def ParseSubshell(self):
left_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next() # skip past (
# Ensure that something $( (cd / && pwd) ) works. If ) is already on the
# translation stack, we want to delay it.
#print('ParseSubshell lexer.PushHint ) -> )')
self.lexer.PushHint(Id.Op_RParen, Id.Right_Subshell)
c_list = self.ParseCommandList()
if not c_list: return None
# Remove singleton CommandList as an optimization.
if len(c_list.children) == 1:
child = c_list.children[0]
else:
child = c_list
node = ast.Subshell(child)
right_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Right_Subshell): return None
node.spids.extend((left_spid, right_spid))
return node
def ParseIf(self):
"""
if_clause : If command_list Then command_list else_part? Fi ;
"""
if_node = ast.If()
self._Next() # skip if
cond = self.ParseCommandList()
if not cond: return None
then_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.KW_Then): return None
body = self.ParseCommandList()
if not body: return None
arm = ast.if_arm(cond.children, body.children)
arm.spids.extend((const.NO_INTEGER, then_spid)) # no if spid at first?
if_node.arms.append(arm)
if self.c_id in (Id.KW_Elif, Id.KW_Else):
if not self._ParseElifElse(if_node):
return None
else:
if_node.spids.append(const.NO_INTEGER) # no else spid
fi_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.KW_Fi): return None
if_node.spids.append(fi_spid)
return if_node
def ParseKshFunctionDef(self):
"""
ksh_function_def : 'function' fname ( '(' ')' )? newline_ok function_body
"""
left_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next() # skip past 'function'
if not self._Peek(): return None
ok, name = word.AsFuncName(self.cur_word)
if not ok:
self.AddErrorContext("Invalid function name: %r", self.cur_word)
return None
after_name_spid = word.LeftMostSpanForWord(self.cur_word) + 1
self._Next() # skip past 'function name
if not self._Peek(): return None
if self.c_id == Id.Op_LParen:
self.lexer.PushHint(Id.Op_RParen, Id.Right_FuncDef)
self._Next()
if not self._Eat(Id.Right_FuncDef): return None
# Change it: after )
after_name_spid = word.LeftMostSpanForWord(self.cur_word) + 1
if not self._NewlineOk(): return None
func = ast.FuncDef()
func.name = name
if not self.ParseFunctionBody(func):
return None
func.spids.append(left_spid)
func.spids.append(after_name_spid)
return func
def ParseCase(self):
"""
case_clause : Case WORD newline_ok in newline_ok case_list? Esac ;
"""
case_node = ast.Case()
case_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next() # skip case
if not self._Peek(): return None
case_node.to_match = self.cur_word
self._Next()
if not self._NewlineOk(): return None
in_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.KW_In): return None
if not self._NewlineOk(): return None
if self.c_id != Id.KW_Esac: # empty case list
if not self.ParseCaseList(case_node.arms):
self.AddErrorContext("ParseCase: error parsing case list")
return None
# TODO: should it return a list of nodes, and extend?
if not self._Peek(): return None
esac_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.KW_Esac): return None
self._Next()
case_node.spids.extend((case_spid, in_spid, esac_spid))
return case_node
def ParseCaseItem(self):
"""
case_item: '('? pattern ('|' pattern)* ')'
newline_ok command_term? trailer? ;
"""
self.lexer.PushHint(Id.Op_RParen, Id.Right_CasePat)
left_spid = word.LeftMostSpanForWord(self.cur_word)
if self.c_id == Id.Op_LParen:
self._Next()
pat_words = []
while True:
if not self._Peek(): return None
pat_words.append(self.cur_word)
self._Next()
if not self._Peek(): return None
if self.c_id == Id.Op_Pipe:
self._Next()
else:
break
rparen_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Right_CasePat): return None
if not self._NewlineOk(): return None
if self.c_id not in (Id.Op_DSemi, Id.KW_Esac):
c_list = self.ParseCommandTerm()
if not c_list: return None
action_children = c_list.children
else:
action_children = []
dsemi_spid = const.NO_INTEGER
last_spid = const.NO_INTEGER
if not self._Peek(): return None
if self.c_id == Id.KW_Esac:
last_spid = word.LeftMostSpanForWord(self.cur_word)
elif self.c_id == Id.Op_DSemi:
dsemi_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next()
else:
self.AddErrorContext('Expected DSEMI or ESAC, got %s', self.cur_word,
word=self.cur_word)
return None
if not self._NewlineOk(): return None
arm = ast.case_arm(pat_words, action_children)
arm.spids.extend((left_spid, rparen_spid, dsemi_spid, last_spid))
return arm
def _SplitSimpleCommandPrefix(self, words):
"""
Second pass of SimpleCommand parsing: look for assignment words.
