1 | #!/usr/bin/env python2
|
2 | # Copyright 2016 Andy Chu. All rights reserved.
|
3 | # Licensed under the Apache License, Version 2.0 (the "License");
|
4 | # you may not use this file except in compliance with the License.
|
5 | # You may obtain a copy of the License at
|
6 | #
|
7 | # http://www.apache.org/licenses/LICENSE-2.0
|
8 | """
|
9 | sh_expr_eval.py -- Shell boolean and arithmetic expressions.
|
10 | """
|
11 |
|
12 | from _devbuild.gen.id_kind_asdl import Id
|
13 | from _devbuild.gen.runtime_asdl import scope_t
|
14 | from _devbuild.gen.syntax_asdl import (
|
15 | word_t,
|
16 | CompoundWord,
|
17 | Token,
|
18 | loc,
|
19 | loc_t,
|
20 | source,
|
21 | arith_expr,
|
22 | arith_expr_e,
|
23 | arith_expr_t,
|
24 | bool_expr,
|
25 | bool_expr_e,
|
26 | bool_expr_t,
|
27 | sh_lhs,
|
28 | sh_lhs_e,
|
29 | sh_lhs_t,
|
30 | BracedVarSub,
|
31 | )
|
32 | from _devbuild.gen.option_asdl import option_i
|
33 | from _devbuild.gen.types_asdl import bool_arg_type_e
|
34 | from _devbuild.gen.value_asdl import (
|
35 | value,
|
36 | value_e,
|
37 | value_t,
|
38 | sh_lvalue,
|
39 | sh_lvalue_e,
|
40 | sh_lvalue_t,
|
41 | LeftName,
|
42 | eggex_ops,
|
43 | regex_match,
|
44 | RegexMatch,
|
45 | )
|
46 | from core import alloc
|
47 | from core import error
|
48 | from core.error import e_die, e_die_status, e_strict, e_usage
|
49 | from core import num
|
50 | from core import state
|
51 | from core import ui
|
52 | from frontend import consts
|
53 | from frontend import lexer
|
54 | from frontend import match
|
55 | from frontend import parse_lib
|
56 | from frontend import reader
|
57 | from mycpp import mops
|
58 | from mycpp import mylib
|
59 | from mycpp.mylib import log, tagswitch, switch, str_cmp
|
60 | from osh import bool_stat
|
61 | from osh import word_eval
|
62 |
|
63 | import libc # for fnmatch
|
64 | # Import these names directly because the C++ translation uses macros literally.
|
65 | from libc import FNM_CASEFOLD, REG_ICASE
|
66 |
|
67 | from typing import Tuple, Optional, cast, TYPE_CHECKING
|
68 | if TYPE_CHECKING:
|
69 | from core.ui import ErrorFormatter
|
70 | from core import optview
|
71 |
|
72 | _ = log
|
73 |
|
74 | #
|
75 | # Arith and Command/Word variants of assignment
|
76 | #
|
77 | # Calls EvalShellLhs()
|
78 | # a[$key]=$val # osh/cmd_eval.py:814 (command_e.ShAssignment)
|
79 | # Calls EvalArithLhs()
|
80 | # (( a[key] = val )) # osh/sh_expr_eval.py:326 (_EvalLhsArith)
|
81 | #
|
82 | # Calls OldValue()
|
83 | # a[$key]+=$val # osh/cmd_eval.py:795 (assign_op_e.PlusEqual)
|
84 | # (( a[key] += val )) # osh/sh_expr_eval.py:308 (_EvalLhsAndLookupArith)
|
85 | #
|
86 | # RHS Indexing
|
87 | # val=${a[$key]} # osh/word_eval.py:639 (bracket_op_e.ArrayIndex)
|
88 | # (( val = a[key] )) # osh/sh_expr_eval.py:509 (Id.Arith_LBracket)
|
89 | #
|
90 |
|
91 |
|
92 | def OldValue(lval, mem, exec_opts):
|
93 | # type: (sh_lvalue_t, state.Mem, Optional[optview.Exec]) -> value_t
|
94 | """Look up for augmented assignment.
|
95 |
|
96 | For s+=val and (( i += 1 ))
|
97 |
|
98 | Args:
|
99 | lval: value we need to
|
100 | exec_opts: can be None if we don't want to check set -u!
|
101 | Because s+=val doesn't check it.
|
102 |
|
103 | TODO: A stricter and less ambiguous version for YSH.
|
104 | - Problem: why does sh_lvalue have Indexed and Keyed, while sh_lhs only has
|
105 | IndexedName?
|
106 | - should I have location.LName and sh_lvalue.Indexed only?
|
107 | - and Indexed uses the index_t type?
|
108 | - well that might be Str or Int
|
109 | """
|
110 | assert isinstance(lval, sh_lvalue_t), lval
|
111 |
|
112 | # TODO: refactor sh_lvalue_t to make this simpler
|
113 | UP_lval = lval
|
114 | with tagswitch(lval) as case:
|
115 | if case(sh_lvalue_e.Var): # (( i++ ))
|
116 | lval = cast(LeftName, UP_lval)
|
117 | var_name = lval.name
|
118 | elif case(sh_lvalue_e.Indexed): # (( a[i]++ ))
|
119 | lval = cast(sh_lvalue.Indexed, UP_lval)
|
120 | var_name = lval.name
|
121 | elif case(sh_lvalue_e.Keyed): # (( A['K']++ )) ? I think this works
|
122 | lval = cast(sh_lvalue.Keyed, UP_lval)
|
123 | var_name = lval.name
|
124 | else:
|
125 | raise AssertionError()
|
126 |
|
127 | val = mem.GetValue(var_name)
|
128 | if exec_opts and exec_opts.nounset() and val.tag() == value_e.Undef:
|
129 | e_die('Undefined variable %r' % var_name) # TODO: location info
|
130 |
|
131 | UP_val = val
|
132 | with tagswitch(lval) as case:
|
133 | if case(sh_lvalue_e.Var):
|
134 | return val
|
135 |
|
136 | elif case(sh_lvalue_e.Indexed):
|
137 | lval = cast(sh_lvalue.Indexed, UP_lval)
|
138 |
|
139 | array_val = None # type: value.BashArray
|
140 | with tagswitch(val) as case2:
|
141 | if case2(value_e.Undef):
|
142 | array_val = value.BashArray([])
|
143 | elif case2(value_e.BashArray):
|
144 | tmp = cast(value.BashArray, UP_val)
|
145 | # mycpp rewrite: add tmp. cast() creates a new var in inner scope
|
146 | array_val = tmp
|
147 | else:
|
148 | e_die("Can't use [] on value of type %s" % ui.ValType(val))
|
149 |
|
150 | s = word_eval.GetArrayItem(array_val.strs, lval.index)
|
151 |
|
152 | if s is None:
|
153 | val = value.Str('') # NOTE: Other logic is value.Undef? 0?
