the list, time to make a new + ;; segment... + ('() (cons (new-time-segment) '())) + ;; If the segment's time is exactly our time, good news + ;; everyone! Let's append our stuff to its queue + (((? segment-equals-time? first) rest ...) + (enq! (time-segment-queue first) proc) + segments) + ;; If the first segment is more than our time, + ;; ours belongs before this one, so add it and + ;; start consing our way back + (((? segment-more-than-time? first) rest ...) + (cons (new-time-segment) segments)) + ;; Otherwise, build up recursive result + ((first rest ... ) + (cons first (loop rest))))) + (set-schedule-segments! + schedule + (loop (schedule-segments schedule))))) + +(define (schedule-empty? schedule) + (eq? (schedule-segments schedule) '())) + +(define (schedule-segments-split schedule time) + "Does a multiple value return of time segments before/at and after TIME" + (let ((time (time-segment-right-format time))) + (define (segment-is-now? segment) + (time-= (time-segment-time segment) time)) + (define (segment-is-before-now? segment) + (time-< (time-segment-time segment) time)) + + (let loop ((segments-before '()) + (segments-left (schedule-segments schedule))) + (match segments-left + ;; end of the line, return + ('() + (values (reverse segments-before) '())) + + ;; It's right now, so time to stop, but include this one in before + ;; but otherwise return + (((? segment-is-now? first) rest ...) + (values (reverse (cons first segments-before)) rest)) + + ;; This is prior or at now, so add it and keep going + (((? segment-is-before-now? first) rest ...) + (loop (cons first segments-before) rest)) + + ;; Otherwise it's past now, just return what we have + (segments-after + (values segments-before segments-after)))))) + +(define (schedule-extract-until! schedule time) + "Extract all segments until TIME from SCHEDULE, and pop old segments off" + (receive (segments-before segments-after) + (schedule-segments-split schedule time) + (set-schedule-segments! schedule segments-after) + segments-before)) + +(define (add-segments-contents-to-queue! segments queue) + (for-each + (lambda (segment) + (let ((seg-queue (time-segment-queue segment))) + (while (not (q-empty? seg-queue)) + (enq! queue (deq! seg-queue))))) + segments)) + + + +;;; Port handling +;;; ============= + +(define (make-port-mapping) + (make-hash-table)) + +(define* (port-mapping-set! port-mapping port #:optional read write except) + "Sets port-mapping for reader / writer / exception handlers" + (if (not (or read write except)) + (throw 'no-handlers-given "No handlers given for port" port)) + (hashq-set! port-mapping port + `#(,read ,write ,except))) + +(define (port-mapping-remove! port-mapping port) + (hashq-remove! port-mapping port)) + +;; TODO: This is O(n), I'm pretty sure :\ +;; ... it might be worthwhile for us to have a +;; port-mapping record that keeps a count of how many +;; handlers (maybe via a promise?) +(define (port-mapping-empty? port-mapping) + "Is this port mapping empty?" + (eq? (hash-count (const #t) port-mapping) 0)) + +(define (port-mapping-non-empty? port-mapping) + "Whether this port-mapping contains any elements" + (not (port-mapping-empty? port-mapping))) + + + +;;; Request to run stuff +;;; ==================== + +(define-record-type + (make-run-request proc when) + run-request? + (proc run-request-proc) + (when run-request-when)) + +(define* (run proc #:optional when) + (make-run-request proc when)) + +(define-syntax-rule (wrap body ...) + (lambda () + body ...)) + +(define-syntax-rule (run-wrap body ...) + (run (wrap body ...))) + +(define-syntax-rule (run-wrap-at body ... when) + (run (wrap body ...) when)) + + +;;; Execution of agenda, and current agenda +;;; ======================================= (define %current-agenda (make-parameter #f)) (define* (start-agenda agenda #:optional stop-condition) (let loop ((agenda agenda)) - (let ((new-agenda + (let ((new-agenda + ;; @@: Hm, maybe here would be a great place to handle + ;; select'ing on ports. + ;; We could compose over agenda-run-once and agenda-read-ports (parameterize ((%current-agenda agenda)) (agenda-run-once agenda)))) (if (and stop-condition (stop-condition agenda)) 'done - (loop new-agenda))))) + (let ((updated-agenda + ;; Adjust the agenda's time just in time + ;; We do this here rather than in agenda-run-once to make + ;; agenda-run-once's behavior fairly predictable + (set-field new-agenda (agenda-time) (gettimeofday)))) + (loop updated-agenda)))))) (define (agenda-run-once agenda) "Run once through the agenda, and produce a new agenda @@ -48,17 +335,36 @@ based on the results" (let* ((proc (q-pop! queue)) (proc-result (call-proc proc)) (enqueue - (lambda (new-proc) - (enq! next-queue new-proc)))) + (lambda (run-request) + (define (schedule-at! time proc) + (schedule-add! (agenda-schedule agenda) time proc)) + (let ((request-time (run-request-when run-request))) + (match request-time + ((? time-delta? time-delta) + (let ((time (time-+ (agenda-time agenda) + time-delta))) + (schedule-at! time (run-request-proc proc)))) + ((? integer? sec) + (let ((time (cons sec 0))) + (schedule-at! time (run-request-proc proc)))) + (((? integer? sec) . (? integer? usec)) + (schedule-at! request-time (run-request-proc proc))) + (#f + (enq! next-queue (run-request-proc run-request)))))))) ;; @@: We might support delay-wrapped procedures here (match proc-result - ((? procedure? new-proc) + ;; TODO: replace procedure with something that indicates + ;; intent to run. Use a (run foo) procedure + ((? run-request? new-proc) (enqueue new-proc)) - (((? procedure? new-procs) ...) + (((? run-request? new-procs) ...) (for-each (lambda (new-proc) (enqueue new-proc)) - new-procs))))) - ;; TODO: Selecting on ports would happen here? + new-procs)) + ;; do nothing + (_ #f)))) + ;; TODO: Alternately, we could return the next-queue + ;; along with changes to be added to the schedule here? ;; Return new agenda, with next queue set (set-field agenda (agenda-queue) next-queue)))