def Echo(argv):
"""
echo builtin. Doesn't depend on executor state.
TODO: Where to put help? docstring?
"""
# NOTE: both getopt and optparse are unsuitable for 'echo' because:
# - 'echo -c' should print '-c', not fail
# - echo '---' should print ---, not fail
arg, i = echo_spec.ParseLikeEcho(argv)
if arg.e:
util.warn('*** echo -e not implemented ***')
#log('echo argv %s', argv)
n = len(argv)
for i in xrange(i, n - 1):
sys.stdout.write(argv[i])
sys.stdout.write(' ') # arg separator
if argv:
sys.stdout.write(argv[-1])
if not arg.n:
sys.stdout.write('\n')
# Do I need the flush? Had a problem here under load, but it might not have
# been because of that.
# File "/home/andy/git/oil/bin/../core/cmd_exec.py", line 251, in _RunBuiltin
# status = builtin.Echo(argv)
# File "/home/andy/git/oil/bin/../core/builtin.py", line 431, in Echo
# sys.stdout.flush()
# IOError: [Errno 32] Broken pipe
sys.stdout.flush()
return 0
def _ValToArithOrError(self,
val,
int_coerce=True,
blame_word=None,
span_id=const.NO_INTEGER):
if span_id == const.NO_INTEGER and blame_word:
span_id = word.LeftMostSpanForWord(blame_word)
#log('_ValToArithOrError span=%s blame=%s', span_id, blame_word)
try:
i = self._ValToArith(val, span_id, int_coerce=int_coerce)
except util.FatalRuntimeError as e:
if self.exec_opts.strict_arith:
raise
else:
i = 0
span_id = word.SpanIdFromError(e)
if self.arena: # BoolEvaluator for test builtin doesn't have it.
if span_id != const.NO_INTEGER:
ui.PrintFilenameAndLine(span_id, self.arena)
else:
log('*** Warning has no location info ***')
warn(e.UserErrorString())
return i
def Wait(self):
# This is a list of async jobs
try:
pid, status = os.wait()
except OSError as e:
if e.errno == errno.ECHILD:
#log('WAIT ECHILD')
return False # nothing to wait for caller should stop
else:
# What else can go wrong?
raise
#log('WAIT got %s %s', pid, status)
# TODO: change status in more cases.
if os.WIFSIGNALED(status):
pass
elif os.WIFEXITED(status):
status = os.WEXITSTATUS(status)
#log('exit status: %s', status)
# This could happen via coding error. But this may legitimately happen
# if a grandchild outlives the child (its parent). Then it is reparented
# under this process, so we might receive notification of its exit, even
# though we didn't start it. We can't have any knowledge of such
# processes, so print a warning.
if pid not in self.callbacks:
util.warn("PID %d stopped, but osh didn't start it", pid)
return True # caller should keep waiting
callback = self.callbacks.pop(pid)
callback(pid, status)
self.last_status = status # for wait -n
return True # caller should keep waiting
def _EvalRedirect(self, n):
fd = REDIR_DEFAULT_FD[n.op_id] if n.fd == const.NO_INTEGER else n.fd
if n.tag == redir_e.Redir:
redir_type = REDIR_ARG_TYPES[n.op_id] # could be static in the LST?
if redir_type == redir_arg_type_e.Path:
# NOTE: no globbing. You can write to a file called '*.py'.
val = self.word_ev.EvalWordToString(n.arg_word)
if val.tag != value_e.Str: # TODO: This error never fires
util.error("Redirect filename must be a string, got %s", val)
return None
filename = val.s
if not filename:
# Whether this is fatal depends on errexit.
util.error("Redirect filename can't be empty")
return None
return runtime.PathRedirect(n.op_id, fd, filename)
elif redir_type == redir_arg_type_e.Desc: # e.g. 1>&2
val = self.word_ev.EvalWordToString(n.arg_word)
if val.tag != value_e.Str: # TODO: This error never fires
util.error("Redirect descriptor should be a string, got %s", val)
return None
t = val.s
if not t:
util.error("Redirect descriptor can't be empty")
return None
try:
target_fd = int(t)
except ValueError:
util.error(
"Redirect descriptor should look like an integer, got %s", val)
return None
return runtime.DescRedirect(n.op_id, fd, target_fd)
elif redir_type == redir_arg_type_e.Here: # here word
# TODO: decay should be controlled by an option
val = self.word_ev.EvalWordToString(n.arg_word, decay=True)
if val.tag != value_e.Str: # TODO: This error never fires
util.warn("Here word body should be a string, got %s", val)
return None
# NOTE: bash and mksh both add \n
return runtime.HereRedirect(fd, val.s + '\n')
else:
raise AssertionError('Unknown redirect op')
elif n.tag == redir_e.HereDoc:
# TODO: decay shoudl be controlled by an option
val = self.word_ev.EvalWordToString(n.body, decay=True)
if val.tag != value_e.Str: # TODO: This error never fires
util.warn("Here doc body should be a string, got %s", val)
return None
return runtime.HereRedirect(fd, val.s)
else:
raise AssertionError('Unknown redirect type')
def Trap(argv, traps):
# TODO: register trap
# Example:
# trap -- 'echo "hi there" | wc ' SIGINT
#
# Then hit Ctrl-C.
#
# Yeah you need the EvalHelper. traps is a list of signals to parsed
# NODES.
util.warn('*** trap not implemented ***')
return 0
def Command(argv, funcs, path_val):
arg, i = COMMAND_SPEC.Parse(argv)
status = 0
if arg.v:
for kind, arg in _ResolveNames(argv[i:], funcs, path_val):
if kind is None:
status = 1 # nothing printed, but we fail
else:
# This is for -v, -V is more detailed.
print(arg)
else:
util.warn('*** command without -v not not implemented ***')
status = 1
return status
def EvalCStringToken(id_, value):
"""
This function is shared between echo -e and $''.
$'' could use it at compile time, much like brace expansion in braces.py.
"""
if id_ == Id.Char_Literals:
return value
elif id_ == Id.Char_BadBackslash:
if 1: # TODO: error in strict mode
# Either \A or trailing \ (A is not a valid backslash escape)
util.warn('Invalid backslash escape')
return value
elif id_ == Id.Char_OneChar:
c = value[1]
return _ONE_CHAR[c]
elif id_ == Id.Char_Stop: # \c returns a special sentinel
return None
elif id_ in (Id.Char_Octal3, Id.Char_Octal4):
if id_ == Id.Char_Octal3: # $'\377'
s = value[1:]
else: # echo -e '\0377'
s = value[2:]
i = int(s, 8)
if i >= 256:
i = i % 256
# NOTE: This is for strict mode
#raise AssertionError('Out of range')
return chr(i)
elif id_ == Id.Char_Hex:
s = value[2:]
i = int(s, 16)
return chr(i)
elif id_ in (Id.Char_Unicode4, Id.Char_Unicode8):
s = value[2:]
i = int(s, 16)
#util.log('i = %d', i)
return libstr.Utf8Encode(i)
else:
raise AssertionError
def Read(argv, mem):
# TODO:
# - Use IFS instead of Python's split().
arg, i = READ_SPEC.Parse(argv)
if not arg.r:
util.warn('*** read without -r not implemented ***')
names = argv[i:]
if arg.n is not None:
try:
name = names[0]
except IndexError:
name = 'REPLY' # default variable name
s = os.read(sys.stdin.fileno(), arg.n)
#log('read -n: %s = %s', name, s)
state.SetLocalString(mem, name, s)
# NOTE: Even if we don't get n bytes back, there is no error?
return 0
line = sys.stdin.readline()
if not line: # EOF
return 1
if line.endswith('\n'): # strip trailing newline
line = line[:-1]
status = 0
else:
# odd bash behavior: fail even if we can set variables.
status = 1
# leftover words assigned to the last name
n = len(names)
strs = line.split(None, n - 1)
# TODO: Use REPLY variable here too?
for i in xrange(n):
try:
s = strs[i]
except IndexError:
s = '' # if there are too many variables
#log('read: %s = %s', names[i], s)
state.SetLocalString(mem, names[i], s)
return status
def _StringToIntegerOrError(self, s, blame_word=None,
span_id=const.NO_INTEGER):
"""Used by both [[ $x -gt 3 ]] and (( $x ))."""
