ulambda/bootstrap/birth.scm

;;;; Prebirth Lisp implemented in Prebirth Lisp (self-hosting)
;;;;
;;;;  Copyright (C) 2017, 2018 Mike Gerwitz
;;;;
;;;;  This file is part of Ulambda Scheme.
;;;;
;;;;  Ulambda Scheme is free software: you can redistribute it and/or modify
;;;;  it under the terms of the GNU Affero General Public License as
;;;;  published by the Free Software Foundation, either version 3 of the
;;;;  License, or (at your option) any later version.
;;;;
;;;;  This program 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 General Public License for more details.
;;;;
;;;;  You should have received a copy of the GNU Affero General Public License
;;;;  along with this program.  If not, see <http://www.gnu.org/licenses/>.
;;;;



;;; Commentary:

;;; THIS IS TEMPORARY CODE that will be REWRITTEN IN REBIRTH LISP ITSELF
;;; after a very basic bootstrap is complete.  It is retained as an
;;; important artifact for those who wish to build Ulambda Scheme from
;;; scratch without using another version of Ulambda itself.  This is called
;;; "self-hosting".
;;;
;;; This is the Prebirth Lisp implementation of the JavaScript Prebirth
;;; compiler, found in `prebirth.js'---that compiler can be used to compile
;;; this compiler, which can then be used to compile itself, completing the
;;; bootstrapping process.  This process is termed "Birth", and the process
;;; is successful if the output of Birth compiling itself is byte-for-byte
;;; identical to the output of compiling Birth with Prebirth.
;;;
;;; This is largely a 1:1 translation of `prebirth.js'.  See that file for
;;; terminology.
;;;
;;; Note that we're dealing with a small subset of Scheme here, so certain
;;; things might be done differently given a proper implementation.
;;;
;;; The next step after this is ``Rebirth'': both Prebirth and Birth require
;;; the manually written `libprebirth.js' to function.  Rebirth will remove
;;; that completely, which bootstraps the runtime in its entirety.  At that
;;; point, all development will be exclusively in Scheme and we can get on
;;; with Ulambda Scheme.
;;;

;;; Code:



;; pair selection
(define (cadr xs)
  (car (cdr xs)))
(define (caadr xs)
  (car (car (cdr xs))))
(define (caddr xs)
  (car (cdr (cdr xs))))
(define (cadddr xs)
  (car (cdr (cdr (cdr xs)))))
(define (caddddr xs)
  (car (cdr (cdr (cdr (cdr xs))))))
(define (cddr xs)
  (cdr (cdr xs)))

(define (not x)
  (if x #f #t))

;; for convenience
(define (es:match-regexp re s)
  (es:match (es:regexp re) s))



;; Convert source input into a string of tokens.
;;
;; This is the lexer.  Whitespace is ignored.  The grammar consists of
;; simple s-expressions.
;;
;; Tokens are produced with `make-token'.  The source SRC will be
;; left-truncated as input is processed.  POS exists for producing metadata
;; for error reporting---it has no impact on parsing.
;;
;; This implementation was originally recursive and immutable, but the stack
;; was being exhausted, so it was refactored into an inferior
;; implementation.  Note the use of `es:while' and `es:break'---these are
;; quick fixes to the problem of stack exhaustion in browsers (where we have
;; no control over the stack limit); proper tail call support will come
;; later when we have a decent architecture in place.
;;
;; The result is a list of tokens.  See `token' for the format.
(define (lex src pos)
  (let ((toks (list)))
    (es:while #t                              ; browser stack workaround
     (let* ((ws     (or (es:match-regexp "^\\s+"
                                         src)
                        (list "")))
            (ws-len (string-length (car ws)))
            (trim   (substring src ws-len))   ; ignore whitespace, if any
            (newpos (+ pos ws-len)))          ; adj pos to account for removed ws

       (if (string=? "" trim)
           (es:break)                         ; EOF and we're done

