OILS / osh / sh_expr_eval.py View on Github | oilshell.org

1147 lines, 743 significant
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"""
9sh_expr_eval.py -- Shell boolean and arithmetic expressions.
10"""
11from __future__ import print_function
12
13from _devbuild.gen.id_kind_asdl import Id
14from _devbuild.gen.runtime_asdl import scope_t
15from _devbuild.gen.syntax_asdl import (
16 word_t,
17 CompoundWord,
18 Token,
19 loc,
20 loc_t,
21 source,
22 arith_expr,
23 arith_expr_e,
24 arith_expr_t,
25 bool_expr,
26 bool_expr_e,
27 bool_expr_t,
28 sh_lhs,
29 sh_lhs_e,
30 sh_lhs_t,
31 BracedVarSub,
32)
33from _devbuild.gen.option_asdl import option_i
34from _devbuild.gen.types_asdl import bool_arg_type_e
35from _devbuild.gen.value_asdl import (
36 value,
37 value_e,
38 value_t,
39 sh_lvalue,
40 sh_lvalue_e,
41 sh_lvalue_t,
42 LeftName,
43 eggex_ops,
44 regex_match,
45 RegexMatch,
46)
47from core import alloc
48from core import error
49from core.error import e_die, e_die_status, e_strict, e_usage
50from core import num
51from core import state
52from core import ui
53from frontend import consts
54from frontend import lexer
55from frontend import location
56from frontend import match
57from frontend import parse_lib
58from frontend import reader
59from mycpp import mops
60from mycpp import mylib
61from mycpp.mylib import log, tagswitch, switch, str_cmp
62from osh import bool_stat
63from osh import word_eval
64
65import libc # for fnmatch
66# Import these names directly because the C++ translation uses macros literally.
67from libc import FNM_CASEFOLD, REG_ICASE
68
69from typing import Tuple, Optional, cast, TYPE_CHECKING
70if TYPE_CHECKING:
71 from core.ui import ErrorFormatter
72 from core import optview
73
74_ = log
75
76#
77# Arith and Command/Word variants of assignment
78#
79# Calls EvalShellLhs()
80# a[$key]=$val # osh/cmd_eval.py:814 (command_e.ShAssignment)
81# Calls EvalArithLhs()
82# (( a[key] = val )) # osh/sh_expr_eval.py:326 (_EvalLhsArith)
83#
84# Calls OldValue()
85# a[$key]+=$val # osh/cmd_eval.py:795 (assign_op_e.PlusEqual)
86# (( a[key] += val )) # osh/sh_expr_eval.py:308 (_EvalLhsAndLookupArith)
87#
88# RHS Indexing
89# val=${a[$key]} # osh/word_eval.py:639 (bracket_op_e.ArrayIndex)
90# (( val = a[key] )) # osh/sh_expr_eval.py:509 (Id.Arith_LBracket)
91#
92
93
94def OldValue(lval, mem, exec_opts):
95 # type: (sh_lvalue_t, state.Mem, Optional[optview.Exec]) -> value_t
96 """Look up for augmented assignment.
97
98 For s+=val and (( i += 1 ))
99
100 Args:
101 lval: value we need to
102 exec_opts: can be None if we don't want to check set -u!
103 Because s+=val doesn't check it.
104
105 TODO: A stricter and less ambiguous version for YSH.
106 - Problem: why does sh_lvalue have Indexed and Keyed, while sh_lhs only has
107 IndexedName?
108 - should I have location.LName and sh_lvalue.Indexed only?
109 - and Indexed uses the index_t type?
110 - well that might be Str or Int
111 """
112 assert isinstance(lval, sh_lvalue_t), lval
113
114 # TODO: refactor sh_lvalue_t to make this simpler
115 UP_lval = lval
116 with tagswitch(lval) as case:
117 if case(sh_lvalue_e.Var): # (( i++ ))
118 lval = cast(LeftName, UP_lval)
119 var_name = lval.name
120 elif case(sh_lvalue_e.Indexed): # (( a[i]++ ))
121 lval = cast(sh_lvalue.Indexed, UP_lval)
122 var_name = lval.name
123 elif case(sh_lvalue_e.Keyed): # (( A['K']++ )) ? I think this works
124 lval = cast(sh_lvalue.Keyed, UP_lval)
125 var_name = lval.name
126 else:
127 raise AssertionError()
128
129 val = mem.GetValue(var_name)
130 if exec_opts and exec_opts.nounset() and val.tag() == value_e.Undef:
131 e_die('Undefined variable %r' % var_name) # TODO: location info
132
133 UP_val = val
134 with tagswitch(lval) as case:
135 if case(sh_lvalue_e.Var):
136 return val
137
138 elif case(sh_lvalue_e.Indexed):
139 lval = cast(sh_lvalue.Indexed, UP_lval)
140
141 array_val = None # type: value.BashArray
142 with tagswitch(val) as case2:
143 if case2(value_e.Undef):
144 array_val = value.BashArray([])
145 elif case2(value_e.BashArray):
146 tmp = cast(value.BashArray, UP_val)
147 # mycpp rewrite: add tmp. cast() creates a new var in inner scope
148 array_val = tmp
149 else:
150 e_die("Can't use [] on value of type %s" % ui.ValType(val))
151
152 s = word_eval.GetArrayItem(array_val.strs, lval.index)
153
154 if s is None:
155 val = value.Str('') # NOTE: Other logic is value.Undef? 0?
