;;; rx.el --- S-exp notation for regexps           --*- lexical-binding: t -*-

;; Copyright (C) 2001-2019 Free Software Foundation, Inc.

;; This file is part of GNU Emacs.

;; GNU Emacs is free software: you can redistribute it and/or modify
;; it under the terms of the GNU General Public License as published by
;; the Free Software Foundation, either version 3 of the License, or
;; (at your option) any later version.

;; GNU Emacs 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 General Public License
;; along with GNU Emacs.  If not, see <https://www.gnu.org/licenses/>.

;;; Commentary:

;; This facility allows writing regexps in a sexp-based language
;; instead of strings.  Regexps in the `rx' notation are easier to
;; read, write and maintain; they can be indented and commented in a
;; natural way, and are easily composed by program code.
;; The translation to string regexp is done by a macro and does not
;; incur any extra processing during run time.  Example:
;;
;;  (rx bos (or (not (any "^"))
;;              (seq "^" (or " *" "["))))
;;
;; => "\\`\\(?:[^^]\\|\\^\\(?: \\*\\|\\[\\)\\)"
;;
;; The notation is much influenced by and retains some compatibility with
;; Olin Shivers's SRE, with concessions to Emacs regexp peculiarities,
;; and the older Emacs package Sregex.

;;; Code:

;; The `rx--translate...' functions below return (REGEXP . PRECEDENCE),
;; where REGEXP is a list of string expressions that will be
;; concatenated into a regexp, and PRECEDENCE is one of
;;
;;  t    -- can be used as argument to postfix operators (eg. "a")
;;  seq  -- can be concatenated in sequence with other seq or higher (eg. "ab")
;;  lseq -- can be concatenated to the left of rseq or higher (eg. "^a")
;;  rseq -- can be concatenated to the right of lseq or higher (eg. "a$")
;;  nil  -- can only be used in alternatives (eg. "a\\|b")
;;
;; They form a lattice:
;;
;;           t          highest precedence
;;           |
;;          seq
;;         /   \
;;      lseq   rseq
;;         \   /
;;          nil         lowest precedence


(defconst rx--char-classes
  '((digit         . digit)
    (numeric       . digit)
    (num           . digit)
    (control       . cntrl)
    (cntrl         . cntrl)
    (hex-digit     . xdigit)
    (hex           . xdigit)
    (xdigit        . xdigit)
    (blank         . blank)
    (graphic       . graph)
    (graph         . graph)
    (printing      . print)
    (print         . print)
    (alphanumeric  . alnum)
    (alnum         . alnum)
    (letter        . alpha)
    (alphabetic    . alpha)
    (alpha         . alpha)
    (ascii         . ascii)
    (nonascii      . nonascii)
    (lower         . lower)
    (lower-case    . lower)
    (punctuation   . punct)
    (punct         . punct)
    (space         . space)
    (whitespace    . space)
    (white         . space)
    (upper         . upper)
    (upper-case    . upper)
    (word          . word)
    (wordchar      . word)
    (unibyte       . unibyte)
    (multibyte     . multibyte))
  "Alist mapping rx symbols to character classes.
Most of the names are from SRE.")

(defvar rx-constituents nil
  "Alist of old-style rx extensions, for compatibility.

Each element is (SYMBOL . DEF).

If DEF is a symbol, then SYMBOL is an alias of DEF.

If DEF is a string, then SYMBOL is a plain rx symbol defined as the
   regexp string DEF.

If DEF is a list on the form (FUN MIN-ARGS MAX-ARGS PRED), then
   SYMBOL is an rx form with at least MIN-ARGS and at most
   MAX-ARGS arguments.  If MAX-ARGS is nil, then there is no upper limit.
   FUN is a function taking the entire rx form as single argument
   and returning the translated regexp string.
   If PRED is non-nil, it is a predicate that all actual arguments must
   satisfy.")

;; TODO: Additions to consider:
;; - A better name for `anything', like `any-char' or `anychar'.
;; - A name for (or), maybe `unmatchable'.
;; - A construct like `or' but without the match order guarantee,
;;   maybe `unordered-or'.  Useful for composition or generation of
;;   alternatives; permits more effective use of regexp-opt.

(defun rx--translate-symbol (sym)
  "Translate an rx symbol.  Return (REGEXP . PRECEDENCE)."
  (pcase sym
    ;; Use `list' instead of a quoted list to wrap the strings here,
    ;; since the return value may be mutated.
    ((or 'nonl 'not-newline 'any) (cons (list ".") t))
    ('anything                    (rx--translate-form '(or nonl "\n")))
    ((or 'bol 'line-start)        (cons (list "^") 'lseq))
    ((or 'eol 'line-end)          (cons (list "$") 'rseq))
    ((or 'bos 'string-start 'bot 'buffer-start) (cons (list "\\`") t))
    ((or 'eos 'string-end   'eot 'buffer-end)   (cons (list "\\'") t))
    ('point                       (cons (list "\\=") t))
    ((or 'bow 'word-start)        (cons (list "\\<") t))
    ((or 'eow 'word-end)          (cons (list "\\>") t))
    ('word-boundary               (cons (list "\\b") t))
    ('not-word-boundary           (cons (list "\\B") t))
    ('symbol-start                (cons (list "\\_<") t))
    ('symbol-end                  (cons (list "\\_>") t))
    ('not-wordchar                (cons (list "\\W") t))
    (_
     (cond
      ((let ((class (cdr (assq sym rx--char-classes))))
         (and class (cons (list (concat "[[:" (symbol-name class) ":]]")) t))))

