ludic/runtime/native/regex_vm.ludic

# ============================================================================
# regex_vm.ludic — the Pike VM (Thompson NFA execution) and the public Regex.*
# API for the engine compiled in regex.ludic. Splits the "run + query" concern
# out of regex.ludic's "parse + compile". Spliced together (regex.ludic imports
# this file), so the two share the Prog/RNode model and OP_* opcodes.
# ============================================================================

# ---- Pike VM ----------------------------------------------------------------
property TList { pc: words, caps: words, n: int }
var rx_seen: words = null
var rx_gen:  int = 0
var rx_nc:   int = 0
var rx_code: words = null

function rx_add(list: TList, pc: int, caps: words, sp: int, s: pointer, slen: int) -> void {
  if rx_seen[pc] == rx_gen { return }
  rx_seen[pc] = rx_gen
  let op = rx_code[3 * pc]
  if op == OP_JMP { rx_add(list, rx_code[3 * pc + 1], caps, sp, s, slen);  return }
  if op == OP_SPLIT {
    rx_add(list, rx_code[3 * pc + 1], caps, sp, s, slen)
    rx_add(list, rx_code[3 * pc + 2], caps, sp, s, slen)
    return
  }
  if op == OP_SAVE {
    let slot = rx_code[3 * pc + 1]
    let old = caps[slot]
    caps[slot] = sp
    rx_add(list, pc + 1, caps, sp, s, slen)
    caps[slot] = old
    return
  }
  if op == OP_BOL {
    if sp == 0 { rx_add(list, pc + 1, caps, sp, s, slen) }
    return
  }
  if op == OP_EOL {
    if sp == slen { rx_add(list, pc + 1, caps, sp, s, slen) }
    else if sp == slen - 1 and (s[sp] & 255) == 10 { rx_add(list, pc + 1, caps, sp, s, slen) }
    return
  }
  # a leaf that consumes (CHAR/ANY/ANYNL/CLASS) or MATCH: record it
  let t = list.n
  list.pc[t] = pc
  var i = 0
  while i < rx_nc { list.caps[t * rx_nc + i] = caps[i];  i += 1 }
  list.n += 1
}

# run prog over s (length slen) from startpos; returns caps words or null
function rx_run(prog: Prog, s: pointer, slen: int, startpos: int) -> words {
  rx_code = prog.code.d
  rx_nc = 2 * (prog.ngroups + 1)
  let ncode = prog.code.n / 3
  rx_seen = words(ncode)
  var i = 0
  while i < ncode { rx_seen[i] = 0;  i += 1 }
  rx_gen = 0
  var clist = new TList
  clist.pc = words(ncode);  clist.caps = words(ncode * rx_nc);  clist.n = 0
  var nlist = new TList
  nlist.pc = words(ncode);  nlist.caps = words(ncode * rx_nc);  nlist.n = 0
  let wcaps = words(rx_nc)
  var matched: words = null

  rx_gen += 1
  i = 0
  while i < rx_nc { wcaps[i] = -1;  i += 1 }
  rx_add(clist, 0, wcaps, startpos, s, slen)

  var sp = startpos
  while true {
    if clist.n == 0 { break }
    var c = -1
    if sp < slen { c = s[sp] & 255 }
    nlist.n = 0
    rx_gen += 1
    var ti = 0
    var stop = false
    while ti < clist.n and not stop {
      let pc = clist.pc[ti]
      var k = 0
      while k < rx_nc { wcaps[k] = clist.caps[ti * rx_nc + k];  k += 1 }
      let op = rx_code[3 * pc]
      if op == OP_CHAR {
        if c >= 0 and c == rx_code[3 * pc + 1] { rx_add(nlist, pc + 1, wcaps, sp + 1, s, slen) }
      } else if op == OP_ANY {
        if c >= 0 and c != '\n' { rx_add(nlist, pc + 1, wcaps, sp + 1, s, slen) }
      } else if op == OP_ANYNL {
        if c >= 0 { rx_add(nlist, pc + 1, wcaps, sp + 1, s, slen) }
      } else if op == OP_CLASS {
        if c >= 0 and rx_class_has(prog, rx_code[3 * pc + 1], c) { rx_add(nlist, pc + 1, wcaps, sp + 1, s, slen) }
      } else if op == OP_MATCH {
        if matched == null { matched = words(rx_nc) }
        k = 0
        while k < rx_nc { matched[k] = wcaps[k];  k += 1 }
        stop = true
      }
      ti += 1
    }
    let tmp = clist;  clist = nlist;  nlist = tmp
    if sp >= slen { break }
    sp += 1
  }
  return matched
}