           ;; normally we'd use `string-ref' here, but then we'd have to
           ;; implement Scheme characters, so let's keep this simple and keep
           ;; with strings
           (let* ((ch (substring trim 0 1))
                  (t (case ch
                       ;; comments extend until the end of the line
                       ((";") (let ((eol (es:match-regexp "^(.*?)(\\n|$)" trim)))
                                (make-token "comment" (cadr eol) trim newpos)))

                       ;; left and right parenthesis are handled in the same
                       ;; manner: they produce distinct tokens with
                       ;; single-character lexemes
                       (("(")  (make-token "open" ch trim newpos))
                       ((")")  (make-token "close" ch trim newpos))

                       ;; strings are delimited by opening and closing ASCII
                       ;; double quotes, which can be escaped with a
                       ;; backslash
                       (("\"") (let ((str (es:match-regexp
                                           "^\"(|(?:.|\\\n)*?[^\\\\])\""
                                           trim)))
                                 (or str (parse-error
                                          src pos
                                          "missing closing string delimiter"))
                                 ;; a string token consists of the entire
                                 ;; string including quotes as its lexeme,
                                 ;; but its value will be the value of the
                                 ;; string without quotes due to the `str'
                                 ;; match group (see `token')
                                 (make-token "string" str trim newpos)))

                       (else
                        ;; anything else is considered a symbol up until
                        ;; whitespace or any of the aforementioned
                        ;; delimiters
                        (let ((symbol (es:match-regexp "^[^\\s()\"]+"
                                                       trim)))
                          (make-token "symbol" symbol trim newpos))))))

             ;; yikes---see notes in docblock with regards to why
             ;; we're using mutators here
             (set! toks (append toks (list (car t))))
             (set! src  (cadr t))
             (set! pos  (caddr t))))))
    toks))



;; Throw an error with a window of surrounding source code.
;;
;; The "window" is simply ten characters to the left and right of the
;; first character of the source input SRC that resulted in the error.
;; It's a little more than useless.
(define (parse-error src pos msg)
  (let ((window (substring src (- pos 10) (+ pos 10))))
    (error (string-append msg " (pos " pos "): " window)
           src)))



;; Produce a token, left-truncate src, and update pos.
;;
;; Unlike the JS Prebirth implementation which uses a key/value object,
;; we're just using a simple list.
;;
;; The expected arguments are: the token type TYPE, the match group or
;; string MATCH, left-truncated source code SRC, and the position POS
;; relative to the original source code.
(define (make-token type match src pos)
  (let* ((parts  (if (list? match) match (list match match)))
         (lexeme (car parts))
         ;; the value is the first group of the match (indicating what we
         ;; are actually interested in), and the lexeme is the full match,
         ;; which might include, for example, string delimiters
         (value  (or (and (pair? (cdr parts))
                          (cadr parts))
                     lexeme))
         (len    (string-length lexeme)))

    ;; produce token and recurse on `lex', left-truncating the source
    ;; string to discard what we have already processed
    (list (list (quote token) type lexeme value pos)
          (substring src len)
          (+ pos len))))


;; various accessor procedures for token lists (we're Prebirth Lisp here,
;; so no record support or anything fancy!)
(define (token? t)       (and (pair? t) (symbol=? (quote token) (car t))))
(define (token-type t)   (cadr t))
(define (token-lexeme t) (caddr t))
(define (token-value t)  (cadddr t))
(define (token-pos t)    (caddddr t))



;; Produce an AST from the given string SRC of sexps
;;
;; This is essentially the CST with whitespace removed.  It first invokes
;; the lexer to produce a token string from the input sexps SRC.  From this,
;; it verifies only proper nesting (that SRC does not close sexps too early
;; and that EOF isn't reached before all sexps are closed) and produces an
;; AST that is an isomorphism of the original sexps.
(define (parse-lisp src)
  ;; accessor methods to make you and me less consfused
  (define (ast-depth ast) (car ast))
  (define (ast-tree  ast) (cadr ast))
  (define (ast-stack ast) (caddr ast))