      ;; For compatibility with old rx.
      ((let ((entry (assq sym rx-constituents)))
         (and (progn
                (while (and entry (not (stringp (cdr entry))))
                  (setq entry
                        (if (symbolp (cdr entry))
                            ;; Alias for another entry.
                            (assq (cdr entry) rx-constituents)
                          ;; Wrong type, try further down the list.
                          (assq (car entry)
                                (cdr (memq entry rx-constituents))))))
                entry)
              (cons (list (cdr entry)) nil))))
      (t (error "Unknown rx symbol `%s'" sym))))))

(defun rx--enclose (left-str rexp right-str)
  "Bracket REXP by LEFT-STR and RIGHT-STR."
  (append (list left-str) rexp (list right-str)))

(defun rx--bracket (rexp)
  (rx--enclose "\\(?:" rexp "\\)"))

(defun rx--sequence (left right)
  "Return the sequence (concatenation) of two translated items,
each on the form (REGEXP . PRECEDENCE), returning (REGEXP . PRECEDENCE)."
  ;; Concatenation rules:
  ;;  seq  ++ seq  -> seq
  ;;  lseq ++ seq  -> lseq
  ;;  seq  ++ rseq -> rseq
  ;;  lseq ++ rseq -> nil
  (cond ((not (car left)) right)
        ((not (car right)) left)
        (t
         (let ((l (if (memq (cdr left) '(nil rseq))
                      (cons (rx--bracket (car left)) t)
                    left))
               (r (if (memq (cdr right) '(nil lseq))
                      (cons (rx--bracket (car right)) t)
                    right)))
           (cons (append (car l) (car r))
                 (if (eq (cdr l) 'lseq)
                     (if (eq (cdr r) 'rseq)
                         nil                   ; lseq ++ rseq
                       'lseq)                  ; lseq ++ seq
                   (if (eq (cdr r) 'rseq)
                       'rseq                   ; seq ++ rseq
                     'seq)))))))               ; seq ++ seq

(defun rx--translate-seq (body)
  "Translate a sequence of one or more rx items.  Return (REGEXP . PRECEDENCE)."
  (if body
      (let* ((items (mapcar #'rx--translate body))
             (result (car items)))
        (dolist (item (cdr items))
          (setq result (rx--sequence result item)))
        result)
    (cons nil 'seq)))

(defun rx--empty ()
  "Regexp that never matches anything."
  (cons (list regexp-unmatchable) 'seq))

;; `cl-every' replacement to avoid bootstrapping problems.
(defun rx--every (pred list)
  "Whether PRED is true for every element of LIST."
  (while (and list (funcall pred (car list)))
    (setq list (cdr list)))
  (null list))

(defun rx--translate-or (body)
  "Translate an or-pattern of one of more rx items.
Return (REGEXP . PRECEDENCE)."
  ;; FIXME: Possible improvements:
  ;;
  ;; - Turn single characters to strings: (or ?a ?b) -> (or "a" "b"),
  ;;   so that they can be candidates for regexp-opt.
  ;;
  ;; - Translate compile-time strings (`eval' forms), again for regexp-opt.
  ;;
  ;; - Flatten sub-patterns first: (or (or A B) (or C D)) -> (or A B C D)
  ;;   in order to improve effectiveness of regexp-opt.
  ;;   This would also help composability.
  ;;
  ;; - Use associativity to run regexp-opt on contiguous subsets of arguments
  ;;   if not all of them are strings.  Example:
  ;;   (or (+ digit) "CHARLIE" "CHAN" (+ blank))
  ;;   -> (or (+ digit) (or "CHARLIE" "CHAN") (+ blank))
  ;;
  ;; - Fuse patterns into a single character alternative if they fit.
  ;;   regexp-opt will do that if all are strings, but we want to do that for:
  ;;     * symbols that expand to classes: space, alpha, ...
  ;;     * character alternatives: (any ...)
  ;;     * (syntax S), for some S (whitespace, word)
  ;;   so that (or "@" "%" digit (any "A-Z" space) (syntax word))
  ;;        -> (any "@" "%" digit "A-Z" space word)
  ;;        -> "[A-Z@%[:digit:][:space:][:word:]]"
  ;;
  ;; Problem: If a subpattern is carefully written to to be
  ;; optimisable by regexp-opt, how do we prevent the transforms
  ;; above from destroying that property?
  ;; Example: (or "a" (or "abc" "abd" "abe"))
  (cond
   ((null body)                    ; No items: a never-matching regexp.
    (rx--empty))
   ((null (cdr body))              ; Single item.
    (rx--translate (car body)))
   ((rx--every #'stringp body)     ; All strings.
    (cons (list (regexp-opt body nil t))
          t))
   (t
    (cons (append (car (rx--translate (car body)))
                  (mapcan (lambda (item)
                            (cons "\\|" (car (rx--translate item))))
                          (cdr body)))
          nil))))