"""
prefix_bindings = []
suffix_words = []
done_prefix = False
for w in words:
if done_prefix:
suffix_words.append(w)
continue
left_spid = word.LeftMostSpanForWord(w)
kov = word.LooksLikeAssignment(w)
if kov:
k, op, v = kov
t = word.TildeDetect(v)
if t:
# t is an unevaluated word with TildeSubPart
prefix_bindings.append((k, op, t, left_spid))
else:
prefix_bindings.append((k, op, v, left_spid)) # v is unevaluated word
else:
done_prefix = True
suffix_words.append(w)
return prefix_bindings, suffix_words
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 = dev.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 PrettyPrintError(parse_error, arena, f):
#print(parse_error)
if parse_error.token:
span_id = parse_error.token.span_id
elif parse_error.word:
# Can be -1
span_id = word.LeftMostSpanForWord(parse_error.word)
else:
span_id = -1
if span_id == -1:
line = '<no position info for token>'
path = '<unknown>'
line_num = -1
col = -1
length = -1
else:
line_span = arena.GetLineSpan(span_id)
line_id = line_span.line_id
line = arena.GetLine(line_id)
path, line_num = arena.GetDebugInfo(line_id)
col = line_span.col
length = line_span.length
print('Line %d of %r' % (line_num + 1, path), file=f)
print(' ' + line.rstrip(), file=f)
if col != -1:
f.write(' ')
# preserve tabs
for c in line[:col]:
f.write('\t' if c == '\t' else ' ')
f.write('^')
f.write('~' * (length - 1))
f.write('\n')
def ParseFunctionDef(self):
"""
function_header : fname '(' ')'
function_def : function_header newline_ok function_body ;
Precondition: Looking at the function name.
Post condition:
NOTE: There is an ambiguity with:
function foo ( echo hi ) and
function foo () ( echo hi )
Bash only accepts the latter, though it doesn't really follow a grammar.
"""
left_spid = word.LeftMostSpanForWord(self.cur_word)
ok, name = word.AsFuncName(self.cur_word)
if not ok:
self.AddErrorContext("Invalid function name: %r",
self.cur_word, word=self.cur_word)
return None
self._Next() # skip function name
# Must be true beacuse of lookahead
if not self._Peek(): return None
assert self.c_id == Id.Op_LParen, self.cur_word
self.lexer.PushHint(Id.Op_RParen, Id.Right_FuncDef)
self._Next()
if not self._Eat(Id.Right_FuncDef): return None
after_name_spid = word.LeftMostSpanForWord(self.cur_word) + 1
if not self._NewlineOk(): return None
func = ast.FuncDef()
func.name = name
if not self.ParseFunctionBody(func):
return None
func.spids.append(left_spid)
func.spids.append(after_name_spid)
return func
def ParseDoGroup(self):
"""
Used by ForEach, ForExpr, While, Until. Should this be a Do node?
