def ShortFlag(self, short_name, arg_type=None, help=None):
# type: (str, Optional[int], Optional[str]) -> None
"""
This is very similar to ShortFlag for FlagSpecAndMore, except we have
separate arity0 and arity1 dicts.
"""
assert short_name.startswith('-'), short_name
assert len(short_name) == 2, short_name
char = short_name[1]
if arg_type is None:
self.arity0.append(char)
else:
self.arity1[char] = SetToArg_(char, _FlagType(arg_type), False)
typ = _FlagType(arg_type)
default_val = _Default(arg_type)
if self.typed:
self.defaults[char] = default_val
else:
# TODO: remove when all builtins converted
self.defaults[char] = value.Undef()
self.fields[char] = typ
def Unset(self, lval, lookup_mode):
"""
Returns:
ok bool, found bool.
ok is false if the cell is read-only.
found is false if the name is not there.
"""
if lval.tag == lvalue_e.Named: # unset x
cell, namespace = self._FindCellAndNamespace(
lval.name, lookup_mode)
if cell:
found = True
if cell.readonly:
return False, found
namespace[lval.name].val = value.Undef()
cell.exported = False
return True, found # found
else:
return True, False
elif lval.tag == lvalue_e.Indexed: # unset a[1]
raise NotImplementedError
else:
raise AssertionError
def ShortOption(self, char, help=None):
# type: (str, Optional[str]) -> None
"""Define an option that can be turned off with + and on with -."""
assert len(char) == 1 # 'r' for -r +r
self.options[char] = True
self.defaults[char] = value.Undef()
# '+' or '-'. TODO: Should we make it a bool?
self.fields[char] = flag_type.Str()
def PlusFlag(self, char, help=None):
# type: (str, Optional[str]) -> None
"""Define an option that can be turned off with + and on with -.
It's actually a ternary value: plus, minus, or unset.
"""
assert len(char) == 1 # 'r' for -r +r
self.plus_flags.append(char)
self.defaults[char] = value.Undef()
# '+' or '-'. TODO: Should we make it a bool?
self.fields[char] = flag_type.Str()
def testGetVar(self):
mem = _InitMem()
# readonly a=x
mem.SetVar(lvalue.Named('a'), value.Str('x'), (var_flags_e.ReadOnly, ),
scope_e.Dynamic)
val = mem.GetVar('a', scope_e.Dynamic)
test_lib.AssertAsdlEqual(self, value.Str('x'), val)
val = mem.GetVar('undef', scope_e.Dynamic)
test_lib.AssertAsdlEqual(self, value.Undef(), val)
def testGetValue(self):
mem = _InitMem()
# readonly a=x
mem.SetValue(lvalue.Named('a'),
value.Str('x'),
scope_e.Dynamic,
flags=state.SetReadOnly)
val = mem.GetValue('a', scope_e.Dynamic)
test_lib.AssertAsdlEqual(self, value.Str('x'), val)
val = mem.GetValue('undef', scope_e.Dynamic)
test_lib.AssertAsdlEqual(self, value.Undef(), val)
def _EvalIndirectArrayExpansion(self, name, index):
"""Expands ${!ref} when $ref has the form `name[index]`.
Args:
name, index: arbitrary strings
Returns:
value, or None if invalid
"""
if not match.IsValidVarName(name):
return None
val = self.mem.GetVar(name)
if val.tag == value_e.StrArray:
if index in ('@', '*'):
# TODO: maybe_decay_array
return value.StrArray(val.strs)
try:
index_num = int(index)
except ValueError:
return None
try:
return value.Str(val.strs[index_num])
except IndexError:
return value.Undef()
elif val.tag == value_e.AssocArray:
if index in ('@', '*'):
raise NotImplementedError
try:
return value.Str(val.d[index])
except KeyError:
return value.Undef()
elif val.tag == value_e.Undef:
return value.Undef()
elif val.tag == value_e.Str:
return None
else:
raise AssertionError
def PyToValue(py_val):
# type: (Any) -> value_t
if py_val is None:
val = value.Undef() # type: value_t
elif isinstance(py_val, bool):
val = value.Bool(py_val)
elif isinstance(py_val, int):
val = value.Int(py_val)
elif isinstance(py_val, float):
val = value.Float() # TODO: ASDL needs float primitive
elif isinstance(py_val, str):
val = value.Str(py_val)
else:
raise AssertionError(py_val)
return val
def GetSpecialVar(self, op_id):
if op_id == Id.VSub_Bang: # $!
n = self.last_job_id
if n == -1:
return value.Undef() # could be an error
elif op_id == Id.VSub_QMark: # $?