(defun rx--string-to-intervals (str)
  "Decode STR as intervals: A-Z becomes (?A . ?Z), and the single
character X becomes (?X . ?X).  Return the intervals in a list."
  ;; We could just do string-to-multibyte on the string and work with
  ;; that instead of this `decode-char' workaround.
  (let ((decode-char
         (if (multibyte-string-p str)
             #'identity
           #'unibyte-char-to-multibyte))
        (len (length str))
        (i 0)
        (intervals nil))
    (while (< i len)
      (cond ((and (< i (- len 2))
                  (= (aref str (1+ i)) ?-))
             ;; Range.
             (let ((start (funcall decode-char (aref str i)))
                   (end   (funcall decode-char (aref str (+ i 2)))))
               (cond ((and (<= start #x7f) (>= end #x3fff80))
                      ;; Ranges between ASCII and raw bytes are split to
                      ;; avoid having them absorb Unicode characters
                      ;; caught in-between.
                      (push (cons start #x7f) intervals)
                      (push (cons #x3fff80 end) intervals))
                     ((<= start end)
                      (push (cons start end) intervals))
                     (t
                      (error "Invalid rx `any' range: %s"
                             (substring str i 3))))
               (setq i (+ i 3))))
            (t
             ;; Single character.
             (let ((char (funcall decode-char (aref str i))))
               (push (cons char char) intervals))
             (setq i (+ i 1)))))
    intervals))

(defun rx--condense-intervals (intervals)
  "Merge adjacent and overlapping intervals by mutation, preserving the order.
INTERVALS is a list of (START . END) with START ≤ END, sorted by START."
  (let ((tail intervals)
        d)
    (while (setq d (cdr tail))
      (if (>= (cdar tail) (1- (caar d)))
          (progn
            (setcdr (car tail) (max (cdar tail) (cdar d)))
            (setcdr tail (cdr d)))
        (setq tail d)))
    intervals))

(defun rx--translate-any (negated body)
  "Translate an (any ...) construct.  Return (REGEXP . PRECEDENCE).
If NEGATED, negate the sense."
  (let ((classes nil)
        (strings nil)
        (conses nil))
    ;; Collect strings, conses and characters, and classes in separate bins.
    (dolist (arg body)
      (cond ((stringp arg)
             (push arg strings))
            ((and (consp arg)
                  (characterp (car arg))
                  (characterp (cdr arg))
                  (<= (car arg) (cdr arg)))
             ;; Copy the cons, in case we need to modify it.
             (push (cons (car arg) (cdr arg)) conses))
            ((characterp arg)
             (push (cons arg arg) conses))
            ((and (symbolp arg)
                  (let ((class (cdr (assq arg rx--char-classes))))
                    (and class (push class classes)))))
            (t (error "Invalid rx `any' argument: %s" arg))))
    (let ((items
           ;; Translate strings and conses into nonoverlapping intervals,
           ;; and add classes as symbols at the end.
           (append
            (rx--condense-intervals
             (sort (append conses
                           (mapcan #'rx--string-to-intervals strings))
                   #'car-less-than-car))
            (reverse classes))))

      ;; Move lone ] and range ]-x to the start.
      (let ((rbrac-l (assq ?\] items)))
        (when rbrac-l
          (setq items (cons rbrac-l (delq rbrac-l items)))))

      ;; Split x-] and move the lone ] to the start.
      (let ((rbrac-r (rassq ?\] items)))
        (when (and rbrac-r (not (eq (car rbrac-r) ?\])))
          (setcdr rbrac-r ?\\)
          (setq items (cons '(?\] . ?\]) items))))

      ;; Split ,-- (which would end up as ,- otherwise).
      (let ((dash-r (rassq ?- items)))
        (when (eq (car dash-r) ?,)
          (setcdr dash-r ?,)
          (setq items (nconc items '((?- . ?-))))))

      ;; Remove - (lone or at start of interval)
      (let ((dash-l (assq ?- items)))
        (when dash-l
          (if (eq (cdr dash-l) ?-)
              (setq items (delq dash-l items))   ; Remove lone -
            (setcar dash-l ?.))                  ; Reduce --x to .-x
          (setq items (nconc items '((?- . ?-))))))

      ;; Deal with leading ^ and range ^-x.
      (when (and (consp (car items))
                 (eq (caar items) ?^)
                 (cdr items))
        ;; Move ^ and ^-x to second place.
        (setq items (cons (cadr items)
                          (cons (car items) (cddr items)))))

