def MakeReplacer(pat, replace_str, slash_spid):
"""Helper for ${x/pat/replace}
Parses 'pat' and returns either a _GlobReplacer or a _ConstStringReplacer.
Using these objects is more efficient when performing the same operation on
multiple strings.
"""
regex, warnings = glob_.GlobToERE(pat)
if warnings:
# TODO: Add strict mode and expose warnings.
pass
if regex is None:
return _ConstStringReplacer(pat, replace_str)
else:
return _GlobReplacer(regex, replace_str, slash_spid)
def testGlobParser(self):
CASES = [
# (glob input, expected AST, expected extended regexp, has error)
('*.py', '.*\.py', False),
('*.?', '.*\..', False),
('<*>', '<.*>', False),
('\**+', '\*.*\+', False),
('\**', '\*.*', False),
('*.[ch]pp', '.*\.[ch]pp', False),
# not globs
('abc', None, False),
('\\*', None, False),
('c:\\foo', None, False),
('strange]one', None, False),
# character class globs
('[[:space:]abc]', '[[:space:]abc]', False),
('[abc]', '[abc]', False),
(r'[\a\b\c]', r'[\a\b\c]', False),
('[abc\[]', '[abc\[]', False),
('[!not]', '[^not]', False),
('[^also_not]', '[^also_not]', False),
('[!*?!\\[]', '[^*?!\\[]', False),
('[!\]foo]', '[^\]foo]', False),
# invalid globs
('not_closed[a-z', None, True),
('[[:spa[ce:]]', None, True),
# Regression test for IndexError.
('[', None, True),
('\\', None, True),
(']', None, False),
]
for glob, expected_ere, expected_err in CASES:
print('===')
print(glob)
regex, warnings = glob_.GlobToERE(glob)
self.assertEqual(
expected_ere, regex, 'Expected %r to translate to %r, got %r' %
(glob, expected_ere, regex))
print('regex : %s' % regex)
print('warnings: %s' % warnings)
def _EvalBracedVarSub(self, part, part_vals, quoted):
"""
Args:
part_vals: output param to append to.
"""
# We have four types of operator that interact.
#
# 1. Bracket: value -> (value, bool maybe_decay_array)
#
# 2. Then these four cases are mutually exclusive:
#
# a. Prefix length: value -> value
# b. Test: value -> part_value[]
# c. Other Suffix: value -> value
# d. no operator: you have a value
#
# That is, we don't have both prefix and suffix operators.
#
# 3. Process maybe_decay_array here before returning.
maybe_decay_array = False # for $*, ${a[*]}, etc.
var_name = None # For ${foo=default}
# 1. Evaluate from (var_name, var_num, token Id) -> value
if part.token.id == Id.VSub_Name:
var_name = part.token.val
val = self.mem.GetVar(var_name)
#log('EVAL NAME %s -> %s', var_name, val)
elif part.token.id == Id.VSub_Number:
var_num = int(part.token.val)
val = self._EvalVarNum(var_num)
else:
# $* decays
val, maybe_decay_array = self._EvalSpecialVar(part.token.id, quoted)
# 2. Bracket: value -> (value v, bool maybe_decay_array)
# maybe_decay_array is for joining ${a[*]} and unquoted ${a[@]} AFTER
# suffix ops are applied. If we take the length with a prefix op, the
# distinction is ignored.
if part.bracket_op:
if part.bracket_op.tag == bracket_op_e.WholeArray:
op_id = part.bracket_op.op_id
if op_id == Id.Lit_At:
if not quoted:
maybe_decay_array = True # ${a[@]} decays but "${a[@]}" doesn't
if val.tag == value_e.Undef:
val = self._EmptyStrArrayOrError(part.token)
elif val.tag == value_e.Str:
e_die("Can't index string with @: %r", val, part=part)
elif val.tag == value_e.StrArray:
# TODO: Is this a no-op? Just leave 'val' alone.
val = value.StrArray(val.strs)
elif op_id == Id.Arith_Star:
maybe_decay_array = True # both ${a[*]} and "${a[*]}" decay
if val.tag == value_e.Undef:
val = self._EmptyStrArrayOrError(part.token)
elif val.tag == value_e.Str:
e_die("Can't index string with *: %r", val, part=part)
elif val.tag == value_e.StrArray:
# TODO: Is this a no-op? Just leave 'val' alone.
# ${a[*]} or "${a[*]}" : maybe_decay_array is always true
val = value.StrArray(val.strs)
else:
raise AssertionError(op_id) # unknown
elif part.bracket_op.tag == bracket_op_e.ArrayIndex:
anode = part.bracket_op.expr
if val.tag == value_e.Undef:
pass # it will be checked later
elif val.tag == value_e.Str:
# Bash treats any string as an array, so we can't add our own
# behavior here without making valid OSH invalid bash.
e_die("Can't index string %r with integer", part.token.val,
token=part.token)
elif val.tag == value_e.StrArray:
index = self.arith_ev.Eval(anode)
try:
# could be None because representation is sparse
s = val.strs[index]
except IndexError:
s = None
if s is None:
val = value.Undef()
else:
val = value.Str(s)
elif val.tag == value_e.AssocArray:
key = self.arith_ev.Eval(anode, int_coerce=False)
try:
val = value.Str(val.d[key])
except KeyError:
val = value.Undef()
else:
raise AssertionError(val.__class__.__name__)
else:
raise AssertionError(part.bracket_op.tag)
if part.prefix_op:
val = self._EmptyStrOrError(val) # maybe error
val = self._ApplyPrefixOp(val, part.prefix_op, token=part.token)
# NOTE: When applying the length operator, we can't have a test or
# suffix afterward. And we don't want to decay the array
elif part.suffix_op:
op = part.suffix_op
if op.tag == suffix_op_e.StringNullary:
if op.op_id == Id.VOp0_P:
prompt = self.prompt_ev.EvalPrompt(val)
val = value.Str(prompt)
elif op.op_id == Id.VOp0_Q:
val = value.Str(string_ops.ShellQuote(val.s))
else:
raise NotImplementedError(op.op_id)
elif op.tag == suffix_op_e.StringUnary:
if LookupKind(part.suffix_op.op_id) == Kind.VTest:
# TODO: Change style to:
# if self._ApplyTestOp(...)
# return
# It should return whether anything was done. If not, we continue to
# the end, where we might throw an error.
assign_part_vals, effect = self._ApplyTestOp(val, part.suffix_op,
quoted, part_vals)
# NOTE: Splicing part_values is necessary because of code like
# ${undef:-'a b' c 'd # e'}. Each part_value can have a different
# do_glob/do_elide setting.
if effect == effect_e.SpliceParts:
return # EARLY RETURN, part_vals mutated
elif effect == effect_e.SpliceAndAssign:
if var_name is None:
# TODO: error context
e_die("Can't assign to special variable")
else:
# NOTE: This decays arrays too! 'set -o strict_array' could
# avoid it.
rhs_str = _DecayPartValuesToString(assign_part_vals,
self.splitter.GetJoinChar())
state.SetLocalString(self.mem, var_name, rhs_str)
return # EARLY RETURN, part_vals mutated
elif effect == effect_e.Error:
raise NotImplementedError
else:
# The old one
#val = self._EmptyStringPartOrError(part_val, quoted)
pass # do nothing, may still be undefined
else:
val = self._EmptyStrOrError(val) # maybe error
# Other suffix: value -> value
val = self._ApplyUnarySuffixOp(val, part.suffix_op)
elif op.tag == suffix_op_e.PatSub: # PatSub, vectorized
val = self._EmptyStrOrError(val) # ${undef//x/y}
# globs are supported in the pattern
pat_val = self.EvalWordToString(op.pat, do_fnmatch=True)
assert pat_val.tag == value_e.Str, pat_val
if op.replace:
replace_val = self.EvalWordToString(op.replace)
assert replace_val.tag == value_e.Str, replace_val
replace_str = replace_val.s
else:
replace_str = ''
regex, warnings = glob_.GlobToERE(pat_val.s)
if warnings:
# TODO:
# - Add 'set -o strict-glob' mode and expose warnings.
# "Glob is not in CANONICAL FORM".
# - Propagate location info back to the 'op.pat' word.
pass
replacer = string_ops.GlobReplacer(regex, replace_str, op.spids[0])
if val.tag == value_e.Str:
s = replacer.Replace(val.s, op)
val = value.Str(s)
elif val.tag == value_e.StrArray:
strs = []
for s in val.strs:
if s is not None:
strs.append(replacer.Replace(s, op))
val = value.StrArray(strs)
else:
raise AssertionError(val.__class__.__name__)
elif op.tag == suffix_op_e.Slice:
val = self._EmptyStrOrError(val) # ${undef:3:1}
if op.begin:
begin = self.arith_ev.Eval(op.begin)
else:
begin = 0
if op.length:
length = self.arith_ev.Eval(op.length)
else:
length = None
if val.tag == value_e.Str: # Slice UTF-8 characters in a string.
s = val.s
try:
if begin < 0:
# It could be negative if we compute unicode length, but that's
# confusing.
# TODO: Instead of attributing it to the word part, it would be
# better if we attributed it to arith_expr begin.
raise util.InvalidSlice(
"The start index of a string slice can't be negative: %d",
begin, part=part)
byte_begin = string_ops.AdvanceUtf8Chars(s, begin, 0)
if length is None:
byte_end = len(s)
else:
if length < 0:
# TODO: Instead of attributing it to the word part, it would be
# better if we attributed it to arith_expr begin.
raise util.InvalidSlice(
"The length of a string slice can't be negative: %d",
length, part=part)
byte_end = string_ops.AdvanceUtf8Chars(s, length, byte_begin)
except (util.InvalidSlice, util.InvalidUtf8) as e:
if self.exec_opts.strict_word_eval:
raise
else:
# TODO:
# - We don't see the error location here, but we see it when set
# -o strict-word-eval.
# - Doesn't make the command exit with 1. It just sets the word
# to empty string.
util.warn(e.UserErrorString())
substr = '' # error condition
else:
substr = s[byte_begin : byte_end]
val = value.Str(substr)
elif val.tag == value_e.StrArray: # Slice array entries.
# NOTE: unset elements don't count towards the length.
strs = []
for s in val.strs[begin:]:
if s is not None:
strs.append(s)
if len(strs) == length: # never true for unspecified length
break
val = value.StrArray(strs)
else:
raise AssertionError(val.__class__.__name__) # Not possible
# After applying suffixes, process maybe_decay_array here.
if maybe_decay_array and val.tag == value_e.StrArray:
val = self._DecayArray(val)
# For the case where there are no prefix or suffix ops.
val = self._EmptyStrOrError(val)
# For example, ${a} evaluates to value_t.Str(), but we want a
# part_value.StringPartValue.
part_val = _ValueToPartValue(val, quoted)
part_vals.append(part_val)