do_group : Do command_list Done ; /* Apply rule 6 */
"""
if not self._Eat(Id.KW_Do): return None
do_spid = word.LeftMostSpanForWord(self.cur_word) # after _Eat
c_list = self.ParseCommandList() # could be any thing
if not c_list: return None
if not self._Eat(Id.KW_Done): return None
done_spid = word.LeftMostSpanForWord(self.cur_word) # after _Eat
node = ast.DoGroup(c_list.children)
node.spids.extend((do_spid, done_spid))
return node
def _ParseForEachLoop(self):
node = ast.ForEach()
node.do_arg_iter = False
ok, iter_name, quoted = word.StaticEval(self.cur_word)
if not ok or quoted:
self.AddErrorContext(
"Invalid for loop variable", word=self.cur_word)
return None
if not VAR_NAME_RE.match(iter_name):
self.AddErrorContext(
"Invalid for loop variable name", word=self.cur_word)
return None
node.iter_name = iter_name
self._Next() # skip past name
if not self._NewlineOk(): return None
in_spid = const.NO_INTEGER
semi_spid = const.NO_INTEGER
if not self._Peek(): return None
if self.c_id == Id.KW_In:
self._Next() # skip in
in_spid = word.LeftMostSpanForWord(self.cur_word) + 1
x = self.ParseForWords()
if x is None:
return None
iter_words, semi_spid = x
words2 = braces.BraceDetectAll(iter_words)
words3 = word.TildeDetectAll(words2)
if iter_words is None: # empty list of words is OK
return None
node.iter_words = words3
elif self.c_id == Id.Op_Semi:
node.do_arg_iter = True # implicit for loop
self._Next()
elif self.c_id == Id.KW_Do:
node.do_arg_iter = True # implicit for loop
# do not advance
else:
self.AddErrorContext("Unexpected word in for loop: %s", self.cur_word,
word=self.cur_word)
return None
node.spids.extend((in_spid, semi_spid))
body_node = self.ParseDoGroup()
if not body_node: return None
node.body = body_node
return node
def _assertSpanForWord(test, code_str):
arena, w = _assertReadWordWithArena(test, code_str)
span_id = word.LeftMostSpanForWord(w)
print(code_str)
print(span_id)
if span_id != -1:
span = arena.GetLineSpan(span_id)
print(span)
def SpanIdFromError(error):
#print(parse_error)
if error.span_id != const.NO_INTEGER:
return error.span_id
if error.token:
return error.token.span_id
if error.part:
return word.LeftMostSpanForPart(error.part)
if error.word:
return word.LeftMostSpanForWord(error.word)
return const.NO_INTEGER
def ParseSubshell(self):
left_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next() # skip past (
# Ensure that something $( (cd / && pwd) ) works. If ) is already on the
# translation stack, we want to delay it.
#print('ParseSubshell lexer.PushHint ) -> )')
self.lexer.PushHint(Id.Op_RParen, Id.Right_Subshell)
child = self.ParseCommandList()
if not child: return None
#print('SUB', self.cur_word)
node = ast.Subshell()
node.children.append(child)
right_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Right_Subshell): return None
node.spids.extend((left_spid, right_spid))
return node
def _ParseElifElse(self, if_node):
"""
else_part: (Elif command_list Then command_list)* Else command_list ;
"""
arms = if_node.arms
self._Peek()
while self.c_id == Id.KW_Elif:
elif_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next() # skip elif
cond = self.ParseCommandList()
if not cond: return None
then_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.KW_Then): return None
body = self.ParseCommandList()
if not body: return None
arm = ast.if_arm(cond.children, body.children)
arm.spids.extend((elif_spid, then_spid))
arms.append(arm)
if self.c_id == Id.KW_Else:
else_spid = word.LeftMostSpanForWord(self.cur_word)
self._Next()
body = self.ParseCommandList()
if not body: return None
if_node.else_action = body.children
else:
else_spid = const.NO_INTEGER
if_node.spids.append(else_spid)
return True
def PrintError(error_stack, arena, f):
"""
NOTE: Parse errors always occur within a single arena. Runtime errors may
span arenas (e.g. the function stack).
"""