      (cond
       ;; Empty set: if negated, any char, otherwise match-nothing.
       ((null items)
        (if negated
            (rx--translate-symbol 'anything)
          (rx--empty)))
       ;; Single non-negated character.
       ((and (null (cdr items))
             (consp (car items))
             (eq (caar items) (cdar items))
             (not negated))
        (cons (list (regexp-quote (char-to-string (caar items))))
              t))
       ;; At least one character or class, possibly negated.
       (t
        (cons
         (list
          (concat
           "["
           (and negated "^")
           (mapconcat (lambda (item)
                        (cond ((symbolp item)
                               (format "[:%s:]" item))
                              ((eq (car item) (cdr item))
                               (char-to-string (car item)))
                              ((eq (1+ (car item)) (cdr item))
                               (string (car item) (cdr item)))
                              (t
                               (string (car item) ?- (cdr item)))))
                      items nil)
           "]"))
         t))))))

(defun rx--translate-not (negated body)
  "Translate a (not ...) construct.  Return (REGEXP . PRECEDENCE).
If NEGATED, negate the sense (thus making it positive)."
  (unless (and body (null (cdr body)))
    (error "rx `not' form takes exactly one argument"))
  (let ((arg (car body)))
    (cond
     ((consp arg)
      (pcase (car arg)
        ((or 'any 'in 'char) (rx--translate-any      (not negated) (cdr arg)))
        ('syntax             (rx--translate-syntax   (not negated) (cdr arg)))
        ('category           (rx--translate-category (not negated) (cdr arg)))
        ('not                (rx--translate-not      (not negated) (cdr arg)))
        (_ (error "Illegal argument to rx `not': %S" arg))))
     ((eq arg 'word-boundary)
      (rx--translate-symbol
       (if negated 'word-boundary 'not-word-boundary)))
     (t
      (let ((class (cdr (assq arg rx--char-classes))))
        (if class
            (rx--translate-any (not negated) (list class))
          (error "Illegal argument to rx `not': %s" arg)))))))

(defun rx--atomic-regexp (item)
  "ITEM is (REGEXP . PRECEDENCE); return a regexp of precedence t."
  (if (eq (cdr item) t)
      (car item)
    (rx--bracket (car item))))

(defun rx--translate-counted-repetition (min-count max-count body)
  (let ((operand (rx--translate-seq body)))
    (if (car operand)
        (cons (append
               (rx--atomic-regexp operand)
               (list (concat "\\{"
                             (number-to-string min-count)
                             (cond ((null max-count) ",")
                                   ((< min-count max-count)
                                    (concat "," (number-to-string max-count))))
                             "\\}")))
              t)
      operand)))

(defun rx--check-repeat-arg (name min-args body)
  (unless (>= (length body) min-args)
    (error "rx `%s' requires at least %d argument%s"
           name min-args (if (= min-args 1) "" "s")))
  ;; There seems to be no reason to disallow zero counts.
  (unless (natnump (car body))
    (error "rx `%s' first argument must be nonnegative" name)))

(defun rx--translate-bounded-repetition (name body)
  (let ((min-count (car body))
        (max-count (cadr body))
        (items (cddr body)))
    (unless (and (natnump min-count)
                 (natnump max-count)
                 (<= min-count max-count))
      (error "rx `%s' range error" name))
    (rx--translate-counted-repetition min-count max-count items)))

(defun rx--translate-repeat (body)
  (rx--check-repeat-arg 'repeat 2 body)
  (if (= (length body) 2)
      (rx--translate-counted-repetition (car body) (car body) (cdr body))
    (rx--translate-bounded-repetition 'repeat body)))

(defun rx--translate-** (body)
  (rx--check-repeat-arg '** 2 body)
  (rx--translate-bounded-repetition '** body))

(defun rx--translate->= (body)
  (rx--check-repeat-arg '>= 1 body)
  (rx--translate-counted-repetition (car body) nil (cdr body)))

(defun rx--translate-= (body)
  (rx--check-repeat-arg '= 1 body)
  (rx--translate-counted-repetition (car body) (car body) (cdr body)))

(defvar rx--greedy t)

(defun rx--translate-rep (op-string greedy body)
  "Translate a repetition; OP-STRING is one of \"*\", \"+\" or \"?\".
GREEDY is a boolean.  Return (REGEXP . PRECEDENCE)."
  (let ((operand (rx--translate-seq body)))
    (if (car operand)
        (cons (append (rx--atomic-regexp operand)
                      (list (concat op-string (unless greedy "?"))))
              ;; The result has precedence seq to avoid (? (* "a")) -> "a*?"
              'seq)
      operand)))

(defun rx--control-greedy (greedy body)
  "Translate the sequence BODY with greediness GREEDY.
Return (REGEXP . PRECEDENCE)."
  (let ((rx--greedy greedy))
    (rx--translate-seq body)))

(defun rx--translate-group (body)
  "Translate the `group' form.  Return (REGEXP . PRECEDENCE)."
  (cons (rx--enclose "\\("
                     (car (rx--translate-seq body))
                     "\\)")
        t))