156 else:
157 assert isinstance(s, str), s
158 val = value.Str(s)
159
160 elif case(sh_lvalue_e.Keyed):
161 lval = cast(sh_lvalue.Keyed, UP_lval)
162
163 assoc_val = None # type: value.BashAssoc
164 with tagswitch(val) as case2:
165 if case2(value_e.Undef):
166 # This never happens, because undef[x]+= is assumed to
167 raise AssertionError()
168 elif case2(value_e.BashAssoc):
169 tmp2 = cast(value.BashAssoc, UP_val)
170 # mycpp rewrite: add tmp. cast() creates a new var in inner scope
171 assoc_val = tmp2
172 else:
173 e_die("Can't use [] on value of type %s" % ui.ValType(val))
174
175 s = assoc_val.d.get(lval.key)
176 if s is None:
177 val = value.Str('')
178 else:
179 val = value.Str(s)
180
181 else:
182 raise AssertionError()
183
184 return val
185
186
187# TODO: Should refactor for int/char-based processing
188if mylib.PYTHON:
189
190 def IsLower(ch):
191 # type: (str) -> bool
192 return 'a' <= ch and ch <= 'z'
193
194 def IsUpper(ch):
195 # type: (str) -> bool
196 return 'A' <= ch and ch <= 'Z'
197
198
199class UnsafeArith(object):
200 """For parsing a[i] at RUNTIME."""
201
202 def __init__(
203 self,
204 mem, # type: state.Mem
205 exec_opts, # type: optview.Exec
206 mutable_opts, # type: state.MutableOpts
207 parse_ctx, # type: parse_lib.ParseContext
208 arith_ev, # type: ArithEvaluator
209 errfmt, # type: ui.ErrorFormatter
210 ):
211 # type: (...) -> None
212 self.mem = mem
213 self.exec_opts = exec_opts
214 self.mutable_opts = mutable_opts
215 self.parse_ctx = parse_ctx
216 self.arith_ev = arith_ev
217 self.errfmt = errfmt
218
219 self.arena = self.parse_ctx.arena
220
221 def ParseLValue(self, s, location):
222 # type: (str, loc_t) -> sh_lvalue_t
223 """Parse sh_lvalue for 'unset' and 'printf -v'.
224
225 It uses the arith parser, so it behaves like the LHS of (( a[i] = x ))
226 """
227 if not self.parse_ctx.parse_opts.parse_sh_arith():
228 # Do something simpler for YSH
229 if not match.IsValidVarName(s):
230 e_die('Invalid variable name %r (parse_sh_arith is off)' % s,
231 location)
232 return LeftName(s, location)
233
234 a_parser = self.parse_ctx.MakeArithParser(s)
235
236 with alloc.ctx_SourceCode(self.arena,
237 source.ArgvWord('dynamic LHS', location)):
238 try:
239 anode = a_parser.Parse()
240 except error.Parse as e:
241 self.errfmt.PrettyPrintError(e)
242 # Exception for builtins 'unset' and 'printf'
243 e_usage('got invalid LHS expression', location)
244
245 # Note: we parse '1+2', and then it becomes a runtime error because
246 # it's not a valid LHS. Could be a parse error.
247
248 if self.exec_opts.eval_unsafe_arith():
249 lval = self.arith_ev.EvalArithLhs(anode)
250 else:
251 # Prevent attacks like these by default:
252 #
253 # unset -v 'A["$(echo K; rm *)"]'
254 with state.ctx_Option(self.mutable_opts,
255 [option_i._allow_command_sub], False):
256 lval = self.arith_ev.EvalArithLhs(anode)
257
258 return lval
259
260 def ParseVarRef(self, ref_str, blame_tok):
261 # type: (str, Token) -> BracedVarSub
262 """Parse and evaluate value for ${!ref}
263
264 This supports:
265 - 0 to 9 for $0 to $9
266 - @ for "$@" etc.
267
268 See grammar in osh/word_parse.py, which is related to grammar in
269 osh/word_parse.py _ReadBracedVarSub
270
271 Note: declare -n allows 'varname' and 'varname[i]' and 'varname[@]', but it
272 does NOT allow 0 to 9, @, *
273
274 NamerefExpr = NAME Subscript? # this allows @ and * too
275
276 _ResolveNameOrRef currently gives you a 'cell'. So it might not support
277 sh_lvalue.Indexed?
278 """
279 line_reader = reader.StringLineReader(ref_str, self.arena)
280 lexer = self.parse_ctx.MakeLexer(line_reader)
281 w_parser = self.parse_ctx.MakeWordParser(lexer, line_reader)
282
283 src = source.VarRef(blame_tok)
284 with alloc.ctx_SourceCode(self.arena, src):
285 try:
286 bvs_part = w_parser.ParseVarRef()
287 except error.Parse as e:
288 # This prints the inner location
289 self.errfmt.PrettyPrintError(e)
290
291 # this affects builtins 'unset' and 'printf'
292 e_die("Invalid var ref expression", blame_tok)
293
294 return bvs_part
295
296
297class ArithEvaluator(object):
298 """Shared between arith and bool evaluators.
299
300 They both:
301
302 1. Convert strings to integers, respecting shopt -s strict_arith.
303 2. Look up variables and evaluate words.