# External commands need WIFEXITED test. What about subshells?
n = self.last_status[-1]
elif op_id == Id.VSub_Pound: # $#
n = self.argv_stack[-1].GetNumArgs()
elif op_id == Id.VSub_Dollar: # $$
n = self.root_pid
else:
raise NotImplementedError(op_id)
return value.Str(str(n))
def ShortFlag(self, short_name, arg_type=None, help=None):
# type: (str, Optional[int], Optional[str]) -> None
"""
This is very similar to ShortFlag for FlagSpecAndMore, except we have
separate arity0 and arity1 dicts.
"""
assert short_name.startswith('-'), short_name
assert len(short_name) == 2, short_name
char = short_name[1]
if arg_type is None:
self.arity0[char] = True
else:
self.arity1[char] = args.SetToArg(char, _FlagType(arg_type))
# TODO: callers should pass flag_type
if arg_type is None:
typ = flag_type.Bool() # type: flag_type_t
default = value.Bool(False) # type: value_t
elif arg_type == args.Int:
typ = flag_type.Int()
default = value.Int(-1)
elif arg_type == args.Float:
typ = flag_type.Float()
default = value.Float(0.0)
elif arg_type == args.String:
typ = flag_type.Str()
default = value.Str('')
elif isinstance(arg_type, list):
typ = flag_type.Enum(arg_type)
default = value.Str('') # This isn't valid
else:
raise AssertionError(arg_type)
if self.typed:
self.defaults[char] = default
else:
# TODO: remove when all builtins converted
self.defaults[char] = value.Undef()
self.fields[char] = typ
def _Default(arg_type, arg_default=None):
# type: (Union[None, int, List[str]], Optional[str]) -> value_t
# for enum or string
# note: not using this for integers yet
if arg_default is not None:
return value.Str(arg_default) # early return
if arg_type is None:
default = value.Bool(False) # type: value_t
elif arg_type == args.Int:
default = value.Int(-1) # positive values aren't allowed now
elif arg_type == args.Float:
default = value.Float(0.0)
elif arg_type == args.String:
default = value.Undef() # e.g. read -d '' is NOT the default
elif isinstance(arg_type, list):
default = value.Str('') # This isn't valid
else:
raise AssertionError(arg_type)
return default
def Eval(self, node):
# type: (arith_expr_t) -> value_t
"""
Args:
node: arith_expr_t
Returns:
None for Undef (e.g. empty cell) TODO: Don't return 0!
int for Str
List[int] for MaybeStrArray
Dict[str, str] for AssocArray (TODO: Should we support this?)
NOTE: (( A['x'] = 'x' )) and (( x = A['x'] )) are syntactically valid in
bash, but don't do what you'd think. 'x' sometimes a variable name and
sometimes a key.
"""
# OSH semantics: Variable NAMES cannot be formed dynamically; but INTEGERS
# can. ${foo:-3}4 is OK. $? will be a compound word too, so we don't have
# to handle that as a special case.
UP_node = node
with tagswitch(node) as case:
if case(arith_expr_e.VarRef): # $(( x )) (can be array)
node = cast(arith_expr__VarRef, UP_node)
tok = node.token
return _LookupVar(tok.val, self.mem, self.exec_opts)
elif case(
arith_expr_e.ArithWord): # $(( $x )) $(( ${x}${y} )), etc.