(defun rx--translate-group-n (body)
  "Translate the `group-n' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and (integerp (car body)) (> (car body) 0))
    (error "rx `group-n' requires a positive number as first argument"))
  (cons (rx--enclose (concat "\\(?" (number-to-string (car body)) ":")
                     (car (rx--translate-seq (cdr body)))
                     "\\)")
        t))

(defun rx--translate-backref (body)
  "Translate the `backref' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and (= (length body) 1) (integerp (car body)) (<= 1 (car body) 9))
    (error "rx `backref' requires an argument in the range 1..9"))
  (cons (list "\\" (number-to-string (car body))) t))

(defconst rx--syntax-codes
  '((whitespace         . ?-)           ; SPC also accepted
    (punctuation        . ?.)
    (word               . ?w)           ; W also accepted
    (symbol             . ?_)
    (open-parenthesis   . ?\()
    (close-parenthesis  . ?\))
    (expression-prefix  . ?\')
    (string-quote       . ?\")
    (paired-delimiter   . ?$)
    (escape             . ?\\)
    (character-quote    . ?/)
    (comment-start      . ?<)
    (comment-end        . ?>)
    (string-delimiter   . ?|)
    (comment-delimiter  . ?!)))

(defun rx--translate-syntax (negated body)
  "Translate the `syntax' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and body (null (cdr body)))
    (error "rx `syntax' form takes exactly one argument"))
  (let* ((sym (car body))
         (syntax (cdr (assq sym rx--syntax-codes))))
    (unless syntax
      (cond
       ;; Syntax character directly (sregex compatibility)
       ((and (characterp sym) (rassq sym rx--syntax-codes))
        (setq syntax sym))
       ;; Syntax character as symbol (sregex compatibility)
       ((symbolp sym)
        (let ((name (symbol-name sym)))
          (when (= (length name) 1)
            (let ((char (string-to-char name)))
              (when (rassq char rx--syntax-codes)
                (setq syntax char)))))))
      (unless syntax
        (error "Unknown rx syntax name `%s'" sym)))
    (cons (list (string ?\\ (if negated ?S ?s) syntax))
          t)))

(defconst rx--categories
  '((space-for-indent           . ?\s)
    (base                       . ?.)
    (consonant                  . ?0)
    (base-vowel                 . ?1)
    (upper-diacritical-mark     . ?2)
    (lower-diacritical-mark     . ?3)
    (tone-mark                  . ?4)
    (symbol                     . ?5)
    (digit                      . ?6)
    (vowel-modifying-diacritical-mark . ?7)
    (vowel-sign                 . ?8)
    (semivowel-lower            . ?9)
    (not-at-end-of-line         . ?<)
    (not-at-beginning-of-line   . ?>)
    (alpha-numeric-two-byte     . ?A)
    (chinese-two-byte           . ?C)
    (chinse-two-byte            . ?C)   ; A typo in Emacs 21.1-24.3.
    (greek-two-byte             . ?G)
    (japanese-hiragana-two-byte . ?H)
    (indian-two-byte            . ?I)
    (japanese-katakana-two-byte . ?K)
    (strong-left-to-right       . ?L)
    (korean-hangul-two-byte     . ?N)
    (strong-right-to-left       . ?R)
    (cyrillic-two-byte          . ?Y)
    (combining-diacritic        . ?^)
    (ascii                      . ?a)
    (arabic                     . ?b)
    (chinese                    . ?c)
    (ethiopic                   . ?e)
    (greek                      . ?g)
    (korean                     . ?h)
    (indian                     . ?i)
    (japanese                   . ?j)
    (japanese-katakana          . ?k)
    (latin                      . ?l)
    (lao                        . ?o)
    (tibetan                    . ?q)
    (japanese-roman             . ?r)
    (thai                       . ?t)
    (vietnamese                 . ?v)
    (hebrew                     . ?w)
    (cyrillic                   . ?y)
    (can-break                  . ?|)))

(defun rx--translate-category (negated body)
  "Translate the `category' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and body (null (cdr body)))
    (error "rx `category' form takes exactly one argument"))
  (let* ((arg (car body))
         (category
          (cond ((symbolp arg)
                 (let ((cat (assq arg rx--categories)))
                   (unless cat
                     (error "Unknown rx category `%s'" arg))
                   (cdr cat)))
                ((characterp arg) arg)
                (t (error "Invalid rx `category' argument `%s'" arg)))))
    (cons (list (string ?\\ (if negated ?C ?c) category))
          t)))

(defvar rx--delayed-evaluation nil
  "Whether to allow certain forms to be evaluated at runtime.")

