| 1 | """
|
| 2 | tdop.py - Library for expression parsing.
|
| 3 | """
|
| 4 |
|
| 5 | from _devbuild.gen.id_kind_asdl import Id, Id_t
|
| 6 | from _devbuild.gen.syntax_asdl import (loc, arith_expr, arith_expr_e,
|
| 7 | arith_expr_t, word_e, word_t,
|
| 8 | CompoundWord, Token)
|
| 9 | from core.error import p_die
|
| 10 | from core import ui
|
| 11 | from mycpp import mylib
|
| 12 | from mycpp.mylib import tagswitch
|
| 13 | from osh import word_
|
| 14 |
|
| 15 | from typing import (Callable, List, Dict, Tuple, Any, cast, TYPE_CHECKING)
|
| 16 |
|
| 17 | if TYPE_CHECKING: # break circular dep
|
| 18 | from osh.word_parse import WordParser
|
| 19 | from core import optview
|
| 20 | LeftFunc = Callable[['TdopParser', word_t, arith_expr_t, int],
|
| 21 | arith_expr_t]
|
| 22 | NullFunc = Callable[['TdopParser', word_t, int], arith_expr_t]
|
| 23 |
|
| 24 |
|
| 25 | def IsIndexable(node):
|
| 26 | # type: (arith_expr_t) -> bool
|
| 27 | """In POSIX shell arith, a[1] is allowed but a[1][1] isn't.
|
| 28 |
|
| 29 | We also allow $a[i] and foo$x[i] (formerly parse_dynamic_arith)
|
| 30 | """
|
| 31 | with tagswitch(node) as case:
|
| 32 | if case(arith_expr_e.VarSub, arith_expr_e.Word):
|
| 33 | return True
|
| 34 | return False
|
| 35 |
|
| 36 |
|
| 37 | def CheckLhsExpr(node, blame_word):
|
| 38 | # type: (arith_expr_t, word_t) -> None
|
| 39 | """Determine if a node is a valid L-value by whitelisting tags.
|
| 40 |
|
| 41 | Valid:
|
| 42 | x = y
|
| 43 | a[1] = y
|
| 44 | Invalid:
|
| 45 | a[0][0] = y
|
| 46 | """
|
| 47 | UP_node = node
|
| 48 | if node.tag() == arith_expr_e.Binary:
|
| 49 | node = cast(arith_expr.Binary, UP_node)
|
| 50 | if node.op.id == Id.Arith_LBracket and IsIndexable(node.left):
|
| 51 | return
|
| 52 | # But a[0][0] = 1 is NOT valid.
|
| 53 |
|
| 54 | if IsIndexable(node):
|
| 55 | return
|
| 56 |
|
| 57 | p_die("Left-hand side of this assignment is invalid", loc.Word(blame_word))
|
| 58 |
|
| 59 |
|
| 60 | #
|
| 61 | # Null Denotation
|
| 62 | #
|
| 63 |
|
| 64 |
|
| 65 | def NullError(p, t, bp):
|
| 66 | # type: (TdopParser, word_t, int) -> arith_expr_t
|
| 67 | # TODO: I need position information
|
| 68 | p_die("Token can't be used in prefix position", loc.Word(t))
|
| 69 | return None # never reached
|
| 70 |
|
| 71 |
|
| 72 | def NullConstant(p, w, bp):
|
| 73 | # type: (TdopParser, word_t, int) -> arith_expr_t
|
| 74 | name_tok = word_.LooksLikeArithVar(w)
|
| 75 | if name_tok:
|
| 76 | return name_tok
|
| 77 |
|
| 78 | # Id.Word_Compound in the spec ensures this cast is valid
|
| 79 | return cast(CompoundWord, w)
|
| 80 |
|
| 81 |
|
| 82 | def NullParen(p, t, bp):
|
| 83 | # type: (TdopParser, word_t, int) -> arith_expr_t
|
| 84 | """Arithmetic grouping."""
|
| 85 | r = p.ParseUntil(bp)
|
| 86 | p.Eat(Id.Arith_RParen)
|
| 87 | return r
|
| 88 |
|
| 89 |
|
| 90 | def NullPrefixOp(p, w, bp):
|
| 91 | # type: (TdopParser, word_t, int) -> arith_expr_t
|
| 92 | """Prefix operator.