304 """
305
306 def __init__(
307 self,
308 mem, # type: state.Mem
309 exec_opts, # type: optview.Exec
310 mutable_opts, # type: state.MutableOpts
311 parse_ctx, # type: Optional[parse_lib.ParseContext]
312 errfmt, # type: ErrorFormatter
313 ):
314 # type: (...) -> None
315 self.word_ev = None # type: word_eval.StringWordEvaluator
316 self.mem = mem
317 self.exec_opts = exec_opts
318 self.mutable_opts = mutable_opts
319 self.parse_ctx = parse_ctx
320 self.errfmt = errfmt
321
322 def CheckCircularDeps(self):
323 # type: () -> None
324 assert self.word_ev is not None
325
326 def _StringToBigInt(self, s, blame_loc):
327 # type: (str, loc_t) -> mops.BigInt
328 """Use bash-like rules to coerce a string to an integer.
329
330 Runtime parsing enables silly stuff like $(( $(echo 1)$(echo 2) + 1 )) => 13
331
332 0xAB -- hex constant
333 042 -- octal constant
334 42 -- decimal constant
335 64#z -- arbitrary base constant
336
337 bare word: variable
338 quoted word: string (not done?)
339 """
340 if s.startswith('0x'):
341 try:
342 integer = mops.FromStr(s, 16)
343 except ValueError:
344 e_strict('Invalid hex constant %r' % s, blame_loc)
345 # TODO: don't truncate
346 return integer
347
348 if s.startswith('0'):
349 try:
350 integer = mops.FromStr(s, 8)
351 except ValueError:
352 e_strict('Invalid octal constant %r' % s, blame_loc)
353 return integer
354
355 b, digits = mylib.split_once(s, '#') # see if it has #
356 if digits is not None:
357 try:
358 base = int(b) # machine integer, not BigInt
359 except ValueError:
360 e_strict('Invalid base for numeric constant %r' % b, blame_loc)
361
362 integer = mops.ZERO
363 for ch in digits:
364 if IsLower(ch):
365 digit = ord(ch) - ord('a') + 10
366 elif IsUpper(ch):
367 digit = ord(ch) - ord('A') + 36
368 elif ch == '@': # horrible syntax
369 digit = 62
370 elif ch == '_':
371 digit = 63
372 elif ch.isdigit():
373 digit = int(ch)
374 else:
375 e_strict('Invalid digits for numeric constant %r' % digits,
376 blame_loc)
377
378 if digit >= base:
379 e_strict(
380 'Digits %r out of range for base %d' % (digits, base),
381 blame_loc)
382
383 #integer = integer * base + digit
384 integer = mops.Add(mops.Mul(integer, mops.BigInt(base)),
385 mops.BigInt(digit))
386 return integer
387
388 try:
389 # Normal base 10 integer. This includes negative numbers like '-42'.
390 integer = mops.FromStr(s)
391 except ValueError:
392 # doesn't look like an integer
393
394 # note: 'test' and '[' never evaluate recursively
395 if self.parse_ctx:
396 arena = self.parse_ctx.arena
397
398 # Special case so we don't get EOF error
399 if len(s.strip()) == 0:
400 return mops.ZERO
401
402 # For compatibility: Try to parse it as an expression and evaluate it.
403 a_parser = self.parse_ctx.MakeArithParser(s)
404
405 # TODO: Fill in the variable name
406 with alloc.ctx_SourceCode(arena,
407 source.Variable(None, blame_loc)):
408 try:
409 node2 = a_parser.Parse() # may raise error.Parse
410 except error.Parse as e:
411 self.errfmt.PrettyPrintError(e)
412 e_die('Parse error in recursive arithmetic',
413 e.location)
414
415 # Prevent infinite recursion of $(( 1x )) -- it's a word that evaluates
416 # to itself, and you don't want to reparse it as a word.
417 if node2.tag() == arith_expr_e.Word:
418 e_die("Invalid integer constant %r" % s, blame_loc)
419
420 if self.exec_opts.eval_unsafe_arith():
421 integer = self.EvalToBigInt(node2)
422 else:
423 # BoolEvaluator doesn't have parse_ctx or mutable_opts
424 assert self.mutable_opts is not None
425
426 # We don't need to flip _allow_process_sub, because they can't be
427 # parsed. See spec/bugs.test.sh.
428 with state.ctx_Option(self.mutable_opts,
429 [option_i._allow_command_sub],
430 False):
431 integer = self.EvalToBigInt(node2)
432
433 else:
434 if len(s.strip()) == 0 or match.IsValidVarName(s):
435 # x42 could evaluate to 0
436 e_strict("Invalid integer constant %r" % s, blame_loc)
437 else:
438 # 42x is always fatal!
439 e_die("Invalid integer constant %r" % s, blame_loc)
440
441 return integer
442
443 def _ValToIntOrError(self, val, blame):
444 # type: (value_t, arith_expr_t) -> mops.BigInt
445 try:
446 UP_val = val
447 with tagswitch(val) as case:
448 if case(value_e.Undef):
449 # 'nounset' already handled before got here
450 # Happens upon a[undefined]=42, which unfortunately turns into a[0]=42.
451 e_strict('Undefined value in arithmetic context',
452 loc.Arith(blame))
453
454 elif case(value_e.Int):
455 val = cast(value.Int, UP_val)
456 return val.i
457
458 elif case(value_e.Str):
459 val = cast(value.Str, UP_val)
460 # calls e_strict
461 return self._StringToBigInt(val.s, loc.Arith(blame))
462
463 except error.Strict as e:
464 if self.exec_opts.strict_arith():
465 raise
466 else:
467 return mops.ZERO
468
469 # Arrays and associative arrays always fail -- not controlled by
470 # strict_arith.