node = cast(arith_expr__ArithWord, UP_node)
return self.word_ev.EvalWordToString(node.w)
elif case(arith_expr_e.UnaryAssign): # a++
node = cast(arith_expr__UnaryAssign, UP_node)
op_id = node.op_id
old_int, lval = self._EvalLhsAndLookupArith(node.child)
if op_id == Id.Node_PostDPlus: # post-increment
new_int = old_int + 1
ret = old_int
elif op_id == Id.Node_PostDMinus: # post-decrement
new_int = old_int - 1
ret = old_int
elif op_id == Id.Arith_DPlus: # pre-increment
new_int = old_int + 1
ret = new_int
elif op_id == Id.Arith_DMinus: # pre-decrement
new_int = old_int - 1
ret = new_int
else:
raise AssertionError(op_id)
#log('old %d new %d ret %d', old_int, new_int, ret)
self._Store(lval, new_int)
return value.Int(ret)
elif case(arith_expr_e.BinaryAssign): # a=1, a+=5, a[1]+=5
node = cast(arith_expr__BinaryAssign, UP_node)
op_id = node.op_id
if op_id == Id.Arith_Equal:
lval = _EvalLhsArith(node.left, self.mem, self)
# Disallowing (( a = myarray ))
# It has to be an integer
rhs_int = self.EvalToInt(node.right)
self._Store(lval, rhs_int)
return value.Int(rhs_int)
old_int, lval = self._EvalLhsAndLookupArith(node.left)
rhs = self.EvalToInt(node.right)
if op_id == Id.Arith_PlusEqual:
new_int = old_int + rhs
elif op_id == Id.Arith_MinusEqual:
new_int = old_int - rhs
elif op_id == Id.Arith_StarEqual:
new_int = old_int * rhs
elif op_id == Id.Arith_SlashEqual:
if rhs == 0:
e_die('Divide by zero') # TODO: location
new_int = old_int / rhs
elif op_id == Id.Arith_PercentEqual:
if rhs == 0:
e_die('Divide by zero') # TODO: location
new_int = old_int % rhs
elif op_id == Id.Arith_DGreatEqual:
new_int = old_int >> rhs
elif op_id == Id.Arith_DLessEqual:
new_int = old_int << rhs
elif op_id == Id.Arith_AmpEqual:
new_int = old_int & rhs
elif op_id == Id.Arith_PipeEqual:
new_int = old_int | rhs
elif op_id == Id.Arith_CaretEqual:
new_int = old_int ^ rhs
else:
raise AssertionError(op_id) # shouldn't get here
self._Store(lval, new_int)
return value.Int(new_int)
elif case(arith_expr_e.Unary):
node = cast(arith_expr__Unary, UP_node)
op_id = node.op_id
i = self.EvalToInt(node.child)
if op_id == Id.Node_UnaryPlus:
ret = i
elif op_id == Id.Node_UnaryMinus:
ret = -i
elif op_id == Id.Arith_Bang: # logical negation
ret = 1 if i == 0 else 0
elif op_id == Id.Arith_Tilde: # bitwise complement
ret = ~i
else:
raise AssertionError(op_id) # shouldn't get here
return value.Int(ret)
elif case(arith_expr_e.Binary):
node = cast(arith_expr__Binary, UP_node)
op_id = node.op_id
# Short-circuit evaluation for || and &&.
if op_id == Id.Arith_DPipe:
lhs = self.EvalToInt(node.left)
if lhs == 0:
rhs = self.EvalToInt(node.right)
ret = int(rhs != 0)
else:
ret = 1 # true
return value.Int(ret)
if op_id == Id.Arith_DAmp:
lhs = self.EvalToInt(node.left)
if lhs == 0:
ret = 0 # false
else:
rhs = self.EvalToInt(node.right)
ret = int(rhs != 0)
return value.Int(ret)
if op_id == Id.Arith_LBracket:
# NOTE: Similar to bracket_op_e.ArrayIndex in osh/word_eval.py
left = self.Eval(node.left)
UP_left = left
with tagswitch(left) as case:
if case(value_e.MaybeStrArray):
left = cast(value__MaybeStrArray, UP_left)
rhs_int = self.EvalToInt(node.right)
try:
# could be None because representation is sparse
s = left.strs[rhs_int]
except IndexError:
s = None
elif case(value_e.AssocArray):
left = cast(value__AssocArray, UP_left)
key = self.EvalWordToString(node.right)
s = left.d.get(key)
else:
# TODO: Add error context
e_die(
'Expected array or assoc in index expression, got %s',
ui.ValType(left))
if s is None:
val = value.Undef() # type: value_t
else:
val = value.Str(s)
return val
if op_id == Id.Arith_Comma:
self.Eval(node.left) # throw away result
return self.Eval(node.right)
# Rest are integers
lhs = self.EvalToInt(node.left)
rhs = self.EvalToInt(node.right)
if op_id == Id.Arith_Plus:
ret = lhs + rhs
elif op_id == Id.Arith_Minus:
ret = lhs - rhs
elif op_id == Id.Arith_Star:
ret = lhs * rhs
elif op_id == Id.Arith_Slash:
if rhs == 0:
# TODO: Could also blame /
e_die('Divide by zero',
span_id=location.SpanForArithExpr(node.right))
ret = lhs / rhs
elif op_id == Id.Arith_Percent:
if rhs == 0:
# TODO: Could also blame /
e_die('Divide by zero',
span_id=location.SpanForArithExpr(node.right))
ret = lhs % rhs
elif op_id == Id.Arith_DStar:
# OVM is stripped of certain functions that are somehow necessary for
# exponentiation.
# Python/ovm_stub_pystrtod.c:21: PyOS_double_to_string: Assertion `0'
# failed.
if rhs < 0:
e_die("Exponent can't be less than zero"
) # TODO: error location
ret = 1
for i in xrange(rhs):
ret *= lhs
elif op_id == Id.Arith_DEqual:
ret = int(lhs == rhs)
elif op_id == Id.Arith_NEqual:
ret = int(lhs != rhs)
elif op_id == Id.Arith_Great:
ret = int(lhs > rhs)
elif op_id == Id.Arith_GreatEqual:
ret = int(lhs >= rhs)
elif op_id == Id.Arith_Less:
ret = int(lhs < rhs)
elif op_id == Id.Arith_LessEqual:
ret = int(lhs <= rhs)
elif op_id == Id.Arith_Pipe:
ret = lhs | rhs
elif op_id == Id.Arith_Amp:
ret = lhs & rhs
elif op_id == Id.Arith_Caret:
ret = lhs ^ rhs
# Note: how to define shift of negative numbers?
elif op_id == Id.Arith_DLess:
ret = lhs << rhs
elif op_id == Id.Arith_DGreat:
ret = lhs >> rhs
else:
raise AssertionError(op_id)
return value.Int(ret)
elif case(arith_expr_e.TernaryOp):
node = cast(arith_expr__TernaryOp, UP_node)
cond = self.EvalToInt(node.cond)
if cond: # nonzero
return self.Eval(node.true_expr)
else:
return self.Eval(node.false_expr)
else:
raise AssertionError(node.tag_())
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)
def GetVar(self, name, lookup_mode=scope_e.Dynamic):
assert isinstance(name, str), name
# TODO: Short-circuit down to _FindCellAndNamespace by doing a single hash
# lookup:
# COMPUTED_VARS = {'PIPESTATUS': 1, 'FUNCNAME': 1, ...}
# if name not in COMPUTED_VARS: ...
if name == 'PIPESTATUS':
return value.StrArray([str(i) for i in self.pipe_status[-1]])
# Do lookup of system globals before looking at user variables. Note: we
# could optimize this at compile-time like $?. That would break
# ${!varref}, but it's already broken for $?.
if name == 'FUNCNAME':
# bash wants it in reverse order. This is a little inefficient but we're
# not depending on deque().
strs = []
for func_name, source_name, _, _, _ in reversed(self.debug_stack):
if func_name:
strs.append(func_name)
if source_name:
strs.append('source') # bash doesn't give name
# Temp stacks are ignored
if self.has_main:
strs.append('main') # bash does this
return value.StrArray(strs) # TODO: Reuse this object too?
# This isn't the call source, it's the source of the function DEFINITION
# (or the sourced # file itself).
if name == 'BASH_SOURCE':
return value.StrArray(list(reversed(self.bash_source)))
# This is how bash source SHOULD be defined, but it's not!
if name == 'CALL_SOURCE':
strs = []
for func_name, source_name, call_spid, _, _ in reversed(self.debug_stack):
# should only happen for the first entry
if call_spid == const.NO_INTEGER:
continue
span = self.arena.GetLineSpan(call_spid)
source_str = self.arena.GetLineSourceString(span.line_id)
strs.append(source_str)
if self.has_main:
strs.append('-') # Bash does this to line up with main?
return value.StrArray(strs) # TODO: Reuse this object too?
if name == 'BASH_LINENO':
strs = []
for _, _, call_spid, _, _ in reversed(self.debug_stack):
# should only happen for the first entry
if call_spid == const.NO_INTEGER:
continue
span = self.arena.GetLineSpan(call_spid)
line_num = self.arena.GetLineNumber(span.line_id)
strs.append(str(line_num))
if self.has_main:
strs.append('0') # Bash does this to line up with main?
return value.StrArray(strs) # TODO: Reuse this object too?