# ---- public API -------------------------------------------------------------
property Match { str: pointer = null, ng: int = 0, caps: words = null }

function regex_matches(str: pointer, pattern: pointer) -> bool {
  let p = regex_compile(pattern)
  if p == null { return false }
  return rx_run(p, str, rx_slen(str), 0) != null
}
function regex_test(str: pointer, re: Prog) -> bool {
  if re == null { return false }
  return rx_run(re, str, rx_slen(str), 0) != null
}
function regex_exec(str: pointer, re: Prog) -> Match {
  if re == null { return null }
  let caps = rx_run(re, str, rx_slen(str), 0)
  if caps == null { return null }
  let m = new Match
  m.str = str;  m.ng = re.ngroups;  m.caps = caps
  return m
}
function regex_find(str: pointer, pattern: pointer) -> Match {
  let p = regex_compile(pattern)
  if p == null { return null }
  return regex_exec(str, p)
}
function regex_next(str: pointer, re: Prog, from: int) -> Match {
  if re == null { return null }
  let n = rx_slen(str)
  if from > n { return null }
  let caps = rx_run(re, str, n, from)
  if caps == null { return null }
  let m = new Match
  m.str = str;  m.ng = re.ngroups;  m.caps = caps
  return m
}
function regex_ok(m: Match) -> bool { return m != null }
function regex_group_count(m: Match) -> int { if m == null { return 0 };  return m.ng }
function regex_start(m: Match, n: int) -> int {
  if m == null or n < 0 or n > m.ng { return -1 }
  return m.caps[2 * n]
}
function regex_end(m: Match, n: int) -> int {
  if m == null or n < 0 or n > m.ng { return -1 }
  return m.caps[2 * n + 1]
}
function regex_group(m: Match, n: int) -> string {
  if m == null or n < 0 or n > m.ng { return "" }
  let a = m.caps[2 * n]
  let b = m.caps[2 * n + 1]
  if a < 0 or b < 0 { return "" }
  let out = bytes(b - a + 1)
  var i = 0
  while i < b - a { out[i] = m.str[a + i];  i += 1 }
  out[b - a] = 0
  return out
}
function regex_valid(pattern: pointer) -> bool { return regex_compile(pattern) != null }

# regex_replace(str, pattern, repl): replace all non-overlapping matches.
# repl expands \0..\9 (groups; \0 = whole match) and \\ (a literal backslash).
function rx_append(out: IVec, s: pointer, a: int, b: int) -> void {
  var i = a
  while i < b { iv_push(out, s[i] & 255);  i += 1 }
}
function regex_replace(str: pointer, pattern: pointer, repl: pointer) -> string {
  let p = regex_compile(pattern)
  if p == null { return str }
  let n = rx_slen(str)
  let rn = rx_slen(repl)
  let out = iv_new()                              # bytes accumulated as ints
  var pos = 0
  var prev = 0
  while pos <= n {
    let caps = rx_run(p, str, n, pos)
    if caps == null { break }
    let ms = caps[0]
    let me = caps[1]
    rx_append(out, str, prev, ms)                 # text before the match
    # expand replacement
    var i = 0
    while i < rn {
      if repl[i] == '\\' and i + 1 < rn {
        let d = repl[i + 1] & 255
        if d >= '0' and d <= '9' {
          let g = d - 48
          if g <= p.ngroups { let a = caps[2 * g];  let b = caps[2 * g + 1];  if a >= 0 and b >= 0 { rx_append(out, str, a, b) } }
          i += 2
        } else if d == '\\' { iv_push(out, 92);  i += 2 }
        else { iv_push(out, d);  i += 2 }
      } else {
        iv_push(out, repl[i] & 255);  i += 1
      }
    }
    prev = me
    if me > pos { pos = me } else { if me < n { iv_push(out, str[me] & 255) };  prev = me + 1;  pos = me + 1 }
  }
  rx_append(out, str, prev, n)                     # trailing text
  let res = bytes(out.n + 1)
  var i = 0
  while i < out.n { res[i] = out.d[i];  i += 1 }
  res[out.n] = 0
  return res
}