471 # In bash, (( a )) is like (( a[0] )), but I don't want that.
472 # And returning '0' gives different results.
473 e_die(
474 "Expected a value convertible to integer, got %s" %
475 ui.ValType(val), loc.Arith(blame))
476
477 def _EvalLhsAndLookupArith(self, node):
478 # type: (arith_expr_t) -> Tuple[mops.BigInt, sh_lvalue_t]
479 """ For x = y and x += y and ++x """
480
481 lval = self.EvalArithLhs(node)
482 val = OldValue(lval, self.mem, self.exec_opts)
483
484 # BASH_LINENO, arr (array name without strict_array), etc.
485 if (val.tag() in (value_e.BashArray, value_e.BashAssoc) and
486 lval.tag() == sh_lvalue_e.Var):
487 named_lval = cast(LeftName, lval)
488 if word_eval.ShouldArrayDecay(named_lval.name, self.exec_opts):
489 if val.tag() == value_e.BashArray:
490 lval = sh_lvalue.Indexed(named_lval.name, 0, loc.Missing)
491 elif val.tag() == value_e.BashAssoc:
492 lval = sh_lvalue.Keyed(named_lval.name, '0', loc.Missing)
493 val = word_eval.DecayArray(val)
494
495 # This error message could be better, but we already have one
496 #if val.tag() == value_e.BashArray:
497 # e_die("Can't use assignment like ++ or += on arrays")
498
499 i = self._ValToIntOrError(val, node)
500 return i, lval
501
502 def _Store(self, lval, new_int):
503 # type: (sh_lvalue_t, mops.BigInt) -> None
504 val = value.Str(mops.ToStr(new_int))
505 state.OshLanguageSetValue(self.mem, lval, val)
506
507 def EvalToBigInt(self, node):
508 # type: (arith_expr_t) -> mops.BigInt
509 """Used externally by ${a[i+1]} and ${a:start:len}.
510
511 Also used internally.
512 """
513 val = self.Eval(node)
514
515 # BASH_LINENO, arr (array name without strict_array), etc.
516 if (val.tag() in (value_e.BashArray, value_e.BashAssoc) and
517 node.tag() == arith_expr_e.VarSub):
518 vsub = cast(Token, node)
519 if word_eval.ShouldArrayDecay(lexer.LazyStr(vsub), self.exec_opts):
520 val = word_eval.DecayArray(val)
521
522 i = self._ValToIntOrError(val, node)
523 return i
524
525 def EvalToInt(self, node):
526 # type: (arith_expr_t) -> int
527 return mops.BigTruncate(self.EvalToBigInt(node))
528
529 def Eval(self, node):
530 # type: (arith_expr_t) -> value_t
531 """
532 Returns:
533 None for Undef (e.g. empty cell) TODO: Don't return 0!
534 int for Str
535 List[int] for BashArray
536 Dict[str, str] for BashAssoc (TODO: Should we support this?)
537
538 NOTE: (( A['x'] = 'x' )) and (( x = A['x'] )) are syntactically valid in
539 bash, but don't do what you'd think. 'x' sometimes a variable name and
540 sometimes a key.
541 """
542 # OSH semantics: Variable NAMES cannot be formed dynamically; but INTEGERS
543 # can. ${foo:-3}4 is OK. $? will be a compound word too, so we don't have
544 # to handle that as a special case.
545
546 UP_node = node
547 with tagswitch(node) as case:
548 if case(arith_expr_e.EmptyZero): # $(( ))
549 return value.Int(mops.ZERO) # Weird axiom
550
551 elif case(arith_expr_e.EmptyOne): # for (( ; ; ))
552 return value.Int(mops.ONE)
553
554 elif case(arith_expr_e.VarSub): # $(( x )) (can be array)
555 vsub = cast(Token, UP_node)
556 var_name = lexer.LazyStr(vsub)
557 val = self.mem.GetValue(var_name)
558 if val.tag() == value_e.Undef and self.exec_opts.nounset():
559 e_die('Undefined variable %r' % var_name, vsub)
560 return val
561
562 elif case(arith_expr_e.Word): # $(( $x )) $(( ${x}${y} )), etc.
563 w = cast(CompoundWord, UP_node)
564 return self.word_ev.EvalWordToString(w)
565
566 elif case(arith_expr_e.UnaryAssign): # a++
567 node = cast(arith_expr.UnaryAssign, UP_node)
568
569 op_id = node.op_id
570 old_big, lval = self._EvalLhsAndLookupArith(node.child)
571
572 if op_id == Id.Node_PostDPlus: # post-increment
573 new_big = mops.Add(old_big, mops.ONE)
574 result = old_big
575
576 elif op_id == Id.Node_PostDMinus: # post-decrement
577 new_big = mops.Sub(old_big, mops.ONE)
578 result = old_big
579
580 elif op_id == Id.Arith_DPlus: # pre-increment
581 new_big = mops.Add(old_big, mops.ONE)
582 result = new_big
583
584 elif op_id == Id.Arith_DMinus: # pre-decrement
585 new_big = mops.Sub(old_big, mops.ONE)
586 result = new_big
587
588 else:
589 raise AssertionError(op_id)
590
591 self._Store(lval, new_big)
592 return value.Int(result)
593
594 elif case(arith_expr_e.BinaryAssign): # a=1, a+=5, a[1]+=5
595 node = cast(arith_expr.BinaryAssign, UP_node)
596 op_id = node.op_id
597
598 if op_id == Id.Arith_Equal:
599 # Don't really need a span ID here, because tdop.CheckLhsExpr should
600 # have done all the validation.