|
154 | else:
|
155 | assert isinstance(s, str), s
|
156 | val = value.Str(s)
|
157 |
|
158 | elif case(sh_lvalue_e.Keyed):
|
159 | lval = cast(sh_lvalue.Keyed, UP_lval)
|
160 |
|
161 | assoc_val = None # type: value.BashAssoc
|
162 | with tagswitch(val) as case2:
|
163 | if case2(value_e.Undef):
|
164 | # This never happens, because undef[x]+= is assumed to
|
165 | raise AssertionError()
|
166 | elif case2(value_e.BashAssoc):
|
167 | tmp2 = cast(value.BashAssoc, UP_val)
|
168 | # mycpp rewrite: add tmp. cast() creates a new var in inner scope
|
169 | assoc_val = tmp2
|
170 | else:
|
171 | e_die("Can't use [] on value of type %s" % ui.ValType(val))
|
172 |
|
173 | s = assoc_val.d.get(lval.key)
|
174 | if s is None:
|
175 | val = value.Str('')
|
176 | else:
|
177 | val = value.Str(s)
|
178 |
|
179 | else:
|
180 | raise AssertionError()
|
181 |
|
182 | return val
|
183 |
|
184 |
|
185 | # TODO: Should refactor for int/char-based processing
|
186 | if mylib.PYTHON:
|
187 |
|
188 | def IsLower(ch):
|
189 | # type: (str) -> bool
|
190 | return 'a' <= ch and ch <= 'z'
|
191 |
|
192 | def IsUpper(ch):
|
193 | # type: (str) -> bool
|
194 | return 'A' <= ch and ch <= 'Z'
|
195 |
|
196 |
|
197 | class UnsafeArith(object):
|
198 | """For parsing a[i] at RUNTIME."""
|
199 |
|
200 | def __init__(
|
201 | self,
|
202 | mem, # type: state.Mem
|
203 | exec_opts, # type: optview.Exec
|
204 | mutable_opts, # type: state.MutableOpts
|
205 | parse_ctx, # type: parse_lib.ParseContext
|
206 | arith_ev, # type: ArithEvaluator
|
207 | errfmt, # type: ui.ErrorFormatter
|
208 | ):
|
209 | # type: (...) -> None
|
210 | self.mem = mem
|
211 | self.exec_opts = exec_opts
|
212 | self.mutable_opts = mutable_opts
|
213 | self.parse_ctx = parse_ctx
|
214 | self.arith_ev = arith_ev
|
215 | self.errfmt = errfmt
|
216 |
|
217 | self.arena = self.parse_ctx.arena
|
218 |
|
219 | def ParseLValue(self, s, location):
|
220 | # type: (str, loc_t) -> sh_lvalue_t
|
221 | """Parse sh_lvalue for 'unset' and 'printf -v'.
|
222 |
|
223 | It uses the arith parser, so it behaves like the LHS of (( a[i] = x ))
|
224 | """
|
225 | if not self.parse_ctx.parse_opts.parse_sh_arith():
|
226 | # Do something simpler for YSH
|
227 | if not match.IsValidVarName(s):
|
228 | e_die('Invalid variable name %r (parse_sh_arith is off)' % s,
|
229 | location)
|
230 | return LeftName(s, location)
|
231 |
|
232 | a_parser = self.parse_ctx.MakeArithParser(s)
|
233 |
|
234 | with alloc.ctx_SourceCode(self.arena,
|
235 | source.ArgvWord('dynamic LHS', location)):
|
236 | try:
|
237 | anode = a_parser.Parse()
|
238 | except error.Parse as e:
|
239 | self.errfmt.PrettyPrintError(e)
|
240 | # Exception for builtins 'unset' and 'printf'
|
241 | e_usage('got invalid LHS expression', location)
|
242 |
|
243 | # Note: we parse '1+2', and then it becomes a runtime error because
|
244 | # it's not a valid LHS. Could be a parse error.
|
245 |
|
246 | if self.exec_opts.eval_unsafe_arith():
|
247 | lval = self.arith_ev.EvalArithLhs(anode)
|
248 | else:
|
249 | # Prevent attacks like these by default:
|
250 | #
|
251 | # unset -v 'A["$(echo K; rm *)"]'
|
252 | with state.ctx_Option(self.mutable_opts,
|
253 | [option_i._allow_command_sub], False):
|
254 | lval = self.arith_ev.EvalArithLhs(anode)
|
255 |
|
256 | return lval
|
257 |
|
258 | def ParseVarRef(self, ref_str, blame_tok):
|
259 | # type: (str, Token) -> BracedVarSub
|
260 | """Parse and evaluate value for ${!ref}
|
261 |
|
262 | This supports:
|
263 | - 0 to 9 for $0 to $9
|
264 | - @ for "$@" etc.
|
265 |
|
266 | See grammar in osh/word_parse.py, which is related to grammar in
|
267 | osh/word_parse.py _ReadBracedVarSub
|
268 |
|
269 | Note: declare -n allows 'varname' and 'varname[i]' and 'varname[@]', but it
|
270 | does NOT allow 0 to 9, @, *
|
271 |
|
272 | NamerefExpr = NAME Subscript? # this allows @ and * too
|
273 |
|
274 | _ResolveNameOrRef currently gives you a 'cell'. So it might not support
|
275 | sh_lvalue.Indexed?
|
276 | """
|
277 | line_reader = reader.StringLineReader(ref_str, self.arena)
|
278 | lexer = self.parse_ctx.MakeLexer(line_reader)
|
279 | w_parser = self.parse_ctx.MakeWordParser(lexer, line_reader)
|
280 |
|
281 | src = source.VarRef(blame_tok)
|
282 | with alloc.ctx_SourceCode(self.arena, src):
|
283 | try:
|
284 | bvs_part = w_parser.ParseVarRef()
|
285 | except error.Parse as e:
|
286 | # This prints the inner location
|
287 | self.errfmt.PrettyPrintError(e)
|
288 |
|
289 | # this affects builtins 'unset' and 'printf'
|
290 | e_die("Invalid var ref expression", blame_tok)
|
291 |
|
292 | return bvs_part
|
293 |
|
294 |
|
295 | class ArithEvaluator(object):
|
296 | """Shared between arith and bool evaluators.