(defun rx--translate-literal (body)
  "Translate the `literal' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and body (null (cdr body)))
    (error "rx `literal' form takes exactly one argument"))
  (let ((arg (car body)))
    (cond ((stringp arg)
           (cons (list (regexp-quote arg)) (if (= (length arg) 1) t 'seq)))
          (rx--delayed-evaluation
           (cons (list (list 'regexp-quote arg)) 'seq))
          (t (error "rx `literal' form with non-string argument")))))

(defun rx--translate-eval (body)
  "Translate the `eval' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and body (null (cdr body)))
    (error "rx `eval' form takes exactly one argument"))
  (rx--translate (eval (car body))))

(defvar rx--regexp-atomic-regexp nil)

(defun rx--translate-regexp (body)
  "Translate the `regexp' form.  Return (REGEXP . PRECEDENCE)."
  (unless (and body (null (cdr body)))
    (error "rx `regexp' form takes exactly one argument"))
  (let ((arg (car body)))
    (cond ((stringp arg)
           ;; Generate the regexp when needed, since rx isn't
           ;; necessarily present in the byte-compilation environment.
           (unless rx--regexp-atomic-regexp
             (setq rx--regexp-atomic-regexp
                   ;; Match atomic (precedence t) regexps: may give
                   ;; false negatives but no false positives, assuming
                   ;; the target string is syntactically correct.
                   (rx-to-string
                    '(seq
                      bos
                      (or (seq "["
                               (opt "^")
                               (opt "]")
                               (* (or (seq "[:" (+ (any "a-z")) ":]")
                                      (not (any "]"))))
                               "]")
                          anything
                          (seq "\\"
                               (or anything
                                   (seq (any "sScC_") anything)
                                   (seq "("
                                        (* (or (not (any "\\"))
                                               (seq "\\" (not (any ")")))))
                                        "\\)"))))
                      eos)
                    t)))
           (cons (list arg)
                 (if (string-match-p rx--regexp-atomic-regexp arg) t nil)))
          (rx--delayed-evaluation
           (cons (list arg) nil))
          (t (error "rx `regexp' form with non-string argument")))))

(defun rx--translate-compat-form (def form)
  "Translate a compatibility form from `rx-constituents'.
DEF is the definition tuple.  Return (REGEXP . PRECEDENCE)."
  (let* ((fn (nth 0 def))
         (min-args (nth 1 def))
         (max-args (nth 2 def))
         (predicate (nth 3 def))
         (nargs (1- (length form))))
    (when (< nargs min-args)
      (error "The `%s' form takes at least %d argument(s)"
             (car form) min-args))
    (when (and max-args (> nargs max-args))
      (error "The `%s' form takes at most %d argument(s)"
             (car form) max-args))
    (when (and predicate (not (rx--every predicate (cdr form))))
      (error "The `%s' form requires arguments satisfying `%s'"
             (car form) predicate))
    (let ((regexp (funcall fn form)))
      (unless (stringp regexp)
        (error "The `%s' form did not expand to a string" (car form)))
      (cons (list regexp) nil))))

(defun rx--translate-form (form)
  "Translate an rx form (list structure).  Return (REGEXP . PRECEDENCE)."
  (let ((body (cdr form)))
    (pcase (car form)
      ((or 'seq : 'and 'sequence) (rx--translate-seq body))
      ((or 'or '|)              (rx--translate-or body))
      ((or 'any 'in 'char)      (rx--translate-any nil body))
      ('not-char                (rx--translate-any t body))
      ('not                     (rx--translate-not nil body))

      ('repeat                  (rx--translate-repeat body))
      ('=                       (rx--translate-= body))
      ('>=                      (rx--translate->= body))
      ('**                      (rx--translate-** body))

      ((or 'zero-or-more '0+)           (rx--translate-rep "*" rx--greedy body))
      ((or 'one-or-more '1+)            (rx--translate-rep "+" rx--greedy body))
      ((or 'zero-or-one 'opt 'optional) (rx--translate-rep "?" rx--greedy body))

      ('*                       (rx--translate-rep "*" t body))
      ('+                       (rx--translate-rep "+" t body))
      ((or '\? ?\s)             (rx--translate-rep "?" t body))

      ('*?                      (rx--translate-rep "*" nil body))
      ('+?                      (rx--translate-rep "+" nil body))
      ((or '\?? ??)             (rx--translate-rep "?" nil body))

      ('minimal-match           (rx--control-greedy nil body))
      ('maximal-match           (rx--control-greedy t   body))

      ((or 'group 'submatch)     (rx--translate-group body))
      ((or 'group-n 'submatch-n) (rx--translate-group-n body))
      ('backref                  (rx--translate-backref body))

      ('syntax                  (rx--translate-syntax nil body))
      ('not-syntax              (rx--translate-syntax t body))
      ('category                (rx--translate-category nil body))

      ('literal                 (rx--translate-literal body))
      ('eval                    (rx--translate-eval body))
      ((or 'regexp 'regex)      (rx--translate-regexp body))

      (op
       (unless (symbolp op)
         (error "Bad rx operator `%S'" op))

       ;; For compatibility with old rx.
       (let ((entry (assq op rx-constituents)))
         (if (progn
               (while (and entry (not (consp (cdr entry))))
                 (setq entry
                       (if (symbolp (cdr entry))
                           ;; Alias for another entry.
                           (assq (cdr entry) rx-constituents)
                         ;; Wrong type, try further down the list.
                         (assq (car entry)
                               (cdr (memq entry rx-constituents))))))
               entry)
             (rx--translate-compat-form (cdr entry) form)
           (error "Unknown rx form `%s'" op)))))))

;; Defined here rather than in re-builder to lower the odds that it
;; will be kept in sync with changes.
(defconst rx--builtin-forms
  '(seq sequence : and or | any in char not-char not
    repeat = >= **
    zero-or-more 0+ *
    one-or-more 1+ +
    zero-or-one opt optional \?
    *? +? \??
    minimal-match maximal-match
    group submatch group-n submatch-n backref
    syntax not-syntax category
    literal eval regexp regex)
  "List of built-in rx forms.  For use in re-builder only.")