601 lval = self.EvalArithLhs(node.left)
602 rhs_big = self.EvalToBigInt(node.right)
603
604 self._Store(lval, rhs_big)
605 return value.Int(rhs_big)
606
607 old_big, lval = self._EvalLhsAndLookupArith(node.left)
608 rhs_big = self.EvalToBigInt(node.right)
609
610 if op_id == Id.Arith_PlusEqual:
611 new_big = mops.Add(old_big, rhs_big)
612 elif op_id == Id.Arith_MinusEqual:
613 new_big = mops.Sub(old_big, rhs_big)
614 elif op_id == Id.Arith_StarEqual:
615 new_big = mops.Mul(old_big, rhs_big)
616
617 elif op_id == Id.Arith_SlashEqual:
618 if mops.Equal(rhs_big, mops.ZERO):
619 e_die('Divide by zero') # TODO: location
620 new_big = num.IntDivide(old_big, rhs_big)
621
622 elif op_id == Id.Arith_PercentEqual:
623 if mops.Equal(rhs_big, mops.ZERO):
624 e_die('Divide by zero') # TODO: location
625 new_big = num.IntRemainder(old_big, rhs_big)
626
627 elif op_id == Id.Arith_DGreatEqual:
628 new_big = mops.RShift(old_big, rhs_big)
629 elif op_id == Id.Arith_DLessEqual:
630 new_big = mops.LShift(old_big, rhs_big)
631 elif op_id == Id.Arith_AmpEqual:
632 new_big = mops.BitAnd(old_big, rhs_big)
633 elif op_id == Id.Arith_PipeEqual:
634 new_big = mops.BitOr(old_big, rhs_big)
635 elif op_id == Id.Arith_CaretEqual:
636 new_big = mops.BitXor(old_big, rhs_big)
637 else:
638 raise AssertionError(op_id) # shouldn't get here
639
640 self._Store(lval, new_big)
641 return value.Int(new_big)
642
643 elif case(arith_expr_e.Unary):
644 node = cast(arith_expr.Unary, UP_node)
645 op_id = node.op_id
646
647 i = self.EvalToBigInt(node.child)
648
649 if op_id == Id.Node_UnaryPlus: # +i
650 result = i
651 elif op_id == Id.Node_UnaryMinus: # -i
652 result = mops.Sub(mops.ZERO, i)
653
654 elif op_id == Id.Arith_Bang: # logical negation
655 if mops.Equal(i, mops.ZERO):
656 result = mops.ONE
657 else:
658 result = mops.ZERO
659 elif op_id == Id.Arith_Tilde: # bitwise complement
660 result = mops.BitNot(i)
661 else:
662 raise AssertionError(op_id) # shouldn't get here
663
664 return value.Int(result)
665
666 elif case(arith_expr_e.Binary):
667 node = cast(arith_expr.Binary, UP_node)
668 op_id = node.op.id
669
670 # Short-circuit evaluation for || and &&.
671 if op_id == Id.Arith_DPipe:
672 lhs_big = self.EvalToBigInt(node.left)
673 if mops.Equal(lhs_big, mops.ZERO):
674 rhs_big = self.EvalToBigInt(node.right)
675 if mops.Equal(rhs_big, mops.ZERO):
676 result = mops.ZERO # false
677 else:
678 result = mops.ONE # true
679 else:
680 result = mops.ONE # true
681 return value.Int(result)
682
683 if op_id == Id.Arith_DAmp:
684 lhs_big = self.EvalToBigInt(node.left)
685 if mops.Equal(lhs_big, mops.ZERO):
686 result = mops.ZERO # false
687 else:
688 rhs_big = self.EvalToBigInt(node.right)
689 if mops.Equal(rhs_big, mops.ZERO):
690 result = mops.ZERO # false
691 else:
692 result = mops.ONE # true
693 return value.Int(result)
694
695 if op_id == Id.Arith_LBracket:
696 # NOTE: Similar to bracket_op_e.ArrayIndex in osh/word_eval.py
697
698 left = self.Eval(node.left)
699 UP_left = left
700 with tagswitch(left) as case:
701 if case(value_e.BashArray):
702 array_val = cast(value.BashArray, UP_left)
703 index = mops.BigTruncate(
704 self.EvalToBigInt(node.right))
705 s = word_eval.GetArrayItem(array_val.strs, index)
706
707 elif case(value_e.BashAssoc):
708 left = cast(value.BashAssoc, UP_left)
709 key = self.EvalWordToString(node.right)
710 s = left.d.get(key)
711
712 elif case(value_e.Str):
713 left = cast(value.Str, UP_left)
714 if self.exec_opts.strict_arith():
715 e_die(
716 "Value of type Str can't be indexed (strict_arith)",
717 node.op)
718 index = self.EvalToBigInt(node.right)
719 # s[0] evaluates to s
720 # s[1] evaluates to Undef
721 s = left.s if mops.Equal(index,
722 mops.ZERO) else None
723
724 elif case(value_e.Undef):
725 if self.exec_opts.strict_arith():
726 e_die(
727 "Value of type Undef can't be indexed (strict_arith)",
728 node.op)
729 s = None # value.Undef
730
731 # There isn't a way to distinguish Undef vs. empty
732 # string, even with set -o nounset?