|
297 |
|
298 | They both:
|
299 |
|
300 | 1. Convert strings to integers, respecting shopt -s strict_arith.
|
301 | 2. Look up variables and evaluate words.
|
302 | """
|
303 |
|
304 | def __init__(
|
305 | self,
|
306 | mem, # type: state.Mem
|
307 | exec_opts, # type: optview.Exec
|
308 | mutable_opts, # type: state.MutableOpts
|
309 | parse_ctx, # type: Optional[parse_lib.ParseContext]
|
310 | errfmt, # type: ErrorFormatter
|
311 | ):
|
312 | # type: (...) -> None
|
313 | self.word_ev = None # type: word_eval.StringWordEvaluator
|
314 | self.mem = mem
|
315 | self.exec_opts = exec_opts
|
316 | self.mutable_opts = mutable_opts
|
317 | self.parse_ctx = parse_ctx
|
318 | self.errfmt = errfmt
|
319 |
|
320 | def CheckCircularDeps(self):
|
321 | # type: () -> None
|
322 | assert self.word_ev is not None
|
323 |
|
324 | def _StringToBigInt(self, s, blame_loc):
|
325 | # type: (str, loc_t) -> mops.BigInt
|
326 | """Use bash-like rules to coerce a string to an integer.
|
327 |
|
328 | Runtime parsing enables silly stuff like $(( $(echo 1)$(echo 2) + 1 )) => 13
|
329 |
|
330 | 0xAB -- hex constant
|
331 | 042 -- octal constant
|
332 | 42 -- decimal constant
|
333 | 64#z -- arbitrary base constant
|
334 |
|
335 | bare word: variable
|
336 | quoted word: string (not done?)
|
337 | """
|
338 | if s.startswith('0x'):
|
339 | try:
|
340 | integer = mops.FromStr(s, 16)
|
341 | except ValueError:
|
342 | e_strict('Invalid hex constant %r' % s, blame_loc)
|
343 | # TODO: don't truncate
|
344 | return integer
|
345 |
|
346 | if s.startswith('0'):
|
347 | try:
|
348 | integer = mops.FromStr(s, 8)
|
349 | except ValueError:
|
350 | e_strict('Invalid octal constant %r' % s, blame_loc)
|
351 | return integer
|
352 |
|
353 | b, digits = mylib.split_once(s, '#') # see if it has #
|
354 | if digits is not None:
|
355 | try:
|
356 | base = int(b) # machine integer, not BigInt
|
357 | except ValueError:
|
358 | e_strict('Invalid base for numeric constant %r' % b, blame_loc)
|
359 |
|
360 | integer = mops.ZERO
|
361 | for ch in digits:
|
362 | if IsLower(ch):
|
363 | digit = ord(ch) - ord('a') + 10
|
364 | elif IsUpper(ch):
|
365 | digit = ord(ch) - ord('A') + 36
|
366 | elif ch == '@': # horrible syntax
|
367 | digit = 62
|
368 | elif ch == '_':
|
369 | digit = 63
|
370 | elif ch.isdigit():
|
371 | digit = int(ch)
|
372 | else:
|
373 | e_strict('Invalid digits for numeric constant %r' % digits,
|
374 | blame_loc)
|
375 |
|
376 | if digit >= base:
|
377 | e_strict(
|
378 | 'Digits %r out of range for base %d' % (digits, base),
|
379 | blame_loc)
|
380 |
|
381 | #integer = integer * base + digit
|
382 | integer = mops.Add(mops.Mul(integer, mops.BigInt(base)),
|
383 | mops.BigInt(digit))
|
384 | return integer
|
385 |
|
386 | try:
|
387 | # Normal base 10 integer. This includes negative numbers like '-42'.
|
388 | integer = mops.FromStr(s)
|
389 | except ValueError:
|
390 | # doesn't look like an integer
|
391 |
|
392 | # note: 'test' and '[' never evaluate recursively
|
393 | if self.parse_ctx:
|
394 | arena = self.parse_ctx.arena
|
395 |
|
396 | # Special case so we don't get EOF error
|
397 | if len(s.strip()) == 0:
|
398 | return mops.ZERO
|
399 |
|
400 | # For compatibility: Try to parse it as an expression and evaluate it.
|
401 | a_parser = self.parse_ctx.MakeArithParser(s)
|
402 |
|
403 | # TODO: Fill in the variable name
|
404 | with alloc.ctx_SourceCode(arena,
|
405 | source.Variable(None, blame_loc)):
|
406 | try:
|
407 | node2 = a_parser.Parse() # may raise error.Parse
|
408 | except error.Parse as e:
|
409 | self.errfmt.PrettyPrintError(e)
|
410 | e_die('Parse error in recursive arithmetic',
|
411 | e.location)
|
412 |
|
413 | # Prevent infinite recursion of $(( 1x )) -- it's a word that evaluates
|
414 | # to itself, and you don't want to reparse it as a word.
|
415 | if node2.tag() == arith_expr_e.Word:
|
416 | e_die("Invalid integer constant %r" % s, blame_loc)
|
417 |
|
418 | if self.exec_opts.eval_unsafe_arith():
|
419 | integer = self.EvalToBigInt(node2)
|
420 | else:
|
421 | # BoolEvaluator doesn't have parse_ctx or mutable_opts
|
422 | assert self.mutable_opts is not None
|
423 |
|
424 | # We don't need to flip _allow_process_sub, because they can't be
|
425 | # parsed. See spec/bugs.test.sh.
|
426 | with state.ctx_Option(self.mutable_opts,
|
427 | [option_i._allow_command_sub],
|
428 | False):
|
429 | integer = self.EvalToBigInt(node2)
|
430 |
|
431 | else:
|
432 | if len(s.strip()) == 0 or match.IsValidVarName(s):
|
433 | # x42 could evaluate to 0
|
434 | e_strict("Invalid integer constant %r" % s, blame_loc)
|
435 | else:
|
436 | # 42x is always fatal!