(defun rx--translate (item)
  "Translate the rx-expression ITEM.  Return (REGEXP . PRECEDENCE)."
  (cond
   ((stringp item)
    (if (= (length item) 0)
        (cons nil 'seq)
      (cons (list (regexp-quote item)) (if (= (length item) 1) t 'seq))))
   ((characterp item)
    (cons (list (regexp-quote (char-to-string item))) t))
   ((symbolp item)
    (rx--translate-symbol item))
   ((consp item)
    (rx--translate-form item))
   (t (error "Bad rx expression: %S" item))))


;;;###autoload
(defun rx-to-string (form &optional no-group)
  "Translate FORM from `rx' sexp syntax into a string regexp.
The arguments to `literal' and `regexp' forms inside FORM must be
constant strings.
If NO-GROUP is non-nil, don't bracket the result in a non-capturing
group."
  (let* ((item (rx--translate form))
         (exprs (if no-group
                    (car item)
                  (rx--atomic-regexp item))))
    (apply #'concat exprs)))

(defun rx--to-expr (form)
  "Translate the rx-expression FORM to a Lisp expression yielding a regexp."
  (let* ((rx--delayed-evaluation t)
         (elems (car (rx--translate form)))
         (args nil))
    ;; Merge adjacent strings.
    (while elems
      (let ((strings nil))
        (while (and elems (stringp (car elems)))
          (push (car elems) strings)
          (setq elems (cdr elems)))
        (let ((s (apply #'concat (nreverse strings))))
          (unless (zerop (length s))
            (push s args))))
      (when elems
        (push (car elems) args)
        (setq elems (cdr elems))))
    (cond ((null args) "")                             ; 0 args
          ((cdr args) (cons 'concat (nreverse args)))  ; ≥2 args
          (t (car args)))))                            ; 1 arg


;;;###autoload
(defmacro rx (&rest regexps)
  "Translate regular expressions REGEXPS in sexp form to a regexp string.
Each argument is one of the forms below; RX is a subform, and RX... stands
for one or more RXs.  For details, see Info node `(elisp) Rx Notation'.
See `rx-to-string' for the corresponding function.

STRING         Match a literal string.
CHAR           Match a literal character.

(seq RX...)    Match the RXs in sequence.  Alias: :, sequence, and.
(or RX...)     Match one of the RXs.  Alias: |.

(zero-or-more RX...) Match RXs zero or more times.  Alias: 0+.
(one-or-more RX...)  Match RXs one or more times.  Alias: 1+.
(zero-or-one RX...)  Match RXs or the empty string.  Alias: opt, optional.
(* RX...)       Match RXs zero or more times; greedy.
(+ RX...)       Match RXs one or more times; greedy.
(? RX...)       Match RXs or the empty string; greedy.
(*? RX...)      Match RXs zero or more times; non-greedy.
(+? RX...)      Match RXs one or more times; non-greedy.
(?? RX...)      Match RXs or the empty string; non-greedy.
(= N RX...)     Match RXs exactly N times.
(>= N RX...)    Match RXs N or more times.
(** N M RX...)  Match RXs N to M times.  Alias: repeat.
(minimal-match RX)  Match RX, with zero-or-more, one-or-more, zero-or-one
                and aliases using non-greedy matching.
(maximal-match RX)  Match RX, with zero-or-more, one-or-more, zero-or-one
                and aliases using greedy matching, which is the default.

(any SET...)    Match a character from one of the SETs.  Each SET is a
                character, a string, a range as string \"A-Z\" or cons
                (?A . ?Z), or a character class (see below).  Alias: in, char.
(not CHARSPEC)  Match one character not matched by CHARSPEC.  CHARSPEC
                can be (any ...), (syntax ...), (category ...),
                or a character class.
not-newline     Match any character except a newline.  Alias: nonl.
anything        Match any character.

CHARCLASS       Match a character from a character class.  One of:
 alpha, alphabetic, letter   Alphabetic characters (defined by Unicode).
 alnum, alphanumeric         Alphabetic or decimal digit chars (Unicode).
 digit numeric, num          0-9.
 xdigit, hex-digit, hex      0-9, A-F, a-f.
 cntrl, control              ASCII codes 0-31.
 blank                       Horizontal whitespace (Unicode).
 space, whitespace, white    Chars with whitespace syntax.
 lower, lower-case           Lower-case chars, from current case table.
 upper, upper-case           Upper-case chars, from current case table.
 graph, graphic              Graphic characters (Unicode).
 print, printing             Whitespace or graphic (Unicode).
 punct, punctuation          Not control, space, letter or digit (ASCII);
                              not word syntax (non-ASCII).
 word, wordchar              Characters with word syntax.
 ascii                       ASCII characters (codes 0-127).
 nonascii                    Non-ASCII characters (but not raw bytes).