733 # s = ''
734
735 else:
736 # TODO: Add error context
737 e_die(
738 "Value of type %s can't be indexed" %
739 ui.ValType(left), node.op)
740
741 if s is None:
742 val = value.Undef
743 else:
744 val = value.Str(s)
745
746 return val
747
748 if op_id == Id.Arith_Comma:
749 self.EvalToBigInt(node.left) # throw away result
750 result = self.EvalToBigInt(node.right)
751 return value.Int(result)
752
753 # Rest are integers
754 lhs_big = self.EvalToBigInt(node.left)
755 rhs_big = self.EvalToBigInt(node.right)
756
757 if op_id == Id.Arith_Plus:
758 result = mops.Add(lhs_big, rhs_big)
759 elif op_id == Id.Arith_Minus:
760 result = mops.Sub(lhs_big, rhs_big)
761 elif op_id == Id.Arith_Star:
762 result = mops.Mul(lhs_big, rhs_big)
763 elif op_id == Id.Arith_Slash:
764 if mops.Equal(rhs_big, mops.ZERO):
765 e_die('Divide by zero', node.op)
766 result = num.IntDivide(lhs_big, rhs_big)
767
768 elif op_id == Id.Arith_Percent:
769 if mops.Equal(rhs_big, mops.ZERO):
770 e_die('Divide by zero', node.op)
771 result = num.IntRemainder(lhs_big, rhs_big)
772
773 elif op_id == Id.Arith_DStar:
774 if mops.Greater(mops.ZERO, rhs_big):
775 e_die("Exponent can't be a negative number",
776 loc.Arith(node.right))
777 result = num.Exponent(lhs_big, rhs_big)
778
779 elif op_id == Id.Arith_DEqual:
780 result = mops.FromBool(mops.Equal(lhs_big, rhs_big))
781 elif op_id == Id.Arith_NEqual:
782 result = mops.FromBool(not mops.Equal(lhs_big, rhs_big))
783 elif op_id == Id.Arith_Great:
784 result = mops.FromBool(mops.Greater(lhs_big, rhs_big))
785 elif op_id == Id.Arith_GreatEqual:
786 result = mops.FromBool(
787 mops.Greater(lhs_big, rhs_big) or
788 mops.Equal(lhs_big, rhs_big))
789 elif op_id == Id.Arith_Less:
790 result = mops.FromBool(mops.Greater(rhs_big, lhs_big))
791 elif op_id == Id.Arith_LessEqual:
792 result = mops.FromBool(
793 mops.Greater(rhs_big, lhs_big) or
794 mops.Equal(lhs_big, rhs_big))
795
796 elif op_id == Id.Arith_Pipe:
797 result = mops.BitOr(lhs_big, rhs_big)
798 elif op_id == Id.Arith_Amp:
799 result = mops.BitAnd(lhs_big, rhs_big)
800 elif op_id == Id.Arith_Caret:
801 result = mops.BitXor(lhs_big, rhs_big)
802
803 # Note: how to define shift of negative numbers?
804 elif op_id == Id.Arith_DLess:
805 result = mops.LShift(lhs_big, rhs_big)
806 elif op_id == Id.Arith_DGreat:
807 result = mops.RShift(lhs_big, rhs_big)
808 else:
809 raise AssertionError(op_id)
810
811 return value.Int(result)
812
813 elif case(arith_expr_e.TernaryOp):
814 node = cast(arith_expr.TernaryOp, UP_node)
815
816 cond = self.EvalToBigInt(node.cond)
817 if mops.Equal(cond, mops.ZERO):
818 return self.Eval(node.false_expr)
819 else:
820 return self.Eval(node.true_expr)
821
822 else:
823 raise AssertionError(node.tag())
824
825 raise AssertionError('for -Wreturn-type in C++')
826
827 def EvalWordToString(self, node, blame_loc=loc.Missing):
828 # type: (arith_expr_t, loc_t) -> str
829 """
830 Raises:
831 error.FatalRuntime if the expression isn't a string
832 or if it contains a bare variable like a[x]
833
834 These are allowed because they're unambiguous, unlike a[x]
835
836 a[$x] a["$x"] a["x"] a['x']
837 """
838 UP_node = node
839 if node.tag() == arith_expr_e.Word: # $(( $x )) $(( ${x}${y} )), etc.
840 w = cast(CompoundWord, UP_node)
841 val = self.word_ev.EvalWordToString(w)
842 return val.s
843 else:
844 # A[x] is the "Parsing Bash is Undecidable" problem
845 # It is a string or var name?
846 # (It's parsed as arith_expr.VarSub)
847 e_die(
848 "Assoc array keys must be strings: $x 'x' \"$x\" etc. (OILS-ERR-101)",
849 blame_loc)
850
851 def EvalShellLhs(self, node, which_scopes):