|
| 93 |
|
| 94 | Low precedence: return, raise, etc.
|
| 95 | return x+y is return (x+y), not (return x) + y
|
| 96 |
|
| 97 | High precedence: logical negation, bitwise complement, etc.
|
| 98 | !x && y is (!x) && y, not !(x && y)
|
| 99 | """
|
| 100 | right = p.ParseUntil(bp)
|
| 101 | return arith_expr.Unary(word_.ArithId(w), right)
|
| 102 |
|
| 103 |
|
| 104 | #
|
| 105 | # Left Denotation
|
| 106 | #
|
| 107 |
|
| 108 |
|
| 109 | def LeftError(p, t, left, rbp):
|
| 110 | # type: (TdopParser, word_t, arith_expr_t, int) -> arith_expr_t
|
| 111 | # Hm is this not called because of binding power?
|
| 112 | p_die("Token can't be used in infix position", loc.Word(t))
|
| 113 | return None # never reached
|
| 114 |
|
| 115 |
|
| 116 | def LeftBinaryOp(p, w, left, rbp):
|
| 117 | # type: (TdopParser, word_t, arith_expr_t, int) -> arith_expr_t
|
| 118 | """Normal binary operator like 1+2 or 2*3, etc."""
|
| 119 |
|
| 120 | assert w.tag() == word_e.Operator, w
|
| 121 | tok = cast(Token, w)
|
| 122 |
|
| 123 | return arith_expr.Binary(tok, left, p.ParseUntil(rbp))
|
| 124 |
|
| 125 |
|
| 126 | def LeftAssign(p, w, left, rbp):
|
| 127 | # type: (TdopParser, word_t, arith_expr_t, int) -> arith_expr_t
|
| 128 | """Normal binary operator like 1+2 or 2*3, etc."""
|
| 129 | # x += 1, or a[i] += 1
|
| 130 |
|
| 131 | CheckLhsExpr(left, w)
|
| 132 | return arith_expr.BinaryAssign(word_.ArithId(w), left, p.ParseUntil(rbp))
|
| 133 |
|
| 134 |
|
| 135 | #
|
| 136 | # Parser definition
|
| 137 | # TODO: To be consistent, move this to osh/tdop_def.py.
|
| 138 | #
|
| 139 |
|
| 140 | if mylib.PYTHON:
|
| 141 |
|
| 142 | def _ModuleAndFuncName(f):
|
| 143 | # type: (Any) -> Tuple[str, str]
|
| 144 | namespace = f.__module__.split('.')[-1]
|
| 145 | return namespace, f.__name__
|
| 146 |
|
| 147 | def _CppFuncName(f):
|
| 148 | # type: (Any) -> str
|
| 149 | return '%s::%s' % _ModuleAndFuncName(f)
|
| 150 |
|
| 151 | class LeftInfo(object):
|
| 152 | """Row for operator.
|
| 153 |
|
| 154 | In C++ this should be a big array.
|
| 155 | """
|
| 156 |
|
| 157 | def __init__(self, led=None, lbp=0, rbp=0):
|
| 158 | # type: (LeftFunc, int, int) -> None
|
| 159 | self.led = led or LeftError
|
| 160 | self.lbp = lbp
|
| 161 | self.rbp = rbp
|
| 162 |
|
| 163 | def __str__(self):
|
| 164 | # type: () -> str
|
| 165 | """Used by C++ code generation."""
|
| 166 | return '{ %s, %d, %d },' % (_CppFuncName(
|
| 167 | self.led), self.lbp, self.rbp)
|
| 168 |
|
| 169 | def ModuleAndFuncName(self):
|
| 170 | # type: () -> Tuple[str, str]
|
| 171 | """Used by C++ code generation."""
|
| 172 | return _ModuleAndFuncName(self.led)
|
| 173 |
|
| 174 | class NullInfo(object):
|
| 175 | """Row for operator.
|
| 176 |
|
| 177 | In C++ this should be a big array.