|
437 | e_die("Invalid integer constant %r" % s, blame_loc)
|
438 |
|
439 | return integer
|
440 |
|
441 | def _ValToIntOrError(self, val, blame):
|
442 | # type: (value_t, arith_expr_t) -> mops.BigInt
|
443 | try:
|
444 | UP_val = val
|
445 | with tagswitch(val) as case:
|
446 | if case(value_e.Undef):
|
447 | # 'nounset' already handled before got here
|
448 | # Happens upon a[undefined]=42, which unfortunately turns into a[0]=42.
|
449 | e_strict('Undefined value in arithmetic context',
|
450 | loc.Arith(blame))
|
451 |
|
452 | elif case(value_e.Int):
|
453 | val = cast(value.Int, UP_val)
|
454 | return val.i
|
455 |
|
456 | elif case(value_e.Str):
|
457 | val = cast(value.Str, UP_val)
|
458 | # calls e_strict
|
459 | return self._StringToBigInt(val.s, loc.Arith(blame))
|
460 |
|
461 | except error.Strict as e:
|
462 | if self.exec_opts.strict_arith():
|
463 | raise
|
464 | else:
|
465 | return mops.ZERO
|
466 |
|
467 | # Arrays and associative arrays always fail -- not controlled by
|
468 | # strict_arith.
|
469 | # In bash, (( a )) is like (( a[0] )), but I don't want that.
|
470 | # And returning '0' gives different results.
|
471 | e_die(
|
472 | "Expected a value convertible to integer, got %s" %
|
473 | ui.ValType(val), loc.Arith(blame))
|
474 |
|
475 | def _EvalLhsAndLookupArith(self, node):
|
476 | # type: (arith_expr_t) -> Tuple[mops.BigInt, sh_lvalue_t]
|
477 | """ For x = y and x += y and ++x """
|
478 |
|
479 | lval = self.EvalArithLhs(node)
|
480 | val = OldValue(lval, self.mem, self.exec_opts)
|
481 |
|
482 | # BASH_LINENO, arr (array name without strict_array), etc.
|
483 | if (val.tag() in (value_e.BashArray, value_e.BashAssoc) and
|
484 | lval.tag() == sh_lvalue_e.Var):
|
485 | named_lval = cast(LeftName, lval)
|
486 | if word_eval.ShouldArrayDecay(named_lval.name, self.exec_opts):
|
487 | if val.tag() == value_e.BashArray:
|
488 | lval = sh_lvalue.Indexed(named_lval.name, 0, loc.Missing)
|
489 | elif val.tag() == value_e.BashAssoc:
|
490 | lval = sh_lvalue.Keyed(named_lval.name, '0', loc.Missing)
|
491 | val = word_eval.DecayArray(val)
|
492 |
|
493 | # This error message could be better, but we already have one
|
494 | #if val.tag() == value_e.BashArray:
|
495 | # e_die("Can't use assignment like ++ or += on arrays")
|
496 |
|
497 | i = self._ValToIntOrError(val, node)
|
498 | return i, lval
|
499 |
|
500 | def _Store(self, lval, new_int):
|
501 | # type: (sh_lvalue_t, mops.BigInt) -> None
|
502 | val = value.Str(mops.ToStr(new_int))
|
503 | state.OshLanguageSetValue(self.mem, lval, val)
|
504 |
|
505 | def EvalToBigInt(self, node):
|
506 | # type: (arith_expr_t) -> mops.BigInt
|
507 | """Used externally by ${a[i+1]} and ${a:start:len}.
|
508 |
|
509 | Also used internally.
|
510 | """
|
511 | val = self.Eval(node)
|
512 |
|
513 | # BASH_LINENO, arr (array name without strict_array), etc.
|
514 | if (val.tag() in (value_e.BashArray, value_e.BashAssoc) and
|
515 | node.tag() == arith_expr_e.VarSub):
|
516 | vsub = cast(Token, node)
|
517 | if word_eval.ShouldArrayDecay(lexer.LazyStr(vsub), self.exec_opts):
|
518 | val = word_eval.DecayArray(val)
|
519 |
|
520 | i = self._ValToIntOrError(val, node)
|
521 | return i
|
522 |
|
523 | def EvalToInt(self, node):
|
524 | # type: (arith_expr_t) -> int
|
525 | return mops.BigTruncate(self.EvalToBigInt(node))
|
526 |
|
527 | def Eval(self, node):
|
528 | # type: (arith_expr_t) -> value_t
|
529 | """
|
530 | Returns:
|
531 | None for Undef (e.g. empty cell) TODO: Don't return 0!
|
532 | int for Str
|
533 | List[int] for BashArray
|
534 | Dict[str, str] for BashAssoc (TODO: Should we support this?)
|
535 |
|
536 | NOTE: (( A['x'] = 'x' )) and (( x = A['x'] )) are syntactically valid in
|
537 | bash, but don't do what you'd think. 'x' sometimes a variable name and
|
538 | sometimes a key.
|
539 | """
|
540 | # OSH semantics: Variable NAMES cannot be formed dynamically; but INTEGERS
|
541 | # can. ${foo:-3}4 is OK. $? will be a compound word too, so we don't have
|
542 | # to handle that as a special case.
|
543 |
|
544 | UP_node = node
|
545 | with tagswitch(node) as case:
|
546 | if case(arith_expr_e.VarSub): # $(( x )) (can be array)
|
547 | vsub = cast(Token, UP_node)
|
548 | var_name = lexer.LazyStr(vsub)
|
549 | val = self.mem.GetValue(var_name)
|
550 | if val.tag() == value_e.Undef and self.exec_opts.nounset():
|
551 | e_die('Undefined variable %r' % var_name, vsub)
|
552 | return val
|
553 |
|
554 | elif case(arith_expr_e.Word): # $(( $x )) $(( ${x}${y} )), etc.