if name == 'LINENO':
span = self.arena.GetLineSpan(self.current_spid)
# TODO: maybe use interned GetLineNumStr?
s = str(self.arena.GetLineNumber(span.line_id))
# Perf bug: why is this slow? Commenting it out reduces line count by
if 1:
self.line_num.s = s # Python's configure takes 75 seconds!
else:
# WTF this does not show the per bug?
self.line_num.s2 = s # Python's configure takes 13 seconds!
return self.line_num
# This is OSH-specific. Get rid of it in favor of ${BASH_SOURCE[0]} ?
if name == 'SOURCE_NAME':
# Update and reuse an object.
span = self.arena.GetLineSpan(self.current_spid)
self.source_name.s = self.arena.GetLineSourceString(span.line_id)
return self.source_name
cell, _ = self._FindCellAndNamespace(name, lookup_mode, writing=False)
if cell:
return cell.val
return value.Undef()
def SetVar(self, lval, val, new_flags, lookup_mode):
"""
Args:
lval: lvalue
val: value, or None if only changing flags
new_flags: tuple of flags to set: ReadOnly | Exported
() means no flags to start with
None means unchanged?
scope:
Local | Global | Dynamic - for builtins, PWD, etc.
NOTE: in bash, PWD=/ changes the directory. But not in dash.
"""
# STRICTNESS / SANENESS:
#
# 1) Don't create arrays automatically, e.g. a[1000]=x
# 2) Never change types? yeah I think that's a good idea, at least for oil
# (not sh, for compatibility). set -o strict-types or something. That
# means arrays have to be initialized with let arr = [], which is fine.
# This helps with stuff like IFS. It starts off as a string, and assigning
# it to a list is en error. I guess you will have to turn this no for
# bash?
#
# TODO:
# - COMPUTED_VARS can't be set
# - What about PWD / OLDPWD / UID / EUID ? You can simply make them
# readonly.
# - $PS1 and $PS4 should be PARSED when they are set, to avoid the error on use
# - Other validity: $HOME could be checked for existence
assert new_flags is not None
if lval.tag == lvalue_e.LhsName:
#if lval.name == 'ldflags':
# TODO: Turn this into a tracing feature. Like osh --tracevar ldflags
# --tracevar foo. Has to respect environment variables too.
if 0:
util.log('--- SETTING ldflags to %s', val)
if lval.spids:
span_id = lval.spids[0]
line_span = self.arena.GetLineSpan(span_id)
line_id = line_span.line_id
source_str = self.arena.GetLineSourceString(line_id)
line_num = self.arena.GetLineNumber(line_id)
#length = line_span.length
util.log('--- spid %s: %s, line %d, col %d', span_id, source_str,
line_num+1, line_span.col)
# TODO: Need the arena to look it up the line spid and line number.
cell, namespace = self._FindCellAndNamespace(lval.name, lookup_mode)
if cell:
if val is not None:
if cell.readonly:
# TODO: error context
e_die("Can't assign to readonly value %r", lval.name)
cell.val = val
if var_flags_e.Exported in new_flags:
cell.exported = True
if var_flags_e.ReadOnly in new_flags:
cell.readonly = True
if var_flags_e.AssocArray in new_flags:
cell.is_assoc_array = True
else:
if val is None:
# set -o nounset; local foo; echo $foo # It's still undefined!
val = value.Undef() # export foo, readonly foo
cell = runtime_asdl.cell(val,
var_flags_e.Exported in new_flags,
var_flags_e.ReadOnly in new_flags,
var_flags_e.AssocArray in new_flags)
namespace[lval.name] = cell
if (cell.val is not None and cell.val.tag == value_e.StrArray and
cell.exported):
e_die("Can't export array") # TODO: error context
elif lval.tag == lvalue_e.LhsIndexedName:
# TODO: All paths should have this? We can get here by a[x]=1 or
# (( a[ x ] = 1 )). Maybe we should make them different?
if lval.spids:
left_spid = lval.spids[0]
else:
left_spid = const.NO_INTEGER
# TODO: This is a parse error!
# a[1]=(1 2 3)
if val.tag == value_e.StrArray:
e_die("Can't assign array to array member", span_id=left_spid)
# bash/mksh have annoying behavior of letting you do LHS assignment to
# Undef, which then turns into an INDEXED array. (Undef means that set
# -o nounset fails.)
cell, namespace = self._FindCellAndNamespace(lval.name, lookup_mode)
if not cell:
self._BindNewArrayWithEntry(namespace, lval, val, new_flags)
return
cell_tag = cell.val.tag
if cell_tag == value_e.Str:
# s=x
# s[1]=y # invalid
e_die("Entries in value of type %s can't be assigned to",
cell.val.__class__.__name__, span_id=left_spid)
if cell.readonly:
e_die("Can't assign to readonly value", span_id=left_spid)