if span_id == const.NO_INTEGER and blame_word:
span_id = word.LeftMostSpanForWord(blame_word)
try:
i = _StringToInteger(s, span_id=span_id)
except util.FatalRuntimeError as e:
if self.exec_opts.strict_arith:
raise
else:
i = 0
# TODO: Need the arena for printing this error?
#ui.PrettyPrintError(e)
warn(e.UserErrorString())
return i
def Wait(self):
# This is a list of async jobs
while True:
try:
pid, status = posix.wait()
except OSError as e:
#log('wait() error: %s', e)
if e.errno == errno.ECHILD:
return False # nothing to wait for caller should stop
elif e.errno == errno.EINTR:
# This happens when we register a handler for SIGINT, and thus never
# get the KeyboardInterrupt exception? Not sure why.
# Try
# $ cat # Now hit Ctrl-C
#log('Continuing')
continue # try again
else:
# An error we don't know about.
raise
else:
break # no exception thrown, so no need to retry
#log('WAIT got %s %s', pid, status)
# TODO: change status in more cases.
if posix.WIFSIGNALED(status):
if posix.WTERMSIG(status) == signal.SIGINT:
print()
elif posix.WIFEXITED(status):
status = posix.WEXITSTATUS(status)
#log('exit status: %s', status)
# This could happen via coding error. But this may legitimately happen
# if a grandchild outlives the child (its parent). Then it is reparented
# under this process, so we might receive notification of its exit, even
# though we didn't start it. We can't have any knowledge of such
# processes, so print a warning.
if pid not in self.callbacks:
util.warn("PID %d stopped, but osh didn't start it", pid)
return True # caller should keep waiting
callback = self.callbacks.pop(pid)
callback(pid, status)
self.last_status = status # for wait -n
return True # caller should keep waiting
def Type(argv, funcs, path_val):
arg, i = TYPE_SPEC.Parse(argv)
if path_val.tag == value_e.Str:
path_list = path_val.s.split(':')
else:
path_list = [] # treat as empty path
status = 0
if not arg.t:
util.warn("*** 'type' builtin called without -t ***")
status = 1
# Keep going anyway
for name in argv[i:]:
if name in funcs:
print('function')
elif Resolve(name) != EBuiltin.NONE:
print('builtin')
elif ResolveSpecial(name) != EBuiltin.NONE:
print('builtin')
elif lex.IsOtherBuiltin(name): # declare, continue, etc.
print('builtin')
elif lex.IsKeyword(name):
print('keyword')
else:
# Now look for files.
found = False
for path_dir in path_list:
full_path = os.path.join(path_dir, name)
if os.path.exists(full_path):
print('file')
found = True
break
if not found: # Nothing printed, but status is 1.
status = 1
return status
def Umask(argv):
if len(argv) == 0:
# umask() has a dumb API: you can't get it without modifying it first!
# NOTE: dash disables interrupts around the two umask() calls, but that
# shouldn't be a concern for us. Signal handlers won't call umask().
mask = posix.umask(0)
posix.umask(mask) #
print('0%03o' % mask) # octal format
return 0
if len(argv) == 1:
a = argv[0]
try:
new_mask = int(a, 8)
except ValueError:
# NOTE: This happens if we have '8' or '9' in the input too.
util.warn('*** umask with symbolic input not implemented ***')
return 1
else:
posix.umask(new_mask)
return 0
raise args.UsageError('umask: unexpected arguments')
def _Dispatch(self, node, fork_external):
# If we call RunCommandSub in a recursive call to the executor, this will
# be set true (if strict-errexit is false). But it only lasts for one
# command.
self.check_command_sub_status = False
#argv0 = None # for error message
check_errexit = False # for errexit
if node.tag == command_e.SimpleCommand:
check_errexit = True
# Find span_id for a basic implementation of $LINENO, e.g.
# PS4='+$SOURCE_NAME:$LINENO:'
# NOTE: osh2oil uses node.more_env, but we don't need that.
span_id = const.NO_INTEGER
if node.words:
first_word = node.words[0]
span_id = word.LeftMostSpanForWord(first_word)
self.mem.SetCurrentSpanId(span_id)