|
555 | w = cast(CompoundWord, UP_node)
|
556 | return self.word_ev.EvalWordToString(w)
|
557 |
|
558 | elif case(arith_expr_e.UnaryAssign): # a++
|
559 | node = cast(arith_expr.UnaryAssign, UP_node)
|
560 |
|
561 | op_id = node.op_id
|
562 | old_big, lval = self._EvalLhsAndLookupArith(node.child)
|
563 |
|
564 | if op_id == Id.Node_PostDPlus: # post-increment
|
565 | new_big = mops.Add(old_big, mops.ONE)
|
566 | result = old_big
|
567 |
|
568 | elif op_id == Id.Node_PostDMinus: # post-decrement
|
569 | new_big = mops.Sub(old_big, mops.ONE)
|
570 | result = old_big
|
571 |
|
572 | elif op_id == Id.Arith_DPlus: # pre-increment
|
573 | new_big = mops.Add(old_big, mops.ONE)
|
574 | result = new_big
|
575 |
|
576 | elif op_id == Id.Arith_DMinus: # pre-decrement
|
577 | new_big = mops.Sub(old_big, mops.ONE)
|
578 | result = new_big
|
579 |
|
580 | else:
|
581 | raise AssertionError(op_id)
|
582 |
|
583 | self._Store(lval, new_big)
|
584 | return value.Int(result)
|
585 |
|
586 | elif case(arith_expr_e.BinaryAssign): # a=1, a+=5, a[1]+=5
|
587 | node = cast(arith_expr.BinaryAssign, UP_node)
|
588 | op_id = node.op_id
|
589 |
|
590 | if op_id == Id.Arith_Equal:
|
591 | # Don't really need a span ID here, because tdop.CheckLhsExpr should
|
592 | # have done all the validation.
|
593 | lval = self.EvalArithLhs(node.left)
|
594 | rhs_big = self.EvalToBigInt(node.right)
|
595 |
|
596 | self._Store(lval, rhs_big)
|
597 | return value.Int(rhs_big)
|
598 |
|
599 | old_big, lval = self._EvalLhsAndLookupArith(node.left)
|
600 | rhs_big = self.EvalToBigInt(node.right)
|
601 |
|
602 | if op_id == Id.Arith_PlusEqual:
|
603 | new_big = mops.Add(old_big, rhs_big)
|
604 | elif op_id == Id.Arith_MinusEqual:
|
605 | new_big = mops.Sub(old_big, rhs_big)
|
606 | elif op_id == Id.Arith_StarEqual:
|
607 | new_big = mops.Mul(old_big, rhs_big)
|
608 |
|
609 | elif op_id == Id.Arith_SlashEqual:
|
610 | if mops.Equal(rhs_big, mops.ZERO):
|
611 | e_die('Divide by zero') # TODO: location
|
612 | new_big = num.IntDivide(old_big, rhs_big)
|
613 |
|
614 | elif op_id == Id.Arith_PercentEqual:
|
615 | if mops.Equal(rhs_big, mops.ZERO):
|
616 | e_die('Divide by zero') # TODO: location
|
617 | new_big = num.IntRemainder(old_big, rhs_big)
|
618 |
|
619 | elif op_id == Id.Arith_DGreatEqual:
|
620 | new_big = mops.RShift(old_big, rhs_big)
|
621 | elif op_id == Id.Arith_DLessEqual:
|
622 | new_big = mops.LShift(old_big, rhs_big)
|
623 | elif op_id == Id.Arith_AmpEqual:
|
624 | new_big = mops.BitAnd(old_big, rhs_big)
|
625 | elif op_id == Id.Arith_PipeEqual:
|
626 | new_big = mops.BitOr(old_big, rhs_big)
|
627 | elif op_id == Id.Arith_CaretEqual:
|
628 | new_big = mops.BitXor(old_big, rhs_big)
|
629 | else:
|
630 | raise AssertionError(op_id) # shouldn't get here
|
631 |
|
632 | self._Store(lval, new_big)
|
633 | return value.Int(new_big)
|
634 |
|
635 | elif case(arith_expr_e.Unary):
|
636 | node = cast(arith_expr.Unary, UP_node)
|
637 | op_id = node.op_id
|
638 |
|
639 | i = self.EvalToBigInt(node.child)
|
640 |
|
641 | if op_id == Id.Node_UnaryPlus: # +i
|
642 | result = i
|
643 | elif op_id == Id.Node_UnaryMinus: # -i
|
644 | result = mops.Sub(mops.ZERO, i)
|
645 |
|
646 | elif op_id == Id.Arith_Bang: # logical negation
|
647 | if mops.Equal(i, mops.ZERO):
|
648 | result = mops.ONE
|
649 | else:
|
650 | result = mops.ZERO
|
651 | elif op_id == Id.Arith_Tilde: # bitwise complement
|
652 | result = mops.BitNot(i)
|
653 | else:
|
654 | raise AssertionError(op_id) # shouldn't get here
|
655 |
|
656 | return value.Int(result)
|
657 |
|
658 | elif case(arith_expr_e.Binary):
|
659 | node = cast(arith_expr.Binary, UP_node)
|
660 | op_id = node.op_id
|
661 |
|
662 | # Short-circuit evaluation for || and &&.