# TODO:
# - rename to PrintParseError()
# - although parse errors happen at runtime because of 'source'
# - should there be a distinction then?
for parse_error in error_stack:
print(parse_error)
if parse_error.token:
span_id = parse_error.token.span_id
elif parse_error.word:
# Can be -1
span_id = word.LeftMostSpanForWord(parse_error.word)
else:
span_id = -1
if span_id == -1:
line = '<token had no position info>'
path = '<unknown>'
line_num = -1
col = -1
length = -1
else:
line_span = arena.GetLineSpan(span_id)
line_id = line_span.line_id
line = arena.GetLine(line_id)
path, line_num = arena.GetDebugInfo(line_id)
col = line_span.col
length = line_span.length
print('Line %d of %r' % (line_num + 1, path))
print(' ' + line.rstrip())
if col == -1:
print('NO COL')
else:
sys.stdout.write(' ')
# preserve tabs
for c in line[:col]:
sys.stdout.write('\t' if c == '\t' else ' ')
sys.stdout.write('^')
sys.stdout.write('~' * (length - 1))
sys.stdout.write('\n')
print(parse_error.UserErrorString(), file=f)
print('---')
def _ParseForEachLoop(self):
node = ast.ForEach()
node.do_arg_iter = False
ok, value, quoted = word.StaticEval(self.cur_word)
if not ok or quoted:
self.AddErrorContext(
"Invalid for loop variable: %s", self.cur_word, word=self.cur_word)
return None
node.iter_name = value
self._Next() # skip past name
if not self._NewlineOk(): return None
in_spid = -1
semi_spid = -1
if not self._Peek(): return None
if self.c_id == Id.KW_In:
self._Next() # skip in
in_spid = word.LeftMostSpanForWord(self.cur_word) + 1
iter_words, semi_spid = self.ParseForWords()
if iter_words is None: # empty list of words is OK
return None
node.iter_words = iter_words
elif self.c_id == Id.Op_Semi:
node.do_arg_iter = True # implicit for loop
self._Next()
elif self.c_id == Id.KW_Do:
node.do_arg_iter = True # implicit for loop
# do not advance
else:
self.AddErrorContext("Unexpected word in for loop: %s", self.cur_word,
word=self.cur_word)
return None
node.spids.extend((in_spid, semi_spid))
body_node = self.ParseDoGroup()
if not body_node: return None
node.body = body_node
return node
def ParseBraceGroup(self):
"""
brace_group : LBrace command_list RBrace ;
"""
left_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Lit_LBrace): return None
c_list = self.ParseCommandList()
if not c_list: return None
# Not needed
#right_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Lit_RBrace): return None
node = ast.BraceGroup(c_list.children)
node.spids.append(left_spid)
return node
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 DoArithExpr(self, node, local_symbols):
if node.tag == arith_expr_e.ArithBinary:
# Maybe I should just write the left span and right span for each word?
#self.f.write(str(node.left))
if node.op_id == Id.Arith_Plus:
# NOTE: Right isn't necessarily a word!
r_id = word.LeftMostSpanForWord(node.right.w)
#self.cursor.SkipUntil(r_id)
#self.f.write('PLUS')
#self.f.write(str(node.right))
elif node.tag == arith_expr_e.ArithWord:
self.DoWordInCommand(node.w, local_symbols)
else:
log("Unhandled: %s", node)
raise AssertionError(node.tag)
def PrettyPrintError(parse_error, arena, f=sys.stderr):
#print(parse_error)
if parse_error.span_id != const.NO_INTEGER:
span_id = parse_error.span_id
elif parse_error.token:
span_id = parse_error.token.span_id
elif parse_error.part:
span_id = word.LeftMostSpanForPart(parse_error.part)
elif parse_error.word:
span_id = word.LeftMostSpanForWord(parse_error.word)
else:
span_id = const.NO_INTEGER
if span_id == const.NO_INTEGER: # Any clause above might return this.
# This is usually a bug.
print('*** Error has no source location info ***', file=f)
return
PrintFilenameAndLine(span_id, arena, f=f)
def ParseBraceGroup(self):
"""
brace_group : LBrace command_list RBrace ;
"""
left_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Lit_LBrace): return None
node = self.ParseCommandList()
if not node: return None
# Not needed
#right_spid = word.LeftMostSpanForWord(self.cur_word)
if not self._Eat(Id.Lit_RBrace): return None
# OPTIMIZATION of AST.
# CommandList has no redirects; BraceGroup may have redirects.
if node.tag == command_e.CommandList:
n = ast.BraceGroup(node.children)
else:
n = ast.BraceGroup([node])
n.spids.append(left_spid)
return n
def _MakeAssignment(self, assign_kw, suffix_words):
bindings = []
for i, w in enumerate(suffix_words):
if i == 0:
continue # skip over local, export, etc.
left_spid = word.LeftMostSpanForWord(w)
kv = word.LooksLikeAssignment(w)
if kv:
k, v = kv
t = word.TildeDetect(v)
if t:
# t is an unevaluated word with TildeSubPart
pair = (k, t, left_spid)
else:
pair = (k, v, left_spid) # v is unevaluated word
else:
# In aboriginal in variables/sources: export_if_blank does export "$1".
# We should allow that.
ok, value, quoted = word.StaticEval(w)
if not ok or quoted:
self.AddErrorContext(
'Variable names must be constant strings, got %s', w, word=w)
return None
pair = (value, None, left_spid) # No value is equivalent to ''
bindings.append(pair)
pairs = []
for lhs, rhs, spid in bindings:
p = ast.assign_pair(ast.LeftVar(lhs), rhs)
p.spids.append(spid)
pairs.append(p)
node = ast.Assignment(assign_kw, pairs)
return node
def DoRedirect(self, node, local_symbols):
#print(node, file=sys.stderr)
op_spid = node.op.span_id
op_id = node.op.id
self.cursor.PrintUntil(op_spid)
# TODO:
# - Do < and <& the same way.
# - How to handle here docs and here docs?
# - >> becomes >+ or >-, or maybe >>>
if node.tag == redir_e.Redir:
if node.fd == const.NO_INTEGER:
if op_id == Id.Redir_Great:
self.f.write('>') # Allow us to replace the operator
self.cursor.SkipUntil(op_spid + 1)
elif op_id == Id.Redir_GreatAnd:
self.f.write('> !') # Replace >& 2 with > !2
spid = word.LeftMostSpanForWord(node.arg_word)
self.cursor.SkipUntil(spid)
#self.DoWordInCommand(node.arg_word)
else:
# NOTE: Spacing like !2>err.txt vs !2 > err.txt can be done in the
# formatter.
self.f.write('!%d ' % node.fd)
if op_id == Id.Redir_Great:
self.f.write('>')
self.cursor.SkipUntil(op_spid + 1)
elif op_id == Id.Redir_GreatAnd:
self.f.write('> !') # Replace 1>& 2 with !1 > !2
spid = word.LeftMostSpanForWord(node.arg_word)
self.cursor.SkipUntil(spid)
self.DoWordInCommand(node.arg_word, local_symbols)
elif node.tag == redir_e.HereDoc:
ok, delimiter, delim_quoted = word.StaticEval(node.here_begin)
if not ok:
p_die('Invalid here doc delimiter', word=node.here_begin)
# Turn everything into <<. We just change the quotes
self.f.write('<<')
#here_begin_spid2 = word.RightMostSpanForWord(node.here_begin)
if delim_quoted:
self.f.write(" '''")
else:
self.f.write(' """')
delim_end_spid = word.RightMostSpanForWord(node.here_begin)
self.cursor.SkipUntil(delim_end_spid + 1)
#self.cursor.SkipUntil(here_begin_spid + 1)
# Now print the lines. TODO: Have a flag to indent these to the level of
# the owning command, e.g.
# cat <<EOF
# EOF
# Or since most here docs are the top level, you could just have a hack
# for a fixed indent? TODO: Look at real use cases.
for part in node.stdin_parts:
self.DoWordPart(part, local_symbols)
self.cursor.SkipUntil(node.here_end_span_id + 1)
if delim_quoted:
self.f.write("'''\n")
else:
self.f.write('"""\n')
# Need
#self.cursor.SkipUntil(here_end_spid2)
else:
raise AssertionError(node.__class__.__name__)
# <<< 'here word'
# << 'here word'
#
# 2> out.txt
# !2 > out.txt
# cat 1<< EOF
# hello $name
# EOF
# cat !1 << """
# hello $name
# """
#
# cat << 'EOF'
# no expansion
# EOF
# cat <<- 'EOF'
# no expansion and indented
#
# cat << '''
# no expansion
# '''
# cat << '''
# no expansion and indented
# '''
# Warn about multiple here docs on a line.
# As an obscure feature, allow
# cat << \'ONE' << \"TWO"
# 123
# ONE
# 234
# TWO
# The _ is an indicator that it's not a string to be piped in.
pass
def ParseSimpleCommand(self):
"""
Fixed transcription of the POSIX grammar (TODO: port to grammar/Shell.g)
io_file : '<' filename
| LESSAND filename
...
io_here : DLESS here_end
| DLESSDASH here_end
redirect : IO_NUMBER (io_redirect | io_here)
prefix_part : ASSIGNMENT_WORD | redirect
cmd_part : WORD | redirect
assign_kw : Declare | Export | Local | Readonly
# Without any words it is parsed as a command, not an assigment
assign_listing : assign_kw
# Now we have something to do (might be changing assignment flags too)
# NOTE: any prefixes should be a warning, but they are allowed in shell.
assignment : prefix_part* assign_kw (WORD | ASSIGNMENT_WORD)+
# an external command, a function call, or a builtin -- a "word_command"
word_command : prefix_part* cmd_part+
simple_command : assign_listing
| assignment
| proc_command
Simple imperative algorithm:
1) Read a list of words and redirects. Append them to separate lists.