|
| 178 | """
|
| 179 |
|
| 180 | def __init__(self, nud=None, bp=0):
|
| 181 | # type: (NullFunc, int) -> None
|
| 182 | self.nud = nud or LeftError
|
| 183 | self.bp = bp
|
| 184 |
|
| 185 | def __str__(self):
|
| 186 | # type: () -> str
|
| 187 | """Used by C++ code generation."""
|
| 188 | return '{ %s, %d },' % (_CppFuncName(self.nud), self.bp)
|
| 189 |
|
| 190 | def ModuleAndFuncName(self):
|
| 191 | # type: () -> Tuple[str, str]
|
| 192 | """Used by C++ code generation."""
|
| 193 | return _ModuleAndFuncName(self.nud)
|
| 194 |
|
| 195 | class ParserSpec(object):
|
| 196 | """Specification for a TDOP parser.
|
| 197 |
|
| 198 | This can be compiled to a table in C++.
|
| 199 | """
|
| 200 |
|
| 201 | def __init__(self):
|
| 202 | # type: () -> None
|
| 203 | self.nud_lookup = {} # type: Dict[Id_t, NullInfo]
|
| 204 | self.led_lookup = {} # type: Dict[Id_t, LeftInfo]
|
| 205 |
|
| 206 | def Null(self, bp, nud, tokens):
|
| 207 | # type: (int, NullFunc, List[Id_t]) -> None
|
| 208 | """Register a token that doesn't take anything on the left.
|
| 209 |
|
| 210 | Examples: constant, prefix operator, error.
|
| 211 | """
|
| 212 | for token in tokens:
|
| 213 | self.nud_lookup[token] = NullInfo(nud=nud, bp=bp)
|
| 214 | if token not in self.led_lookup:
|
| 215 | self.led_lookup[token] = LeftInfo() # error
|
| 216 |
|
| 217 | def _RegisterLed(self, lbp, rbp, led, tokens):
|
| 218 | # type: (int, int, LeftFunc, List[Id_t]) -> None
|
| 219 | for token in tokens:
|
| 220 | if token not in self.nud_lookup:
|
| 221 | self.nud_lookup[token] = NullInfo(NullError)
|
| 222 | self.led_lookup[token] = LeftInfo(lbp=lbp, rbp=rbp, led=led)
|
| 223 |
|
| 224 | def Left(self, bp, led, tokens):
|
| 225 | # type: (int, LeftFunc, List[Id_t]) -> None
|
| 226 | """Register a token that takes an expression on the left."""
|
| 227 | self._RegisterLed(bp, bp, led, tokens)
|
| 228 |
|
| 229 | def LeftRightAssoc(self, bp, led, tokens):
|
| 230 | # type: (int, LeftFunc, List[Id_t]) -> None
|
| 231 | """Register a right associative operator."""
|
| 232 | self._RegisterLed(bp, bp - 1, led, tokens)
|
| 233 |
|
| 234 | def LookupNud(self, token):
|
| 235 | # type: (Id_t) -> NullInfo
|
| 236 |
|
| 237 | # As long as the table is complete, this shouldn't fail
|
| 238 | return self.nud_lookup[token]
|
| 239 |
|
| 240 | def LookupLed(self, token):
|
| 241 | # type: (Id_t) -> LeftInfo
|
| 242 | """Get a left_info for the token."""
|
| 243 |
|
| 244 | # As long as the table is complete, this shouldn't fail
|
| 245 | return self.led_lookup[token]
|
| 246 |
|
| 247 |
|
| 248 | class TdopParser(object):
|
| 249 |
|
| 250 | def __init__(self, spec, w_parser, parse_opts):
|
| 251 | # type: (ParserSpec, WordParser, optview.Parse) -> None
|
| 252 | self.spec = spec
|
| 253 | self.w_parser = w_parser
|
| 254 | self.parse_opts = parse_opts
|
| 255 |
|
| 256 | # NOTE: Next() overwrites this state, so we don't need a Reset() method in
|
| 257 | # between reuses of this TdopParser instance.
|
| 258 | self.cur_word = None # type: word_t # current token
|
| 259 | self.op_id = Id.Undefined_Tok
|
| 260 |
|
| 261 | def CurrentId(self):
|
| 262 | # type: () -> Id_t
|
| 263 | """Glue used by the WordParser to check for extra tokens."""