|
663 | if op_id == Id.Arith_DPipe:
|
664 | lhs_big = self.EvalToBigInt(node.left)
|
665 | if mops.Equal(lhs_big, mops.ZERO):
|
666 | rhs_big = self.EvalToBigInt(node.right)
|
667 | if mops.Equal(rhs_big, mops.ZERO):
|
668 | result = mops.ZERO # false
|
669 | else:
|
670 | result = mops.ONE # true
|
671 | else:
|
672 | result = mops.ONE # true
|
673 | return value.Int(result)
|
674 |
|
675 | if op_id == Id.Arith_DAmp:
|
676 | lhs_big = self.EvalToBigInt(node.left)
|
677 | if mops.Equal(lhs_big, mops.ZERO):
|
678 | result = mops.ZERO # false
|
679 | else:
|
680 | rhs_big = self.EvalToBigInt(node.right)
|
681 | if mops.Equal(rhs_big, mops.ZERO):
|
682 | result = mops.ZERO # false
|
683 | else:
|
684 | result = mops.ONE # true
|
685 | return value.Int(result)
|
686 |
|
687 | if op_id == Id.Arith_LBracket:
|
688 | # NOTE: Similar to bracket_op_e.ArrayIndex in osh/word_eval.py
|
689 |
|
690 | left = self.Eval(node.left)
|
691 | UP_left = left
|
692 | with tagswitch(left) as case:
|
693 | if case(value_e.BashArray):
|
694 | array_val = cast(value.BashArray, UP_left)
|
695 | index = mops.BigTruncate(
|
696 | self.EvalToBigInt(node.right))
|
697 | s = word_eval.GetArrayItem(array_val.strs, index)
|
698 |
|
699 | elif case(value_e.BashAssoc):
|
700 | left = cast(value.BashAssoc, UP_left)
|
701 | key = self.EvalWordToString(node.right)
|
702 | s = left.d.get(key)
|
703 |
|
704 | else:
|
705 | # TODO: Add error context
|
706 | e_die(
|
707 | 'Expected array or assoc in index expression, got %s'
|
708 | % ui.ValType(left))
|
709 |
|
710 | if s is None:
|
711 | val = value.Undef
|
712 | else:
|
713 | val = value.Str(s)
|
714 |
|
715 | return val
|
716 |
|
717 | if op_id == Id.Arith_Comma:
|
718 | self.EvalToBigInt(node.left) # throw away result
|
719 | result = self.EvalToBigInt(node.right)
|
720 | return value.Int(result)
|
721 |
|
722 | # Rest are integers
|
723 | lhs_big = self.EvalToBigInt(node.left)
|
724 | rhs_big = self.EvalToBigInt(node.right)
|
725 |
|
726 | if op_id == Id.Arith_Plus:
|
727 | result = mops.Add(lhs_big, rhs_big)
|
728 | elif op_id == Id.Arith_Minus:
|
729 | result = mops.Sub(lhs_big, rhs_big)
|
730 | elif op_id == Id.Arith_Star:
|
731 | result = mops.Mul(lhs_big, rhs_big)
|
732 | elif op_id == Id.Arith_Slash:
|
733 | if mops.Equal(rhs_big, mops.ZERO):
|
734 | e_die('Divide by zero', loc.Arith(node.right))
|
735 | result = num.IntDivide(lhs_big, rhs_big)
|
736 |
|
737 | elif op_id == Id.Arith_Percent:
|
738 | if mops.Equal(rhs_big, mops.ZERO):
|
739 | e_die('Divide by zero', loc.Arith(node.right))
|
740 | result = num.IntRemainder(lhs_big, rhs_big)
|
741 |
|
742 | elif op_id == Id.Arith_DStar:
|
743 | if mops.Greater(mops.ZERO, rhs_big):
|
744 | e_die("Exponent can't be a negative number",
|
745 | loc.Arith(node.right))
|
746 | result = num.Exponent(lhs_big, rhs_big)
|
747 |
|
748 | elif op_id == Id.Arith_DEqual:
|
749 | result = mops.FromBool(mops.Equal(lhs_big, rhs_big))
|
750 | elif op_id == Id.Arith_NEqual:
|
751 | result = mops.FromBool(not mops.Equal(lhs_big, rhs_big))
|
752 | elif op_id == Id.Arith_Great:
|
753 | result = mops.FromBool(mops.Greater(lhs_big, rhs_big))
|
754 | elif op_id == Id.Arith_GreatEqual:
|
755 | result = mops.FromBool(
|
756 | mops.Greater(lhs_big, rhs_big) or
|
757 | mops.Equal(lhs_big, rhs_big))
|
758 | elif op_id == Id.Arith_Less:
|
759 | result = mops.FromBool(mops.Greater(rhs_big, lhs_big))
|
760 | elif op_id == Id.Arith_LessEqual:
|
761 | result = mops.FromBool(
|
762 | mops.Greater(rhs_big, lhs_big) or
|
763 | mops.Equal(lhs_big, rhs_big))
|
764 |
|
765 | elif op_id == Id.Arith_Pipe:
|
766 | result = mops.BitOr(lhs_big, rhs_big)
|
767 | elif op_id == Id.Arith_Amp:
|
768 | result = mops.BitAnd(lhs_big, rhs_big)
|
769 | elif op_id == Id.Arith_Caret:
|
770 | result = mops.BitXor(lhs_big, rhs_big)
|
771 |
|
772 | # Note: how to define shift of negative numbers?
|
773 | elif op_id == Id.Arith_DLess:
|
774 | result = mops.LShift(lhs_big, rhs_big)
|
775 | elif op_id == Id.Arith_DGreat:
|
776 | result = mops.RShift(lhs_big, rhs_big)
|
777 | else:
|
778 | raise AssertionError(op_id)
|
779 |
|
780 | return value.Int(result)
|
781 |
|
782 | elif case(arith_expr_e.TernaryOp):
|
783 | node = cast(arith_expr.TernaryOp, UP_node)
|
784 |
|
785 | cond = self.EvalToBigInt(node.cond)
|
786 | if mops.Equal(cond, mops.ZERO):
|
787 | return self.Eval(node.false_expr)
|
788 | else:
|
789 | return self.Eval(node.true_expr)
|
790 |
|
791 | else:
|
792 | raise AssertionError(node.tag())
|
793 |
|
794 | raise AssertionError('for -Wreturn-type in C++')
|
795 |
|
796 | def EvalWordToString(self, node):
|
797 | # type: (arith_expr_t) -> str
|
798 | """
|
799 | Raises:
|
800 | error.FatalRuntime if the expression isn't a string
|
801 | or if it contains a bare variable like a[x]
|
802 |
|
803 | These are allowed because they're unambiguous, unlike a[x]
|
804 |
|
805 | a[$x] a["$x"] a["x"] a['x']
|
806 | """
|
807 | UP_node = node
|
808 | if node.tag() == arith_expr_e.Word: # $(( $x )) $(( ${x}${y} )), etc.
|
809 | w = cast(CompoundWord, UP_node)
|
810 | val = self.word_ev.EvalWordToString(w)
|
811 | return val.s
|
812 | else:
|
813 | # TODO: location info for original
|
814 | e_die("Associative array keys must be strings: $x 'x' \"$x\" etc.")
|
815 |
|
816 | def EvalShellLhs(self, node, which_scopes):
|
817 | # type: (sh_lhs_t, scope_t) -> sh_lvalue_t
|
818 | """Evaluate a shell LHS expression
|
819 |
|
820 | For a=b and a[x]=b etc.