852 # type: (sh_lhs_t, scope_t) -> sh_lvalue_t
853 """Evaluate a shell LHS expression
854
855 For a=b and a[x]=b etc.
856 """
857 assert isinstance(node, sh_lhs_t), node
858
859 UP_node = node
860 lval = None # type: sh_lvalue_t
861 with tagswitch(node) as case:
862 if case(sh_lhs_e.Name): # a=x
863 node = cast(sh_lhs.Name, UP_node)
864 assert node.name is not None
865
866 lval1 = LeftName(node.name, node.left)
867 lval = lval1
868
869 elif case(sh_lhs_e.IndexedName): # a[1+2]=x
870 node = cast(sh_lhs.IndexedName, UP_node)
871 assert node.name is not None
872
873 if self.mem.IsBashAssoc(node.name):
874 key = self.EvalWordToString(node.index,
875 blame_loc=node.left)
876 # node.left points to A[ in A[x]=1
877 lval2 = sh_lvalue.Keyed(node.name, key, node.left)
878 lval = lval2
879 else:
880 index = mops.BigTruncate(self.EvalToBigInt(node.index))
881 lval3 = sh_lvalue.Indexed(node.name, index, node.left)
882 lval = lval3
883
884 else:
885 raise AssertionError(node.tag())
886
887 return lval
888
889 def _VarNameOrWord(self, anode):
890 # type: (arith_expr_t) -> Tuple[Optional[str], loc_t]
891 """
892 Returns a variable name if the arith node can be interpreted that way.
893 """
894 UP_anode = anode
895 with tagswitch(anode) as case:
896 if case(arith_expr_e.VarSub):
897 tok = cast(Token, UP_anode)
898 return (lexer.LazyStr(tok), tok)
899
900 elif case(arith_expr_e.Word):
901 w = cast(CompoundWord, UP_anode)
902 var_name = self.EvalWordToString(w)
903 return (var_name, w)
904
905 no_str = None # type: str
906 return (no_str, loc.Missing)
907
908 def EvalArithLhs(self, anode):
909 # type: (arith_expr_t) -> sh_lvalue_t
910 """
911 For (( a[x] = 1 )) etc.
912 """
913 UP_anode = anode
914 if anode.tag() == arith_expr_e.Binary:
915 anode = cast(arith_expr.Binary, UP_anode)
916 if anode.op.id == Id.Arith_LBracket:
917 var_name, blame_loc = self._VarNameOrWord(anode.left)
918
919 # (( 1[2] = 3 )) isn't valid
920 if not match.IsValidVarName(var_name):
921 e_die('Invalid variable name %r' % var_name, blame_loc)
922
923 if var_name is not None:
924 if self.mem.IsBashAssoc(var_name):
925 arith_loc = location.TokenForArith(anode)
926 key = self.EvalWordToString(anode.right,
927 blame_loc=arith_loc)
928 return sh_lvalue.Keyed(var_name, key, blame_loc)
929 else:
930 index = mops.BigTruncate(self.EvalToBigInt(
931 anode.right))
932 return sh_lvalue.Indexed(var_name, index, blame_loc)
933
934 var_name, blame_loc = self._VarNameOrWord(anode)
935 if var_name is not None:
936 return LeftName(var_name, blame_loc)
937
938 # e.g. unset 'x-y'. status 2 for runtime parse error
939 e_die_status(2, 'Invalid LHS to modify', blame_loc)
940
941
942class BoolEvaluator(ArithEvaluator):
943 """This is also an ArithEvaluator because it has to understand.
944
945 [[ x -eq 3 ]]
946
947 where x='1+2'
948 """
949
950 def __init__(
951 self,
952 mem, # type: state.Mem
953 exec_opts, # type: optview.Exec
954 mutable_opts, # type: Optional[state.MutableOpts]
955 parse_ctx, # type: Optional[parse_lib.ParseContext]
956 errfmt, # type: ErrorFormatter
957 always_strict=False # type: bool
958 ):
959 # type: (...) -> None
960 ArithEvaluator.__init__(self, mem, exec_opts, mutable_opts, parse_ctx,
961 errfmt)
962 self.always_strict = always_strict
963
964 def _StringToBigIntOrError(self, s, blame_word=None):
965 # type: (str, Optional[word_t]) -> mops.BigInt
966 """Used by both [[ $x -gt 3 ]] and (( $x ))."""
967 if blame_word:
968 location = loc.Word(blame_word) # type: loc_t
969 else:
970 location = loc.Missing
971
972 try:
973 i = self._StringToBigInt(s, location)
974 except error.Strict as e:
975 if self.always_strict or self.exec_opts.strict_arith():
976 raise
977 else:
978 i = mops.ZERO
979 return i
980
981 def _EvalCompoundWord(self, word, eval_flags=0):
982 # type: (word_t, int) -> str
983 val = self.word_ev.EvalWordToString(word, eval_flags)
984 return val.s
985
986 def EvalB(self, node):
987 # type: (bool_expr_t) -> bool
988
989 UP_node = node
990 with tagswitch(node) as case:
991 if case(bool_expr_e.WordTest):
992 node = cast(bool_expr.WordTest, UP_node)
993 s = self._EvalCompoundWord(node.w)
994 return bool(s)
995
996 elif case(bool_expr_e.LogicalNot):
997 node = cast(bool_expr.LogicalNot, UP_node)
998 b = self.EvalB(node.child)
999 return not b
1000
1001 elif case(bool_expr_e.LogicalAnd):
1002 node = cast(bool_expr.LogicalAnd, UP_node)
1003 # Short-circuit evaluation
1004 if self.EvalB(node.left):
1005 return self.EvalB(node.right)
1006 else:
1007 return False
1008
1009 elif case(bool_expr_e.LogicalOr):
1010 node = cast(bool_expr.LogicalOr, UP_node)
1011 if self.EvalB(node.left):
1012 return True
1013 else:
1014 return self.EvalB(node.right)
1015
1016 elif case(bool_expr_e.Unary):
1017 node = cast(bool_expr.Unary, UP_node)
1018 op_id = node.op_id
1019 s = self._EvalCompoundWord(node.child)
1020
1021 # Now dispatch on arg type
1022 arg_type = consts.BoolArgType(
1023 op_id) # could be static in the LST?