  ;; perform a leftmost reduction on the token string
  (define (toks->ast toks)
    (fold
     (lambda (token result)
       (let ((depth (ast-depth result))
             (xs    (ast-tree result))
             (stack (ast-stack result))
             (type  (token-type token))
             (pos   (token-pos token)))

         ;; there are very few token types to deal with (again, this is a
         ;; very simple bootstrap lisp)
         (case type
           ;; ignore comments
           (("comment") result)

           ;; when beginning a new expression, place the expression
           ;; currently being processed onto a stack, allocate a new list,
           ;; and we'll continue processing into that new list
           (("open") (list (+ depth 1)
                           (list)
                           (cons xs stack)))

           ;; once we reach the end of the expression, pop the parent off of
           ;; the stack and append the new list to it
           (("close") (if (zero? depth)
                          (parse-error src pos
                                       "unexpected closing parenthesis")
                          (list (- depth 1)
                                (append (car stack) (list xs))
                                (cdr stack))))

           ;; strings and symbols (we cheat and just consider everything,
           ;; including numbers and such, to be symbols) are just copied
           ;; in place
           (("string" "symbol") (list depth
                                      (append xs (list token))
                                      stack))

           ;; we should never encounter anything else unless there's a bug
           ;; in the tokenizer or we forget a token type above
           (else (parse-error
                  src pos (string-append
                           "unexpected token `" type "'"))))))
     (list 0 (list) (list))    ; initial 0 depth; empty tree;  expr stack
     toks))


  ;; lex the input SRC and pass it to `toks->ast' to generate the AST;
  ;; if the depth is non-zero after we're done, then we're unbalanced.
  (let* ((toks (lex src 0))
         (ast  (toks->ast toks)))
    (if (zero? (ast-depth ast))
        (ast-tree ast)
        ;; if we terminate at a non-zero depth, that means there ar still
        ;; open sexps
        (error (string-append
                "unexpected end of input at depth "
                (ast-depth ast))))))


;; Compile Birth Lisp AST into ECMAScript.
;;
;; The AST can be generated with `parse-lisp'.
(define (birth->ecmascript ast)
  ;; Generate ECMAScript-friendly name from the given id.
  ;;
  ;; A subset of special characters that are acceptable in Scheme are
  ;; converted in an identifiable manner; others are simply converted to `$'
  ;; in a catch-all and therefore could result in conflicts and cannot be
  ;; reliably distinguished from one-another.  Remember: this is temporary
  ;; code.
  (define (tname->id name)
    (if (es:match (es:regexp "^\\d+$") name)
        name
        (string-append
         "$$" (es:replace (es:regexp "[^a-zA-Z0-9_]" "g")
                          (lambda (c)
                            (case c
                              (("-") "$_$")
                              (("?") "$7$")
                              (("@") "$a$")
                              (("!") "$b$")
                              ((">") "$g$")
                              (("#") "$h$")
                              (("*") "$k$")
                              (("<") "$l$")
                              (("&") "$n$")
                              (("%") "$o$")
                              (("+") "$p$")
                              (("=") "$q$")
                              (("^") "$v$")
                              (("/") "$w$")
                              (("$") "$$")
                              (else  "$")))
                          name))))

  ;; Join a list of strings XS on a delimiter DELIM
  (define (join delim xs)
    (if (pair? xs)
        (fold (lambda (x str)
                (string-append str delim x))
              (car xs)
              (cdr xs))
        ""))


  ;; Compile parameter list.
  ;;
  ;; This simply takes the value of the symbol and outputs it (formatted),
  ;; delimited by commas.
  (define (params->es params)
    (join ", " (map (lambda (t)
                      (tname->id (token-value t)))
                    params)))