# This is for the case where we did declare -a foo or declare -A foo.
# There IS a cell, but it's still undefined.
if cell_tag == value_e.Undef:
if cell.is_assoc_array:
self._BindNewAssocArrayWithEntry(namespace, lval, val, new_flags)
else:
self._BindNewArrayWithEntry(namespace, lval, val, new_flags)
return
if cell_tag == value_e.StrArray:
strs = cell.val.strs
try:
strs[lval.index] = val.s
except IndexError:
# Fill it in with None. It could look like this:
# ['1', 2, 3, None, None, '4', None]
# Then ${#a[@]} counts the entries that are not None.
#
# TODO: strict-array for Oil arrays won't auto-fill.
n = lval.index - len(strs) + 1
strs.extend([None] * n)
strs[lval.index] = val.s
return
if cell_tag == value_e.AssocArray:
cell.val.d[lval.index] = val.s
return
else:
raise AssertionError(lval.__class__.__name__)
def GetArgNum(self, arg_num):
index = self.num_shifted + arg_num - 1
if index >= len(self.argv):
return value.Undef()
return value.Str(str(self.argv[index]))
def GetVar(self, name, lookup_mode=scope_e.Dynamic):
assert isinstance(name, str), name
# Do lookup of system globals before looking at user variables. Note: we
# could optimize this at compile-time like $?. That would break
# ${!varref}, but it's already broken for $?.
if name == 'FUNCNAME':
# bash wants it in reverse order. This is a little inefficient but we're
# not depending on deque().
strs = []
for func_name, source_name, _, _, _ in reversed(self.debug_stack):
if func_name:
strs.append(func_name)
if source_name:
strs.append('source') # bash doesn't give name
# Temp stacks are ignored
if self.has_main:
strs.append('main') # bash does this
return value.StrArray(strs) # TODO: Reuse this object too?
# This isn't the call source, it's the source of the function DEFINITION
# (or the sourced # file itself).
if name == 'BASH_SOURCE':
return value.StrArray(list(reversed(self.bash_source)))
# This is how bash source SHOULD be defined, but it's not!
if name == 'CALL_SOURCE':
strs = []
for func_name, source_name, call_spid, _, _ in reversed(
self.debug_stack):
# should only happen for the first entry
if call_spid == const.NO_INTEGER:
continue
span = self.arena.GetLineSpan(call_spid)
path, _ = self.arena.GetDebugInfo(span.line_id)
strs.append(path)
if self.has_main:
strs.append('-') # Bash does this to line up with main?
return value.StrArray(strs) # TODO: Reuse this object too?
if name == 'BASH_LINENO':
strs = []
for func_name, source_name, call_spid, _, _ in reversed(
self.debug_stack):
# should only happen for the first entry
if call_spid == const.NO_INTEGER:
continue
span = self.arena.GetLineSpan(call_spid)
_, line_num = self.arena.GetDebugInfo(span.line_id)
strs.append(str(line_num))
if self.has_main:
strs.append('0') # Bash does this to line up with main?
return value.StrArray(strs) # TODO: Reuse this object too?
if name == 'LINENO':
return self.line_num
# This is OSH-specific. Get rid of it in favor of ${BASH_SOURCE[0]} ?
if name == 'SOURCE_NAME':
return self.source_name
cell, _ = self._FindCellAndNamespace(name, lookup_mode, writing=False)
if cell:
return cell.val
return value.Undef()
def SetVar(self,
lval,
val,
flags_to_set,
lookup_mode,
flags_to_clear=(),
keyword_id=None):
"""
Args:
lval: lvalue
val: value, or None if only changing flags
flags_to_set: tuple of flags to set: ReadOnly | Exported
() means no flags to start with
scope:
Local | Global | Dynamic - for builtins, PWD, etc.
NOTE: in bash, PWD=/ changes the directory. But not in dash.