# PROBLEM: We want to log argv in 'xtrace' mode, but we may have already
# redirected here, which screws up logging. For example, 'echo hi
# >/dev/null 2>&1'. We want to evaluate argv and log it BEFORE applying
# redirects.
# Another problem:
# - tracing can be called concurrently from multiple processes, leading
# to overlap. Maybe have a mode that creates a file per process.
# xtrace-proc
# - line numbers for every command would be very nice. But then you have
# to print the filename too.
words = braces.BraceExpandWords(node.words)
argv = self.word_ev.EvalWordSequence(words)
# This comes before evaluating env, in case there are problems evaluating
# it. We could trace the env separately? Also trace unevaluated code
# with set-o verbose?
self.tracer.OnSimpleCommand(argv)
if node.more_env:
self.mem.PushTemp()
try:
for env_pair in node.more_env:
val = self.word_ev.EvalWordToString(env_pair.val)
# Set each var so the next one can reference it. Example:
# FOO=1 BAR=$FOO ls /
self.mem.SetVar(ast.LhsName(env_pair.name), val,
(var_flags_e.Exported,), scope_e.TempEnv)
# NOTE: This might never return! In the case of fork_external=False.
status = self._RunSimpleCommand(argv, fork_external, span_id)
finally:
if node.more_env:
self.mem.PopTemp()
elif node.tag == command_e.Sentence:
# Don't check_errexit since this isn't a real node!
if node.terminator.id == Id.Op_Semi:
status = self._Execute(node.child)
else:
status = self._RunJobInBackground(node.child)
elif node.tag == command_e.Pipeline:
check_errexit = True
if node.stderr_indices:
raise NotImplementedError('|&')
if node.negated:
self._PushErrExit()
try:
status2 = self._RunPipeline(node)
finally:
self._PopErrExit()
# errexit is disabled for !.
check_errexit = False
status = 1 if status2 == 0 else 0
else:
status = self._RunPipeline(node)
elif node.tag == command_e.Subshell:
check_errexit = True
# This makes sure we don't waste a process if we'd launch one anyway.
p = self._MakeProcess(node.child)
status = p.Run(self.waiter)
elif node.tag == command_e.DBracket:
check_errexit = True
result = self.bool_ev.Eval(node.expr)
status = 0 if result else 1
elif node.tag == command_e.DParen:
check_errexit = True
i = self.arith_ev.Eval(node.child)
status = 0 if i != 0 else 1
elif node.tag == command_e.Assignment:
flags = word_compile.ParseAssignFlags(node.flags)
if node.keyword == Id.Assign_Local:
lookup_mode = scope_e.LocalOnly
# typeset and declare are synonyms? I see typeset -a a=() the most.
elif node.keyword in (Id.Assign_Declare, Id.Assign_Typeset):
# declare is like local, except it can also be used outside functions?
if var_flags_e.Global in flags:
lookup_mode = scope_e.GlobalOnly
else:
lookup_mode = scope_e.LocalOnly
elif node.keyword == Id.Assign_Readonly:
lookup_mode = scope_e.Dynamic
flags.append(var_flags_e.ReadOnly)
elif node.keyword == Id.Assign_None: # mutate existing local or global
lookup_mode = scope_e.Dynamic
else:
raise AssertionError(node.keyword)
for pair in node.pairs:
if pair.op == assign_op_e.PlusEqual:
assert pair.rhs, pair.rhs # I don't think a+= is valid?
val = self.word_ev.EvalRhsWord(pair.rhs)
old_val, lval = expr_eval.EvalLhsAndLookup(pair.lhs, self.arith_ev,
self.mem, self.exec_opts)
sig = (old_val.tag, val.tag)
if sig == (value_e.Undef, value_e.Str):
pass # val is RHS
elif sig == (value_e.Undef, value_e.StrArray):
pass # val is RHS
elif sig == (value_e.Str, value_e.Str):
val = runtime.Str(old_val.s + val.s)
elif sig == (value_e.Str, value_e.StrArray):
e_die("Can't append array to string")
elif sig == (value_e.StrArray, value_e.Str):
e_die("Can't append string to array")
elif sig == (value_e.StrArray, value_e.StrArray):
val = runtime.StrArray(old_val.strs + val.strs)
else: # plain assignment
spid = pair.spids[0] # Source location for tracing
lval = self._EvalLhs(pair.lhs, spid, lookup_mode)
# RHS can be a string or array.
if pair.rhs:
val = self.word_ev.EvalRhsWord(pair.rhs)
assert isinstance(val, runtime.value), val
else: # e.g. 'readonly x' or 'local x'