2) Look for the first non-assignment word. If it's declare, etc., then
keep parsing words AND assign words. Otherwise, just parse words.
3) If there are no non-assignment words, then it's a global assignment.
{ redirects, global assignments } OR
{ redirects, prefix_bindings, words } OR
{ redirects, ERROR_prefix_bindings, keyword, assignments, words }
THEN CHECK that prefix bindings don't have any array literal parts!
global assignment and keyword assignments can have the of course.
well actually EXPORT shouldn't have them either -- WARNING
3 cases we want to warn: prefix_bindings for assignment, and array literal
in prefix bindings, or export
A command can be an assignment word, word, or redirect on its own.
ls
>out.txt
>out.txt FOO=bar # this touches the file, and hten
Or any sequence:
ls foo bar
<in.txt ls foo bar >out.txt
<in.txt ls >out.txt foo bar
Or add one or more environment bindings:
VAR=val env
>out.txt VAR=val env
here_end vs filename is a matter of whether we test that it's quoted. e.g.
<<EOF vs <<'EOF'.
"""
result = self._ScanSimpleCommand()
if not result: return None
redirects, words = result
if not words: # e.g. >out.txt # redirect without words
node = ast.SimpleCommand()
node.redirects = redirects
return node
prefix_bindings, suffix_words = self._SplitSimpleCommandPrefix(words)
if not suffix_words: # ONE=1 TWO=2 (with no other words)
# TODO: Have a strict mode to prevent this?
if redirects: # >out.txt g=foo
print('WARNING: Got redirects in assignment: %s', redirects)
pairs = []
for lhs, rhs, spid in prefix_bindings:
p = ast.assign_pair(ast.LeftVar(lhs), rhs)
p.spids.append(spid)
pairs.append(p)
node = ast.Assignment(Id.Assign_None, pairs)
left_spid = word.LeftMostSpanForWord(words[0])
node.spids.append(left_spid) # no keyword spid to skip past
return node
assign_kw, keyword_spid = word.AssignmentBuiltinId(suffix_words[0])
if assign_kw == Id.Undefined_Tok:
node = self._MakeSimpleCommand(prefix_bindings, suffix_words, redirects)
return node
if redirects:
# TODO: Make it a warning, or do it in the second stage?
print(
'WARNING: Got redirects in assignment: %s' % redirects, file=sys.stderr)
if prefix_bindings: # FOO=bar local spam=eggs not allowed
# Use the location of the first value. TODO: Use the whole word before
# splitting.
_, v0, _ = prefix_bindings[0]
self.AddErrorContext(
'Invalid prefix bindings in assignment: %s', prefix_bindings,
word=v0)
return None
node = self._MakeAssignment(assign_kw, suffix_words)
if not node: return None
node.spids.append(keyword_spid)
return node
def ParseSimpleCommand(self):
"""
Fixed transcription of the POSIX grammar (TODO: port to grammar/Shell.g)
io_file : '<' filename
| LESSAND filename
...
io_here : DLESS here_end
| DLESSDASH here_end
redirect : IO_NUMBER (io_redirect | io_here)
prefix_part : ASSIGNMENT_WORD | redirect
cmd_part : WORD | redirect
assign_kw : Declare | Export | Local | Readonly
# Without any words it is parsed as a command, not an assigment
assign_listing : assign_kw
# Now we have something to do (might be changing assignment flags too)
# NOTE: any prefixes should be a warning, but they are allowed in shell.
assignment : prefix_part* assign_kw (WORD | ASSIGNMENT_WORD)+
# an external command, a function call, or a builtin -- a "word_command"
word_command : prefix_part* cmd_part+
simple_command : assign_listing
| assignment
| proc_command
Simple imperative algorithm:
1) Read a list of words and redirects. Append them to separate lists.
2) Look for the first non-assignment word. If it's declare, etc., then
keep parsing words AND assign words. Otherwise, just parse words.
3) If there are no non-assignment words, then it's a global assignment.