  ;; Compile body s-expressions into ECMAScript
  ;;
  ;; This produces a 1:1 mapping of body XS s-expressions to ES statements,
  ;; recursively.  The heavy lifting is done by `sexp->es'.
  (define (body->es xs ret)
    ;; recursively process body XS until we're out of pairs
    (if (not (pair? xs))
        ""
        (let* ((x     (car xs))
               (rest  (cdr xs))
               (more? (or (not ret) (pair? rest))))
          ;; the result is a semicolon-delimited string of statements, with
          ;; the final statement prefixed with `return' unless (not ret)
          (string-append
           "    "
           (if more? "" "return ")   ; prefix with `return' if last body exp
           (sexp->es x) ";"          ; process current body expression
           (if (pair? rest) "\n" "")
           (body->es rest ret)))))   ; recurse


  ;; Compile procedure definition into an ES function definition
  ;;
  ;; This will fail if the given token is not a `define'.
  (define (cdfn t)
    ;; e.g. (define (foo ...) body)
    (let* ((dfn    (cadr t))
           (id     (tname->id (token-value (car dfn))))
           (params (params->es (cdr dfn)))
           (body   (body->es (cddr t) #t)))
      ;; this is the final format---each procedure becomes its own function
      ;; definition in ES
      (string-append
       "function " id "(" params ")\n{\n" body "\n};")))


  ;; Function/procedure aliases and special forms
  ;;
  ;; And here we have what is probably the most grotesque part of this file.
  ;;
  ;; This map allows for a steady transition---items can be removed as they
  ;; are written in Prebirth Lisp.  This should give us a sane (but still
  ;; simple) environment with which we can start to self-host.
  ;;
  ;; String values are simple function aliases.  Function values take over
  ;; the compilation of that function and allow for defining special forms
  ;; (in place of macro support).  The first argument FN is the name of the
  ;; function/procedure/form, and ARGS is the list of arguments.
  ;;
  ;; These are by no means meant to be solid implementations; notable
  ;; deficiencies are documented, but don't expect this to work properly in
  ;; every case.  They will be replaced with proper R7RS implementations in
  ;; the future (post-Rebirth).
  (define (fnmap fn args t)
    (case fn
      (("es:console")
       (string-append "console.log(" (map sexp->es args) ")"))

      ;; only expands `else' clauses; this is just to facilitate its use
      ;; moving forward
      (("cond-expand")
       (join "" (map (lambda (clause)
                       (let ((feature (token-value (car clause)))
                             (body    (cdr clause)))
                         (if (string=? feature "else")
                             (body->es body #f)
                             "")))
                     args)))

      ;; yes, there are more important things to do until we get to the
      ;; point where it's worth implementing proper tail calls
      (("es:while")
       (let ((pred (car args))
             (body (cdr args)))
         (string-append
          "(function(__whilebrk){"
          "while (" (sexp->es pred) "){\n"
          (body->es body #f) " if (__whilebrk) break;\n"
          "}\n"
          "})(false)")))
      (("es:break") "__whilebrk=true")

      ;; fortunately ES6+ has native symbol support :)
      ;; we don't (yet?) need list quoting in Prebirth
      (("quote")
       (if (pair? (cdr args))
           (error "quoting lists is not yet supported; sorry!")
           (string-append "Symbol.for('" (sexp->es args) "')")))

      (("define") (cdfn t))

      (("lambda")
       (let ((fnargs (car args))
             (body   (cdr args)))
         (string-append
          "function(" (join ", " (map sexp->es fnargs)) "){\n"
          (body->es body #t)
          "}")))

      ;; simple if statement with optional else, wrapped in a self-executing
      ;; function to simplify code generation (e.g. returning an if)
      (("if")
       (let ((pred (car args))
             (t    (cadr args))
             (f    (and (pair? (cddr args))
                        (caddr args))))
         (string-append
          "(function(){"
          "if (_truep(" (sexp->es pred) ")){return " (sexp->es t) ";}"
          (if (pair? f)
              (string-append "else{return " (sexp->es f) ";}")
              "")
          "})()")))

      ;; and short-circuits, so we need to implement it as a special form
      ;; rather than an alias
      (("and")
       (string-append
        "(function(__and){\n"
        (join "" (map (lambda (expr)
                        (string-append
                         "__and = " (sexp->es expr) "; "
                         "if (!_truep(__and)) return false;\n"))
                      args))
        "return __and;})()"))