val = None
# NOTE: In bash and mksh, declare -a myarray makes an empty cell with
# Undef value, but the 'array' attribute.
#log('setting %s to %s with flags %s', lval, val, flags)
self.mem.SetVar(lval, val, flags, lookup_mode,
strict_array=self.exec_opts.strict_array)
# Assignment always appears to have a spid.
if node.spids:
current_spid = node.spids[0]
else:
current_spid = const.NO_INTEGER
self.mem.SetCurrentSpanId(current_spid)
self.tracer.OnAssignment(lval, pair.op, val, flags, lookup_mode)
# PATCH to be compatible with existing shells: If the assignment had a
# command sub like:
#
# s=$(echo one; false)
#
# then its status will be in mem.last_status, and we can check it here.
# If there was NOT a command sub in the assignment, then we don't want to
# check it.
if node.keyword == Id.Assign_None: # mutate existing local or global
# Only do this if there was a command sub? How? Look at node?
# Set a flag in mem? self.mem.last_status or
if self.check_command_sub_status:
self._CheckStatus(self.mem.last_status, node)
# A global assignment shouldn't clear $?.
status = self.mem.last_status
else:
status = 0
else:
# To be compatible with existing shells, local assignments DO clear
# $?. Even in strict mode, we don't need to bother setting
# check_errexit = True, because we would have already checked the
# command sub in RunCommandSub.
status = 0
# TODO: maybe we should have a "sane-status" that respects this:
# false; echo $?; local f=x; echo $?
elif node.tag == command_e.ControlFlow:
if node.arg_word: # Evaluate the argument
val = self.word_ev.EvalWordToString(node.arg_word)
assert val.tag == value_e.Str
arg = int(val.s) # They all take integers
else:
arg = 0 # return 0, exit 0, break 0 levels, etc.
# NOTE: We don't do anything about a top-level 'return' here. Unlike in
# bash, that is OK. If you can return from a sourced script, it makes
# sense to return from a main script.
ok = True
tok = node.token
if (tok.id in (Id.ControlFlow_Break, Id.ControlFlow_Continue) and
self.loop_level == 0):
ok = False
msg = 'Invalid control flow at top level'
if ok:
raise _ControlFlow(tok, arg)
if self.exec_opts.strict_control_flow:
e_die(msg, token=tok)
else:
# Only print warnings, never fatal.
# Bash oddly only exits 1 for 'return', but no other shell does.
ui.PrintFilenameAndLine(tok.span_id, self.arena)
util.warn(msg)
status = 0
# The only difference between these two is that CommandList has no
# redirects. We already took care of that above.
elif node.tag in (command_e.CommandList, command_e.BraceGroup):
status = self._ExecuteList(node.children)
check_errexit = False
elif node.tag == command_e.AndOr:
# NOTE: && and || have EQUAL precedence in command mode. See case #13
# in dbracket.test.sh.
left = node.children[0]
# Suppress failure for every child except the last one.
self._PushErrExit()
try:
status = self._Execute(left)
finally:
self._PopErrExit()
i = 1
n = len(node.children)
while i < n:
#log('i %d status %d', i, status)
child = node.children[i]
op_id = node.ops[i-1]
#log('child %s op_id %s', child, op_id)
if op_id == Id.Op_DPipe and status == 0:
i += 1
continue # short circuit
elif op_id == Id.Op_DAmp and status != 0:
i += 1
continue # short circuit
if i == n - 1: # errexit handled differently for last child
status = self._Execute(child)
check_errexit = True
else:
self._PushErrExit()
try:
status = self._Execute(child)
finally:
self._PopErrExit()
i += 1
elif node.tag == command_e.WhileUntil:
if node.keyword.id == Id.KW_While:
_DonePredicate = lambda status: status != 0
else:
_DonePredicate = lambda status: status == 0
status = 0
self.loop_level += 1
try:
while True:
self._PushErrExit()
try:
cond_status = self._ExecuteList(node.cond)
finally:
self._PopErrExit()
done = cond_status != 0
if _DonePredicate(cond_status):
break
try:
status = self._Execute(node.body) # last one wins
except _ControlFlow as e:
if e.IsBreak():
status = 0
break
elif e.IsContinue():
status = 0
continue
else: # return needs to pop up more
raise
finally:
self.loop_level -= 1
elif node.tag == command_e.ForEach:
iter_name = node.iter_name
if node.do_arg_iter:
iter_list = self.mem.GetArgv()
else:
words = braces.BraceExpandWords(node.iter_words)
iter_list = self.word_ev.EvalWordSequence(words)