1024
1025 if arg_type == bool_arg_type_e.Path:
1026 return bool_stat.DoUnaryOp(op_id, s)
1027
1028 if arg_type == bool_arg_type_e.Str:
1029 if op_id == Id.BoolUnary_z:
1030 return not bool(s)
1031 if op_id == Id.BoolUnary_n:
1032 return bool(s)
1033
1034 raise AssertionError(op_id) # should never happen
1035
1036 if arg_type == bool_arg_type_e.Other:
1037 if op_id == Id.BoolUnary_t:
1038 return bool_stat.isatty(s, node.child)
1039
1040 # See whether 'set -o' options have been set
1041 if op_id == Id.BoolUnary_o:
1042 index = consts.OptionNum(s)
1043 if index == 0:
1044 return False
1045 else:
1046 return self.exec_opts.opt0_array[index]
1047
1048 if op_id == Id.BoolUnary_v:
1049 val = self.mem.GetValue(s)
1050 return val.tag() != value_e.Undef
1051
1052 e_die("%s isn't implemented" %
1053 ui.PrettyId(op_id)) # implicit location
1054
1055 raise AssertionError(arg_type)
1056
1057 elif case(bool_expr_e.Binary):
1058 node = cast(bool_expr.Binary, UP_node)
1059
1060 op_id = node.op_id
1061 # Whether to glob escape
1062 eval_flags = 0
1063 with switch(op_id) as case2:
1064 if case2(Id.BoolBinary_GlobEqual, Id.BoolBinary_GlobDEqual,
1065 Id.BoolBinary_GlobNEqual):
1066 eval_flags |= word_eval.QUOTE_FNMATCH
1067 elif case2(Id.BoolBinary_EqualTilde):
1068 eval_flags |= word_eval.QUOTE_ERE
1069
1070 s1 = self._EvalCompoundWord(node.left)
1071 s2 = self._EvalCompoundWord(node.right, eval_flags)
1072
1073 # Now dispatch on arg type
1074 arg_type = consts.BoolArgType(op_id)
1075
1076 if arg_type == bool_arg_type_e.Path:
1077 return bool_stat.DoBinaryOp(op_id, s1, s2)
1078
1079 if arg_type == bool_arg_type_e.Int:
1080 # NOTE: We assume they are constants like [[ 3 -eq 3 ]].
1081 # Bash also allows [[ 1+2 -eq 3 ]].
1082 i1 = self._StringToBigIntOrError(s1, blame_word=node.left)
1083 i2 = self._StringToBigIntOrError(s2, blame_word=node.right)
1084
1085 if op_id == Id.BoolBinary_eq:
1086 return mops.Equal(i1, i2)
1087 if op_id == Id.BoolBinary_ne:
1088 return not mops.Equal(i1, i2)
1089 if op_id == Id.BoolBinary_gt:
1090 return mops.Greater(i1, i2)
1091 if op_id == Id.BoolBinary_ge:
1092 return mops.Greater(i1, i2) or mops.Equal(i1, i2)
1093 if op_id == Id.BoolBinary_lt:
1094 return mops.Greater(i2, i1)
1095 if op_id == Id.BoolBinary_le:
1096 return mops.Greater(i2, i1) or mops.Equal(i1, i2)
1097
1098 raise AssertionError(op_id) # should never happen
1099
1100 if arg_type == bool_arg_type_e.Str:
1101 fnmatch_flags = (FNM_CASEFOLD
1102 if self.exec_opts.nocasematch() else 0)
1103
1104 if op_id in (Id.BoolBinary_GlobEqual,
1105 Id.BoolBinary_GlobDEqual):
1106 #log('Matching %s against pattern %s', s1, s2)
1107 return libc.fnmatch(s2, s1, fnmatch_flags)
1108
1109 if op_id == Id.BoolBinary_GlobNEqual:
1110 return not libc.fnmatch(s2, s1, fnmatch_flags)
1111
1112 if op_id in (Id.BoolBinary_Equal, Id.BoolBinary_DEqual):
1113 return s1 == s2
1114
1115 if op_id == Id.BoolBinary_NEqual:
1116 return s1 != s2
1117
1118 if op_id == Id.BoolBinary_EqualTilde:
1119 # TODO: This should go to --debug-file
1120 #log('Matching %r against regex %r', s1, s2)
1121 regex_flags = (REG_ICASE
1122 if self.exec_opts.nocasematch() else 0)
1123
1124 try:
1125 indices = libc.regex_search(s2, regex_flags, s1, 0)
1126 except ValueError as e:
1127 # Status 2 indicates a regex parse error. This is fatal in OSH but
1128 # not in bash, which treats [[ like a command with an exit code.
1129 e_die_status(2, e.message, loc.Word(node.right))
1130
1131 if indices is not None:
1132 self.mem.SetRegexMatch(
1133 RegexMatch(s1, indices, eggex_ops.No))
1134 return True
1135 else:
1136 self.mem.SetRegexMatch(regex_match.No)
1137 return False
1138
1139 if op_id == Id.Op_Less:
1140 return str_cmp(s1, s2) < 0
1141
1142 if op_id == Id.Op_Great:
1143 return str_cmp(s1, s2) > 0
1144
1145 raise AssertionError(op_id) # should never happen
1146
1147 raise AssertionError(node.tag())