|
821 | """
|
822 | assert isinstance(node, sh_lhs_t), node
|
823 |
|
824 | UP_node = node
|
825 | lval = None # type: sh_lvalue_t
|
826 | with tagswitch(node) as case:
|
827 | if case(sh_lhs_e.Name): # a=x
|
828 | node = cast(sh_lhs.Name, UP_node)
|
829 | assert node.name is not None
|
830 |
|
831 | lval1 = LeftName(node.name, node.left)
|
832 | lval = lval1
|
833 |
|
834 | elif case(sh_lhs_e.IndexedName): # a[1+2]=x
|
835 | node = cast(sh_lhs.IndexedName, UP_node)
|
836 | assert node.name is not None
|
837 |
|
838 | if self.mem.IsBashAssoc(node.name):
|
839 | key = self.EvalWordToString(node.index)
|
840 | lval2 = sh_lvalue.Keyed(node.name, key, node.left)
|
841 | lval = lval2
|
842 | else:
|
843 | index = mops.BigTruncate(self.EvalToBigInt(node.index))
|
844 | lval3 = sh_lvalue.Indexed(node.name, index, node.left)
|
845 | lval = lval3
|
846 |
|
847 | else:
|
848 | raise AssertionError(node.tag())
|
849 |
|
850 | return lval
|
851 |
|
852 | def _VarNameOrWord(self, anode):
|
853 | # type: (arith_expr_t) -> Tuple[Optional[str], loc_t]
|
854 | """Returns a variable name if the arith node can be interpreted that
|
855 | way."""
|
856 | UP_anode = anode
|
857 | with tagswitch(anode) as case:
|
858 | if case(arith_expr_e.VarSub):
|
859 | tok = cast(Token, UP_anode)
|
860 | return (lexer.LazyStr(tok), tok)
|
861 |
|
862 | elif case(arith_expr_e.Word):
|
863 | w = cast(CompoundWord, UP_anode)
|
864 | var_name = self.EvalWordToString(w)
|
865 | return (var_name, w)
|
866 |
|
867 | no_str = None # type: str
|
868 | return (no_str, loc.Missing)
|
869 |
|
870 | def EvalArithLhs(self, anode):
|
871 | # type: (arith_expr_t) -> sh_lvalue_t
|
872 | """
|
873 | For (( a[x] = 1 )) etc.
|
874 | """
|
875 | UP_anode = anode
|
876 | if anode.tag() == arith_expr_e.Binary:
|
877 | anode = cast(arith_expr.Binary, UP_anode)
|
878 | if anode.op_id == Id.Arith_LBracket:
|
879 | var_name, location = self._VarNameOrWord(anode.left)
|
880 |
|
881 | # (( 1[2] = 3 )) isn't valid
|
882 | if not match.IsValidVarName(var_name):
|
883 | e_die('Invalid variable name %r' % var_name, location)
|
884 |
|
885 | if var_name is not None:
|
886 | if self.mem.IsBashAssoc(var_name):
|
887 | key = self.EvalWordToString(anode.right)
|
888 | return sh_lvalue.Keyed(var_name, key, location)
|
889 | else:
|
890 | index = mops.BigTruncate(self.EvalToBigInt(
|
891 | anode.right))
|
892 | return sh_lvalue.Indexed(var_name, index, location)
|
893 |
|
894 | var_name, location = self._VarNameOrWord(anode)
|
895 | if var_name is not None:
|
896 | return LeftName(var_name, location)
|
897 |
|
898 | # e.g. unset 'x-y'. status 2 for runtime parse error
|
899 | e_die_status(2, 'Invalid LHS to modify', location)
|
900 |
|
901 |
|
902 | class BoolEvaluator(ArithEvaluator):
|
903 | """This is also an ArithEvaluator because it has to understand.
|
904 |
|
905 | [[ x -eq 3 ]]
|
906 |
|
907 | where x='1+2'
|
908 | """
|
909 |
|
910 | def __init__(
|
911 | self,
|
912 | mem, # type: state.Mem
|
913 | exec_opts, # type: optview.Exec
|
914 | mutable_opts, # type: Optional[state.MutableOpts]
|
915 | parse_ctx, # type: Optional[parse_lib.ParseContext]
|
916 | errfmt, # type: ErrorFormatter
|
917 | always_strict=False # type: bool
|
918 | ):
|
919 | # type: (...) -> None
|
920 | ArithEvaluator.__init__(self, mem, exec_opts, mutable_opts, parse_ctx,
|
921 | errfmt)
|
922 | self.always_strict = always_strict
|
923 |
|
924 | def _StringToBigIntOrError(self, s, blame_word=None):
|
925 | # type: (str, Optional[word_t]) -> mops.BigInt
|
926 | """Used by both [[ $x -gt 3 ]] and (( $x ))."""
|
927 | if blame_word:
|
928 | location = loc.Word(blame_word) # type: loc_t
|
929 | else:
|
930 | location = loc.Missing
|
931 |
|
932 | try:
|
933 | i = self._StringToBigInt(s, location)
|
934 | except error.Strict as e:
|
935 | if self.always_strict or self.exec_opts.strict_arith():
|
936 | raise
|
937 | else:
|
938 | i = mops.ZERO
|
939 | return i
|
940 |
|
941 | def _EvalCompoundWord(self, word, eval_flags=0):
|
942 | # type: (word_t, int) -> str
|
943 | val = self.word_ev.EvalWordToString(word, eval_flags)
|
944 | return val.s
|
945 |
|
946 | def EvalB(self, node):
|
947 | # type: (bool_expr_t) -> bool
|
948 |
|
949 | UP_node = node
|
950 | with tagswitch(node) as case:
|
951 | if case(bool_expr_e.WordTest):
|
952 | node = cast(bool_expr.WordTest, UP_node)
|
953 | s = self._EvalCompoundWord(node.w)
|
954 | return bool(s)
|
955 |
|
956 | elif case(bool_expr_e.LogicalNot):
|
957 | node = cast(bool_expr.LogicalNot, UP_node)
|
958 | b = self.EvalB(node.child)
|
959 | return not b
|
960 |
|
961 | elif case(bool_expr_e.LogicalAnd):
|
962 | node = cast(bool_expr.LogicalAnd, UP_node)
|
963 | # Short-circuit evaluation
|
964 | if self.EvalB(node.left):
|
965 | return self.EvalB(node.right)
|
966 | else:
|
967 | return False
|
968 |
|
969 | elif case(bool_expr_e.LogicalOr):
|
970 | node = cast(bool_expr.LogicalOr, UP_node)
|
971 | if self.EvalB(node.left):
|
972 | return True
|
973 | else:
|
974 | return self.EvalB(node.right)
|
975 |
|
976 | elif case(bool_expr_e.Unary):
|
977 | node = cast(bool_expr.Unary, UP_node)
|
978 | op_id = node.op_id
|
979 | s = self._EvalCompoundWord(node.child)
|
980 |
|
981 | # Now dispatch on arg type
|
982 | arg_type = consts.BoolArgType(
|
983 | op_id) # could be static in the LST?