# We need word splitting and so forth
# NOTE: This expands globs too. TODO: We should pass in a Globber()
# object.
status = 0 # in case we don't loop
self.loop_level += 1
try:
for x in iter_list:
#log('> ForEach setting %r', x)
state.SetLocalString(self.mem, iter_name, x)
#log('<')
try:
status = self._Execute(node.body) # last one wins
except _ControlFlow as e:
if e.IsBreak():
status = 0
break
elif e.IsContinue():
status = 0
else: # return needs to pop up more
raise
finally:
self.loop_level -= 1
elif node.tag == command_e.ForExpr:
status = 0
init, cond, body, update = node.init, node.cond, node.body, node.update
if init:
self.arith_ev.Eval(init)
self.loop_level += 1
try:
while True:
if cond:
b = self.arith_ev.Eval(cond)
if not b:
break
try:
status = self._Execute(body)
except _ControlFlow as e:
if e.IsBreak():
status = 0
break
elif e.IsContinue():
status = 0
else: # return needs to pop up more
raise
if update:
self.arith_ev.Eval(update)
finally:
self.loop_level -= 1
elif node.tag == command_e.DoGroup:
status = self._ExecuteList(node.children)
check_errexit = False # not real statements
elif node.tag == command_e.FuncDef:
# NOTE: Would it make sense to evaluate the redirects BEFORE entering?
# It will save time on function calls.
self.funcs[node.name] = node
status = 0
elif node.tag == command_e.If:
done = False
for arm in node.arms:
self._PushErrExit()
try:
status = self._ExecuteList(arm.cond)
finally:
self._PopErrExit()
if status == 0:
status = self._ExecuteList(arm.action)
done = True
break
# TODO: The compiler should flatten this
if not done and node.else_action is not None:
status = self._ExecuteList(node.else_action)
elif node.tag == command_e.NoOp:
status = 0 # make it true
elif node.tag == command_e.Case:
val = self.word_ev.EvalWordToString(node.to_match)
to_match = val.s
status = 0 # If there are no arms, it should be zero?
done = False
for arm in node.arms:
for pat_word in arm.pat_list:
# NOTE: Is it OK that we're evaluating these as we go?
# TODO: case "$@") shouldn't succeed? That's a type error?
# That requires strict-array?
pat_val = self.word_ev.EvalWordToString(pat_word, do_fnmatch=True)
#log('Matching word %r against pattern %r', to_match, pat_val.s)
if libc.fnmatch(pat_val.s, to_match):
status = self._ExecuteList(arm.action)
done = True # TODO: Parse ;;& and for fallthrough and such?
break # Only execute action ONCE
if done:
break
elif node.tag == command_e.TimeBlock:
# TODO:
# - When do we need RUSAGE_CHILDREN?
# - Respect TIMEFORMAT environment variable.
# "If this variable is not set, Bash acts as if it had the value"
# $'\nreal\t%3lR\nuser\t%3lU\nsys\t%3lS'
# "A trailing newline is added when the format string is displayed."
start_t = time.time() # calls gettimeofday() under the hood
start_u = resource.getrusage(resource.RUSAGE_SELF)
status = self._Execute(node.pipeline)
end_t = time.time()
end_u = resource.getrusage(resource.RUSAGE_SELF)
real = end_t - start_t
user = end_u.ru_utime - start_u.ru_utime
sys_ = end_u.ru_stime - start_u.ru_stime
libc.print_time(real, user, sys_)
else:
raise NotImplementedError(node.__class__.__name__)
return status, check_errexit
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 _ApplyPrefixOp(self, val, op_id, token):
"""
Returns:
value
"""
assert val.tag != value_e.Undef
if op_id == Id.VSub_Pound: # LENGTH
if val.tag == value_e.Str:
# NOTE: Whether bash counts bytes or chars is affected by LANG
# environment variables.
# Should we respect that, or another way to select? set -o
# count-bytes?
# https://stackoverflow.com/questions/17368067/length-of-string-in-bash
try:
length = string_ops.CountUtf8Chars(val.s)
except util.InvalidUtf8 as e:
# TODO: Add location info from 'part'? Only the caller has it.
if self.exec_opts.strict_word_eval:
raise
else:
# NOTE: Doesn't make the command exit with 1; it just returns a
# length of -1.
util.warn(e.UserErrorString())
return value.Str('-1')
elif val.tag == value_e.StrArray:
# There can be empty placeholder values in the array.
length = sum(1 for s in val.strs if s is not None)
return value.Str(str(length))
elif op_id == Id.VSub_Bang: # ${!foo}, "indirect expansion"
# NOTES:
# - Could translate to eval('$' + name) or eval("\$$name")
# - ${!array[@]} means something completely different. TODO: implement
# that.
# - It might make sense to suggest implementing this with associative
# arrays?
if val.tag == value_e.Str:
# plain variable name, like 'foo'
if match.IsValidVarName(val.s):
return self.mem.GetVar(val.s)
# positional argument, like '1'
try:
return self.mem.GetArgNum(int(val.s))
except ValueError:
pass
if val.s in ('@', '*'):
# TODO maybe_decay_array
return value.StrArray(self.mem.GetArgv())
# otherwise an array reference, like 'arr[0]' or 'arr[xyz]' or 'arr[@]'
i = val.s.find('[')
if i >= 0 and val.s[-1] == ']':
name, index = val.s[:i], val.s[i+1:-1]
result = self._EvalIndirectArrayExpansion(name, index)
if result is not None:
return result
# Note that bash doesn't consider this fatal. It makes the
# command exit with '1', but we don't have that ability yet?
e_die('Bad indirect expansion: %r', val.s, token=token)
elif val.tag == value_e.StrArray:
indices = [str(i) for i, s in enumerate(val.strs) if s is not None]
return value.StrArray(indices)
else:
raise AssertionError
else:
raise AssertionError(op_id)
def Trap(argv, traps, nodes_to_run, ex):
arg, i = TRAP_SPEC.Parse(argv)
if arg.p: # Print registered handlers
for name, value in traps.iteritems():
# The unit tests rely on this being one line.
# bash prints a line that can be re-parsed.
print('%s %s' % (name, value.__class__.__name__))
sys.stdout.flush()
return 0
if arg.l: # List valid signals and hooks
ordered = _SIGNAL_NAMES.items()
ordered.sort(key=lambda x: x[1])
for name in _HOOK_NAMES:
print(' %s' % name)
for name, int_val in ordered:
print('%2d %s' % (int_val, name))
sys.stdout.flush()
return 0
try:
code_str = argv[0]
sig_spec = argv[1]
except IndexError:
raise args.UsageError('trap CODE SIGNAL_SPEC')
# sig_key is NORMALIZED sig_spec: and integer signal number or string hook
# name.
sig_key = None
sig_num = None
if sig_spec in _HOOK_NAMES:
sig_key = sig_spec
elif sig_spec == '0': # Special case
sig_key = 'EXIT'
else:
sig_num = _GetSignalNumber(sig_spec)
if sig_num is not None:
sig_key = sig_num
if sig_key is None:
util.error("Invalid signal or hook %r" % sig_spec)
return 1
# NOTE: sig_spec isn't validated when removing handlers.
if code_str == '-':
if sig_key in _HOOK_NAMES:
try:
del traps[sig_key]
except KeyError:
pass
return 0
if sig_num is not None:
try:
del traps[sig_key]
except KeyError:
pass
# Restore default
if sig_num == signal.SIGINT:
RegisterSigIntHandler()
else:
signal.signal(sig_num, signal.SIG_DFL)
return 0
raise AssertionError('Signal or trap')
# Try parsing the code first.
node = ex.ParseTrapCode(code_str)
if node is None:
return 1 # ParseTrapCode() prints an error for us.