(syntax SYNTAX)  Match a character with syntax SYNTAX, being one of:
  whitespace, punctuation, word, symbol, open-parenthesis,
  close-parenthesis, expression-prefix, string-quote,
  paired-delimiter, escape, character-quote, comment-start,
  comment-end, string-delimiter, comment-delimiter

(category CAT)   Match a character in category CAT, being one of:
  space-for-indent, base, consonant, base-vowel,
  upper-diacritical-mark, lower-diacritical-mark, tone-mark, symbol,
  digit, vowel-modifying-diacritical-mark, vowel-sign,
  semivowel-lower, not-at-end-of-line, not-at-beginning-of-line,
  alpha-numeric-two-byte, chinese-two-byte, greek-two-byte,
  japanese-hiragana-two-byte, indian-two-byte,
  japanese-katakana-two-byte, strong-left-to-right,
  korean-hangul-two-byte, strong-right-to-left, cyrillic-two-byte,
  combining-diacritic, ascii, arabic, chinese, ethiopic, greek,
  korean, indian, japanese, japanese-katakana, latin, lao,
  tibetan, japanese-roman, thai, vietnamese, hebrew, cyrillic,
  can-break

Zero-width assertions: these all match the empty string in specific places.
 line-start         At the beginning of a line.  Alias: bol.
 line-end           At the end of a line.  Alias: eol.
 string-start       At the start of the string or buffer.
                     Alias: buffer-start, bos, bot.
 string-end         At the end of the string or buffer.
                     Alias: buffer-end, eos, eot.
 point              At point.
 word-start         At the beginning of a word.
 word-end           At the end of a word.
 word-boundary      At the beginning or end of a word.
 not-word-boundary  Not at the beginning or end of a word.
 symbol-start       At the beginning of a symbol.
 symbol-end         At the end of a symbol.

(group RX...)  Match RXs and define a capture group.  Alias: submatch.
(group-n N RX...) Match RXs and define capture group N.  Alias: submatch-n.
(backref N)    Match the text that capture group N matched.

(literal EXPR) Match the literal string from evaluating EXPR at run time.
(regexp EXPR)  Match the string regexp from evaluating EXPR at run time.
(eval EXPR)    Match the rx sexp from evaluating EXPR at compile time.

\(fn REGEXPS...)"
  (rx--to-expr (cons 'seq regexps)))


;; During `rx--pcase-transform', list of defined variables in right-to-left
;; order.
(defvar rx--pcase-vars)

(defun rx--pcase-transform (rx)
  "Transform RX, an rx-expression augmented with `let' and named `backref',
into a plain rx-expression, collecting names into `rx--pcase-vars'."
  (pcase rx
    (`(let ,name . ,body)
     (let* ((index (length (memq name rx--pcase-vars)))
            (i (if (zerop index)
                   (length (push name rx--pcase-vars))
                 index)))
       `(group-n ,i ,(rx--pcase-transform (cons 'seq body)))))
    ((and `(backref ,ref)
          (guard (symbolp ref)))
     (let ((index (length (memq ref rx--pcase-vars))))
       (when (zerop index)
         (error "rx `backref' variable must be one of: %s"
                (mapconcat #'symbol-name rx--pcase-vars " ")))
       `(backref ,index)))
    ((and `(,head . ,rest)
          (guard (and (symbolp head)
                      (not (memq head '(literal regexp regex eval))))))
     (cons head (mapcar #'rx--pcase-transform rest)))
    (_ rx)))

(pcase-defmacro rx (&rest regexps)
  "A pattern that matches strings against `rx' REGEXPS in sexp form.
REGEXPS are interpreted as in `rx'.  The pattern matches any
string that is a match for REGEXPS, as if by `string-match'.

In addition to the usual `rx' syntax, REGEXPS can contain the
following constructs:

  (let REF RX...)  binds the symbol REF to a submatch that matches
                   the regular expressions RX.  REF is bound in
                   CODE to the string of the submatch or nil, but
                   can also be used in `backref'.
  (backref REF)    matches whatever the submatch REF matched.
                   REF can be a number, as usual, or a name
                   introduced by a previous (let REF ...)
                   construct."
  (let* ((rx--pcase-vars nil)
         (regexp (rx--to-expr (rx--pcase-transform (cons 'seq regexps)))))
    `(and (pred (string-match ,regexp))
          ,@(let ((i 0))
              (mapcar (lambda (name)
                        (setq i (1+ i))
                        `(app (match-string ,i) ,name))
                      (reverse rx--pcase-vars))))))

(provide 'rx)

;;; rx.el ends here