      ;; or short-circuits, so we need to implement it as a special form
      ;; rather than an alias
      (("or")
       (string-append
        "(function(__or){\n"
        (join "" (map (lambda (expr)
                        (string-append
                         "__or = " (sexp->es expr) "; "
                         "if (_truep(__or)) return __or;\n"))
                      args))
        "return false;})()"))

      ;; (let ((binding val) ...) ...body), compiled as a self-executing
      ;; function which allows us to easily represent the return value of
      ;; the entire expression while maintaining local scope
      (("let*")
       (let ((bindings (car args))
             (body     (cdr args)))
         (string-append
          "(function(){\n"
          (join "" (map (lambda (binding)
                          (let ((var  (car binding))
                                (init (cadr binding)))
                            (string-append "    let " (sexp->es var)
                                           " = " (sexp->es init) ";\n")))
                        bindings))
          (body->es body #t) "\n"
          "    })()")))

      ;; similar to the above, but variables cannot reference one-another
      (("let")
       (let* ((bindings (car args))
              (body     (cdr args))
              (fparams  (join ", " (map sexp->es
                                        (map car bindings))))
              (fargs    (join ", " (map sexp->es
                                        (map cadr bindings)))))
         (string-append "(function(" fparams "){\n"
                        (body->es body #t) "\n"
                        "})(" fargs ")")))

      ;; and here I thought Prebirth Lisp would be simple...but having
      ;; `case' support really keeps things much more tidy, so here we are
      ;; (note that it doesn't support the arrow form, nor does it support
      ;; expressions as data)
      (("case")
       (let ((key     (car args))
             (clauses (cdr args)))
         (string-append
          "(function(){const _key=" (sexp->es key) ";\n"
          "switch (_key){\n"
          (join ""
                (map (lambda (data exprs)
                       (string-append
                        (if (and (token? data)
                                  (string=? "else" (token-lexeme data)))
                            "default:\n"
                            (join ""
                                  (map (lambda (datum)
                                         (string-append
                                          "case " (sexp->es datum) ":\n"))
                                       data)))
                        (body->es exprs #t) "\n"))
                     (map car clauses)
                     (map cdr clauses)))
          "}})()")))

      (("set!")
       (let ((varid (car args))
             (val   (cadr args)))
         (string-append (sexp->es varid) " = " (sexp->es val))))

      ;; normal procedure application
      (else (let* ((idfn   (tname->id fn))
                   (argstr (join ", " (map sexp->es args))))
              (string-append idfn "(" argstr ")")))))


  ;; Convert s-expressions or scalar into ECMAScript
  ;;
  ;; T may be either an array of tokens or a primitive token (e.g. string,
  ;; symbol).  This procedure is applied recursively to T as needed if T is
  ;; a list.
  (define (sexp->es t)
    (if (not (list? t))
        (error "unexpected non-list for sexp->es token"))

    (if (token? t)
        (case (token-type t)
          ;; strings output as-is (note that we don't escape double quotes,
          ;; because the method of escaping them is the same in Scheme as it
          ;; is in ECMAScript---a backslash)
          (("string") (string-append "\"" (token-value t) "\""))

          ;; symbols have the same concerns as procedure definitions: the
          ;; identifiers generated need to be ES-friendly
          (("symbol") (tname->id (token-value t)))

          (else (error
                 (string-append
                  "cannot compile unknown token `" (token-type t) "'"))))

        ;; otherwise, process the expression
        (fnmap (token-value (car t))
               (cdr t)
               t)))

  ;; output libprebirth and compiled output, wrapped in a self-executing
  ;; function to limit scope
  (string-append "(function(){"
                 (es:file->string "libprebirth.js") "\n\n"
                 (join "\n\n" (map sexp->es ast))
                 "})();"))


;; at this point, this program can parse itself and output a CST (sans
;; whitespace)
(es:console (birth->ecmascript
             (parse-lisp
              (es:file->string "/dev/stdin"))))