# Register a hook.
if sig_key in _HOOK_NAMES:
if sig_key in ('ERR', 'RETURN', 'DEBUG'):
util.warn("*** The %r isn't yet implemented in OSH ***", sig_spec)
traps[sig_key] = _TrapHandler(node, nodes_to_run)
return 0
# Register a signal.
sig_num = _GetSignalNumber(sig_spec)
if sig_num is not None:
handler = _TrapHandler(node, nodes_to_run)
# For signal handlers, the traps dictionary is used only for debugging.
traps[sig_key] = handler
signal.signal(sig_num, handler)
return 0
raise AssertionError('Signal or trap')
def _ApplyPrefixOp(self, val, op_id):
"""
Returns:
value
"""
assert val.tag != value_e.Undef
if op_id == Id.VSub_Pound: # LENGTH
if val.tag == value_e.Str:
# NOTE: Whether bash counts bytes or chars is affected by LANG
# environment variables.
# Should we respect that, or another way to select? set -o
# count-bytes?
# https://stackoverflow.com/questions/17368067/length-of-string-in-bash
try:
length = libstr.CountUtf8Chars(val.s)
except util.InvalidUtf8 as e:
# TODO: Add location info from 'part'? Only the caller has it.
if self.exec_opts.strict_word_eval:
raise
else:
# NOTE: Doesn't make the command exit with 1; it just returns a
# length of -1.
util.warn(e.UserErrorString())
return runtime.Str('-1')
elif val.tag == value_e.StrArray:
# There can be empty placeholder values in the array.
length = sum(1 for s in val.strs if s is not None)
return runtime.Str(str(length))
elif op_id == Id.VSub_Bang:
# NOTES:
# - Could translate to eval('$' + name) or eval("\$$name")
# - ${!array[@]} means something completely different. TODO: implement
# that.
# - It might make sense to suggest implementing this with associative
# arrays?
# Treat the value of the variable as a variable name.
if val.tag == value_e.Str:
try:
# e.g. ${!OPTIND} gives $1 when OPTIND is 1
arg_num = int(val.s)
return self.mem.GetArgNum(arg_num)
except ValueError:
if not match.IsValidVarName(val.s):
# TODO: location information.
# Also note that bash doesn't consider this fatal. It makes the
# command exit with '1', but we don't have that ability yet?
e_die('Bad variable name %r in var ref', val.s)
return self.mem.GetVar(val.s)
elif val.tag == value_e.StrArray:
raise NotImplementedError(
'${!a[@]}') # bash gets keys this way
else:
raise AssertionError
else:
raise AssertionError(op_id)
def Trap(argv, traps, nodes_to_run, ex):
arg, i = TRAP_SPEC.Parse(argv)
if arg.p: # Print registered handlers
for name, value in traps.iteritems():
print(name)
print(value)
print()
sys.stdout.flush()
return 0
if arg.l: # List valid signals and hooks
ordered = _SIGNAL_NAMES.items()
ordered.sort(key=lambda x: x[1])
for name in _HOOK_NAMES:
print(' %s' % name)
for name, int_val in ordered:
print('%2d %s' % (int_val, name))
sys.stdout.flush()
return 0
try:
code_str = argv[0]
sig_spec = argv[1]
except IndexError:
raise args.UsageError('trap CODE SIGNAL_SPEC')
# NOTE: sig_spec isn't validated when removing handlers.
if code_str == '-':
if sig_spec in _HOOK_NAMES:
try:
del traps[sig_spec]
except KeyError:
pass
return 0
sig_val = _GetSignalValue(sig_spec)
if sig_val is not None:
try:
del traps[sig_spec]
except KeyError:
pass
# Restore default
if sig_val == signal.SIGINT:
RegisterSigIntHandler()
else:
signal.signal(sig_val, signal.SIG_DFL)
return 0
util.error("Can't remove invalid trap %r" % sig_spec)
return 1
# Try parsing the code first.
node = ex.ParseTrapCode(code_str)
if node is None:
return 1 # ParseTrapCode() prints an error for us.
# Register a hook.
if sig_spec in _HOOK_NAMES:
if sig_spec in ('ERR', 'RETURN', 'DEBUG'):
util.warn("*** The %r isn't yet implemented in OSH ***", sig_spec)
traps[sig_spec] = _TrapHandler(node, nodes_to_run)
return 0
# Register a signal.
sig_val = _GetSignalValue(sig_spec)
if sig_val is not None:
handler = _TrapHandler(node, nodes_to_run)
# For signal handlers, the traps dictionary is used only for debugging.
traps[sig_spec] = handler
signal.signal(sig_val, handler)
return 0
util.error('Invalid trap %r' % sig_spec)
return 1