"""
# STRICTNESS / SANENESS:
#
# 1) Don't create arrays automatically, e.g. a[1000]=x
# 2) Never change types? yeah I think that's a good idea, at least for oil
# (not sh, for compatibility). set -o strict-types or something. That
# means arrays have to be initialized with let arr = [], which is fine.
# This helps with stuff like IFS. It starts off as a string, and assigning
# it to a list is an error. I guess you will have to turn this no for
# bash?
#
# TODO:
# - COMPUTED_VARS can't be set
# - What about PWD / OLDPWD / UID / EUID ? You can simply make them
# readonly.
# - $PS1 and $PS4 should be PARSED when they are set, to avoid the error on use
# - Other validity: $HOME could be checked for existence
assert flags_to_set is not None
if lval.tag == lvalue_e.Named:
cell, namespace = self._FindCellAndNamespace(
lval.name, lookup_mode)
self._CheckOilKeyword(keyword_id, lval, cell)
if cell:
# Clear before checking readonly bit.
# NOTE: Could be cell.flags &= flag_clear_mask
if var_flags_e.Exported in flags_to_clear:
cell.exported = False
if var_flags_e.ReadOnly in flags_to_clear:
cell.readonly = False
if val is not None: # e.g. declare -rx existing
if cell.readonly:
# TODO: error context
e_die("Can't assign to readonly value %r", lval.name)
cell.val = val
# NOTE: Could be cell.flags |= flag_set_mask
if var_flags_e.Exported in flags_to_set:
cell.exported = True
if var_flags_e.ReadOnly in flags_to_set:
cell.readonly = True
else:
if val is None: # declare -rx nonexistent
# set -o nounset; local foo; echo $foo # It's still undefined!
val = value.Undef() # export foo, readonly foo
cell = runtime_asdl.cell(val, var_flags_e.Exported
in flags_to_set, var_flags_e.ReadOnly
in flags_to_set)
namespace[lval.name] = cell
# Maintain invariant that only strings and undefined cells can be
# exported.
if (cell.val is not None
and cell.val.tag not in (value_e.Undef, value_e.Str)
and cell.exported):
e_die("Can't export array") # TODO: error context
elif lval.tag == lvalue_e.Indexed:
# There is no syntax 'declare a[x]'
assert val is not None, val
# TODO: All paths should have this? We can get here by a[x]=1 or
# (( a[ x ] = 1 )). Maybe we should make them different?
left_spid = lval.spids[0] if lval.spids else const.NO_INTEGER
# bash/mksh have annoying behavior of letting you do LHS assignment to
# Undef, which then turns into an INDEXED array. (Undef means that set
# -o nounset fails.)
cell, namespace = self._FindCellAndNamespace(
lval.name, lookup_mode)
self._CheckOilKeyword(keyword_id, lval, cell)
if not cell:
self._BindNewArrayWithEntry(namespace, lval, val, flags_to_set)
return
if cell.readonly:
e_die("Can't assign to readonly array", span_id=left_spid)
cell_tag = cell.val.tag
# undef[0]=y is allowed
if cell_tag == value_e.Undef:
self._BindNewArrayWithEntry(namespace, lval, val, flags_to_set)
return
if cell_tag == value_e.Str:
# s=x
# s[1]=y # invalid
e_die("Entries in value of type %s can't be assigned to",
cell.val.__class__.__name__,
span_id=left_spid)
if cell_tag == value_e.MaybeStrArray:
strs = cell.val.strs
try:
strs[lval.index] = val.s
except IndexError:
# Fill it in with None. It could look like this:
# ['1', 2, 3, None, None, '4', None]
# Then ${#a[@]} counts the entries that are not None.
#
# TODO: strict-array for Oil arrays won't auto-fill.
n = lval.index - len(strs) + 1
strs.extend([None] * n)
strs[lval.index] = val.s
return
# AssocArray shouldn't happen because we query IsAssocArray before
# evaluating lhs_expr.
e_die("Object of this type can't be indexed: %s", cell.val)
elif lval.tag == lvalue_e.Keyed:
# There is no syntax 'declare A["x"]'
assert val is not None, val
left_spid = lval.spids[0] if lval.spids else const.NO_INTEGER
cell, namespace = self._FindCellAndNamespace(
lval.name, lookup_mode)
self._CheckOilKeyword(keyword_id, lval, cell)
# We already looked it up before making the lvalue
assert cell.val.tag == value_e.AssocArray, cell
if cell.readonly:
e_die("Can't assign to readonly associative array",
span_id=left_spid)
cell.val.d[lval.key] = val.s
else:
raise AssertionError(lval.__class__.__name__)