|
| 264 | return word_.ArithId(self.cur_word)
|
| 265 |
|
| 266 | def AtToken(self, token_type):
|
| 267 | # type: (Id_t) -> bool
|
| 268 | return self.op_id == token_type
|
| 269 |
|
| 270 | def Eat(self, token_type):
|
| 271 | # type: (Id_t) -> None
|
| 272 | """Assert that we're at the current token and advance."""
|
| 273 | if not self.AtToken(token_type):
|
| 274 | p_die(
|
| 275 | 'Parser expected %s, got %s' %
|
| 276 | (ui.PrettyId(token_type), ui.PrettyId(self.op_id)),
|
| 277 | loc.Word(self.cur_word))
|
| 278 | self.Next()
|
| 279 |
|
| 280 | def Next(self):
|
| 281 | # type: () -> bool
|
| 282 | self.cur_word = self.w_parser.ReadArithWord()
|
| 283 | self.op_id = word_.ArithId(self.cur_word)
|
| 284 | return True
|
| 285 |
|
| 286 | def ParseUntil(self, rbp):
|
| 287 | # type: (int) -> arith_expr_t
|
| 288 | """Parse to the right, eating tokens until we encounter a token with
|
| 289 | binding power LESS THAN OR EQUAL TO rbp."""
|
| 290 | # TODO: use Kind.Eof
|
| 291 | if self.op_id in (Id.Eof_Real, Id.Eof_RParen, Id.Eof_Backtick):
|
| 292 | p_die('Unexpected end of input', loc.Word(self.cur_word))
|
| 293 |
|
| 294 | t = self.cur_word
|
| 295 | null_info = self.spec.LookupNud(self.op_id)
|
| 296 |
|
| 297 | self.Next() # skip over the token, e.g. ! ~ + -
|
| 298 | node = null_info.nud(self, t, null_info.bp)
|
| 299 |
|
| 300 | while True:
|
| 301 | t = self.cur_word
|
| 302 | left_info = self.spec.LookupLed(self.op_id)
|
| 303 |
|
| 304 | # Examples:
|
| 305 | # If we see 1*2+ , rbp = 27 and lbp = 25, so stop.
|
| 306 | # If we see 1+2+ , rbp = 25 and lbp = 25, so stop.
|
| 307 | # If we see 1**2**, rbp = 26 and lbp = 27, so keep going.
|
| 308 | if rbp >= left_info.lbp:
|
| 309 | break
|
| 310 | self.Next() # skip over the token, e.g. / *
|
| 311 |
|
| 312 | node = left_info.led(self, t, node, left_info.rbp)
|
| 313 |
|
| 314 | return node
|
| 315 |
|
| 316 | def Parse(self):
|
| 317 | # type: () -> arith_expr_t
|
| 318 |
|
| 319 | self.Next() # may raise ParseError
|
| 320 |
|
| 321 | if not self.parse_opts.parse_sh_arith():
|
| 322 | # Affects:
|
| 323 | # echo $(( x ))
|
| 324 | # ${a[i]} which should be $[a[i]] -- could have better error
|
| 325 | #
|
| 326 | # Note: sh_expr_eval.UnsafeArith has a dynamic e_die() check
|
| 327 | #
|
| 328 | # Doesn't affect:
|
| 329 | # printf -v x # unsafe_arith.ParseLValue
|
| 330 | # unset x # unsafe_arith.ParseLValue
|
| 331 | # ${!ref} # unsafe_arith.ParseVarRef
|
| 332 | # declare -n # not fully implemented yet
|
| 333 | #
|
| 334 | # a[i+1]= # parse_sh_assign
|
| 335 | #
|
| 336 | # (( a = 1 )) # parse_dparen
|
| 337 | # for (( i = 0; ... # parse_dparen
|
| 338 |
|
| 339 | p_die("POSIX shell arithmetic isn't allowed (parse_sh_arith)",
|
| 340 | loc.Word(self.cur_word))
|
| 341 |
|
| 342 | return self.ParseUntil(0)
|