|
984 |
|
985 | if arg_type == bool_arg_type_e.Path:
|
986 | return bool_stat.DoUnaryOp(op_id, s)
|
987 |
|
988 | if arg_type == bool_arg_type_e.Str:
|
989 | if op_id == Id.BoolUnary_z:
|
990 | return not bool(s)
|
991 | if op_id == Id.BoolUnary_n:
|
992 | return bool(s)
|
993 |
|
994 | raise AssertionError(op_id) # should never happen
|
995 |
|
996 | if arg_type == bool_arg_type_e.Other:
|
997 | if op_id == Id.BoolUnary_t:
|
998 | return bool_stat.isatty(s, node.child)
|
999 |
|
1000 | # See whether 'set -o' options have been set
|
1001 | if op_id == Id.BoolUnary_o:
|
1002 | index = consts.OptionNum(s)
|
1003 | if index == 0:
|
1004 | return False
|
1005 | else:
|
1006 | return self.exec_opts.opt0_array[index]
|
1007 |
|
1008 | if op_id == Id.BoolUnary_v:
|
1009 | val = self.mem.GetValue(s)
|
1010 | return val.tag() != value_e.Undef
|
1011 |
|
1012 | e_die("%s isn't implemented" %
|
1013 | ui.PrettyId(op_id)) # implicit location
|
1014 |
|
1015 | raise AssertionError(arg_type)
|
1016 |
|
1017 | elif case(bool_expr_e.Binary):
|
1018 | node = cast(bool_expr.Binary, UP_node)
|
1019 |
|
1020 | op_id = node.op_id
|
1021 | # Whether to glob escape
|
1022 | eval_flags = 0
|
1023 | with switch(op_id) as case2:
|
1024 | if case2(Id.BoolBinary_GlobEqual, Id.BoolBinary_GlobDEqual,
|
1025 | Id.BoolBinary_GlobNEqual):
|
1026 | eval_flags |= word_eval.QUOTE_FNMATCH
|
1027 | elif case2(Id.BoolBinary_EqualTilde):
|
1028 | eval_flags |= word_eval.QUOTE_ERE
|
1029 |
|
1030 | s1 = self._EvalCompoundWord(node.left)
|
1031 | s2 = self._EvalCompoundWord(node.right, eval_flags)
|
1032 |
|
1033 | # Now dispatch on arg type
|
1034 | arg_type = consts.BoolArgType(op_id)
|
1035 |
|
1036 | if arg_type == bool_arg_type_e.Path:
|
1037 | return bool_stat.DoBinaryOp(op_id, s1, s2)
|
1038 |
|
1039 | if arg_type == bool_arg_type_e.Int:
|
1040 | # NOTE: We assume they are constants like [[ 3 -eq 3 ]].
|
1041 | # Bash also allows [[ 1+2 -eq 3 ]].
|
1042 | i1 = self._StringToBigIntOrError(s1, blame_word=node.left)
|
1043 | i2 = self._StringToBigIntOrError(s2, blame_word=node.right)
|
1044 |
|
1045 | if op_id == Id.BoolBinary_eq:
|
1046 | return mops.Equal(i1, i2)
|
1047 | if op_id == Id.BoolBinary_ne:
|
1048 | return not mops.Equal(i1, i2)
|
1049 | if op_id == Id.BoolBinary_gt:
|
1050 | return mops.Greater(i1, i2)
|
1051 | if op_id == Id.BoolBinary_ge:
|
1052 | return mops.Greater(i1, i2) or mops.Equal(i1, i2)
|
1053 | if op_id == Id.BoolBinary_lt:
|
1054 | return mops.Greater(i2, i1)
|
1055 | if op_id == Id.BoolBinary_le:
|
1056 | return mops.Greater(i2, i1) or mops.Equal(i1, i2)
|
1057 |
|
1058 | raise AssertionError(op_id) # should never happen
|
1059 |
|
1060 | if arg_type == bool_arg_type_e.Str:
|
1061 | fnmatch_flags = (FNM_CASEFOLD
|
1062 | if self.exec_opts.nocasematch() else 0)
|
1063 |
|
1064 | if op_id in (Id.BoolBinary_GlobEqual,
|
1065 | Id.BoolBinary_GlobDEqual):
|
1066 | #log('Matching %s against pattern %s', s1, s2)
|
1067 | return libc.fnmatch(s2, s1, fnmatch_flags)
|
1068 |
|
1069 | if op_id == Id.BoolBinary_GlobNEqual:
|
1070 | return not libc.fnmatch(s2, s1, fnmatch_flags)
|
1071 |
|
1072 | if op_id in (Id.BoolBinary_Equal, Id.BoolBinary_DEqual):
|
1073 | return s1 == s2
|
1074 |
|
1075 | if op_id == Id.BoolBinary_NEqual:
|
1076 | return s1 != s2
|
1077 |
|
1078 | if op_id == Id.BoolBinary_EqualTilde:
|
1079 | # TODO: This should go to --debug-file
|
1080 | #log('Matching %r against regex %r', s1, s2)
|
1081 | regex_flags = (REG_ICASE
|
1082 | if self.exec_opts.nocasematch() else 0)
|
1083 |
|
1084 | try:
|
1085 | indices = libc.regex_search(s2, regex_flags, s1, 0)
|
1086 | except ValueError as e:
|
1087 | # Status 2 indicates a regex parse error. This is fatal in OSH but
|
1088 | # not in bash, which treats [[ like a command with an exit code.
|
1089 | e_die_status(2, e.message, loc.Word(node.right))
|
1090 |
|
1091 | if indices is not None:
|
1092 | self.mem.SetRegexMatch(
|
1093 | RegexMatch(s1, indices, eggex_ops.No))
|
1094 | return True
|
1095 | else:
|
1096 | self.mem.SetRegexMatch(regex_match.No)
|
1097 | return False
|
1098 |
|
1099 | if op_id == Id.Op_Less:
|
1100 | return str_cmp(s1, s2) < 0
|
1101 |
|
1102 | if op_id == Id.Op_Great:
|
1103 | return str_cmp(s1, s2) > 0
|
1104 |
|
1105 | raise AssertionError(op_id) # should never happen
|
1106 |
|
1107 | raise AssertionError(node.tag())
|