{ redirects, global assignments } OR
{ redirects, prefix_bindings, words } OR
{ redirects, ERROR_prefix_bindings, keyword, assignments, words }
THEN CHECK that prefix bindings don't have any array literal parts!
global assignment and keyword assignments can have the of course.
well actually EXPORT shouldn't have them either -- WARNING
3 cases we want to warn: prefix_bindings for assignment, and array literal
in prefix bindings, or export
A command can be an assignment word, word, or redirect on its own.
ls
>out.txt
>out.txt FOO=bar # this touches the file, and hten
Or any sequence:
ls foo bar
<in.txt ls foo bar >out.txt
<in.txt ls >out.txt foo bar
Or add one or more environment bindings:
VAR=val env
>out.txt VAR=val env
here_end vs filename is a matter of whether we test that it's quoted. e.g.
<<EOF vs <<'EOF'.
"""
result = self._ScanSimpleCommand()
if not result: return None
redirects, words = result
if not words: # e.g. >out.txt # redirect without words
node = ast.SimpleCommand()
node.redirects = redirects
return node
prefix_bindings, suffix_words = self._SplitSimpleCommandPrefix(words)
if not suffix_words: # ONE=1 TWO=2 (with no other words)
if redirects:
binding1 = prefix_bindings[0]
_, _, _, spid = binding1
self.AddErrorContext('Got redirects in global assignment',
span_id=spid)
return None
pairs = []
for lhs, op, rhs, spid in prefix_bindings:
p = ast.assign_pair(ast.LhsName(lhs), op, rhs)
p.spids.append(spid)
pairs.append(p)
node = ast.Assignment(Id.Assign_None, [], pairs)
left_spid = word.LeftMostSpanForWord(words[0])
node.spids.append(left_spid) # no keyword spid to skip past
return node
kind, kw_token = word.KeywordToken(suffix_words[0])
if kind == Kind.Assign:
# Here we StaticEval suffix_words[1] to see if it's a command like
# 'typeset -p'. Then it becomes a SimpleCommand node instead of an
# Assignment. Note we're not handling duplicate flags like 'typeset
# -pf'. I see this in bashdb (bash debugger) but it can just be changed
# to 'typeset -p -f'.
is_command = False
if len(suffix_words) > 1:
ok, val, _ = word.StaticEval(suffix_words[1])
if ok and (kw_token.id, val) in self._ASSIGN_COMMANDS:
is_command = True
if is_command: # declare -f, declare -p, typeset -p, etc.
node = self._MakeSimpleCommand(prefix_bindings, suffix_words,
redirects)
return node
else: # declare str='', declare -a array=()
if redirects:
# Attach the error location to the keyword. It would be more precise
# to attach it to the
self.AddErrorContext('Got redirects in assignment', token=kw_token)
return None
if prefix_bindings: # FOO=bar local spam=eggs not allowed
# Use the location of the first value. TODO: Use the whole word before
# splitting.
_, _, v0, _ = prefix_bindings[0]
self.AddErrorContext(
'Invalid prefix bindings in assignment: %s', prefix_bindings,
word=v0)
return None
node = self._MakeAssignment(kw_token.id, suffix_words)
if not node: return None
node.spids.append(kw_token.span_id)
return node
elif kind == Kind.ControlFlow:
if redirects:
self.AddErrorContext('Got redirects in control flow: %s', redirects)
return None
if prefix_bindings: # FOO=bar local spam=eggs not allowed
# Use the location of the first value. TODO: Use the whole word before
# splitting.
_, _, v0, _ = prefix_bindings[0]
self.AddErrorContext(
'Invalid prefix bindings in control flow: %s', prefix_bindings,
word=v0)
return None
# Attach the token for errors. (Assignment may not need it.)
if len(suffix_words) == 1:
arg_word = None
elif len(suffix_words) == 2:
arg_word = suffix_words[1]
else:
self.AddErrorContext('Too many arguments')
return None
return ast.ControlFlow(kw_token, arg_word)
else:
node = self._MakeSimpleCommand(prefix_bindings, suffix_words, redirects)
return node
def _MakeAssignment(self, assign_kw, suffix_words):
# First parse flags, e.g. -r -x -a -A. None of the flags have arguments.
flags = []
n = len(suffix_words)
i = 1
while i < n:
w = suffix_words[i]
ok, static_val, quoted = word.StaticEval(w)
if not ok or quoted:
break # can't statically evaluate
if static_val.startswith('-'):
flags.append(static_val)
else:
break # not a flag, rest are args
i += 1
# Now parse bindings or variable names
assignments = []
while i < n:
w = suffix_words[i]
left_spid = word.LeftMostSpanForWord(w)
kov = word.LooksLikeAssignment(w)
if kov:
k, op, v = kov
t = word.TildeDetect(v)
if t:
# t is an unevaluated word with TildeSubPart
a = (k, op, t, left_spid)
else:
a = (k, op, v, left_spid) # v is unevaluated word
else:
# In aboriginal in variables/sources: export_if_blank does export "$1".
# We should allow that.
# Parse this differently then?
# dynamic-export?
# It sets global variables.
ok, static_val, quoted = word.StaticEval(w)
if not ok or quoted:
self.AddErrorContext(
'Variable names must be constant strings, got %s', w, word=w)
return None
# No value is equivalent to ''
m = VAR_NAME_RE.match(static_val)
if not m:
self.AddErrorContext('Invalid variable name %r', static_val, word=w)
return None
a = (static_val, assign_op_e.Equal, None, left_spid)
assignments.append(a)
i += 1
# TODO: Also make with LhsIndexedName
pairs = []
for lhs, op, rhs, spid in assignments:
p = ast.assign_pair(ast.LhsName(lhs), op, rhs)
p.spids.append(spid)
pairs.append(p)
node = ast.Assignment(assign_kw, flags, pairs)
return node
def DoRedirect(self, node, local_symbols):
#print(node, file=sys.stderr)
self.cursor.PrintUntil(node.spids[0])
# TODO:
# - Do < and <& the same way.
# - How to handle here docs and here docs?
# - >> becomes >+ or >-, or maybe >>>
if node.fd == -1:
if node.op_id == Id.Redir_Great:
self.f.write('>') # Allow us to replace the operator
self.cursor.SkipUntil(node.spids[0] + 1)
elif node.op_id == Id.Redir_GreatAnd:
self.f.write('> !') # Replace >& 2 with > !2
spid = word.LeftMostSpanForWord(node.arg_word)
self.cursor.SkipUntil(spid)
#self.DoWordInCommand(node.arg_word)
else:
# NOTE: Spacing like !2>err.txt vs !2 > err.txt can be done in the
# formatter.
self.f.write('!%d ' % node.fd)
if node.op_id == Id.Redir_Great:
self.f.write('>')
self.cursor.SkipUntil(node.spids[0] + 1)
elif node.op_id == Id.Redir_GreatAnd:
self.f.write('> !') # Replace 1>& 2 with !1 > !2
spid = word.LeftMostSpanForWord(node.arg_word)
self.cursor.SkipUntil(spid)
self.DoWordInCommand(node.arg_word, local_symbols)
# <<< 'here word'
# << 'here word'
#
# 2> out.txt
# !2 > out.txt
# cat 1<< EOF
# hello $name
# EOF
# cat !1 << """
# hello $name
# """
#
# cat << 'EOF'
# no expansion
# EOF
# cat <<- 'EOF'
# no expansion and indented
#
# cat << '''
# no expansion
# '''
# cat << '''
# no expansion and indented
# '''
# Warn about multiple here docs on a line.
# As an obscure feature, allow
# cat << \'ONE' << \"TWO"
# 123
# ONE
# 234
# TWO
# The _ is an indicator that it's not a string to be piped in.
pass
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 DoCommand(self, node, local_symbols, at_top_level=False):
if node.tag == command_e.CommandList:
# TODO: How to distinguish between echo hi; echo bye; and on separate
# lines
for child in node.children:
self.DoCommand(child, local_symbols, at_top_level=at_top_level)
elif node.tag == command_e.SimpleCommand:
# How to preserve spaces between words? Do you want to do it?
# Well you need to test this:
#
# echo foo \
# bar
# TODO: Need to print until the left most part of the phrase? the phrase
# is a word, binding, redirect.
#self.cursor.PrintUntil()
if node.more_env:
(left_spid,) = node.more_env[0].spids
self.cursor.PrintUntil(left_spid)
self.f.write('env ')
# We only need to transform the right side, not left side.
for pair in node.more_env:
self.DoWordInCommand(pair.val, local_symbols)
# More translations:
# - . to source
# - eval to sh-eval
if node.words:
first_word = node.words[0]
ok, val, quoted = word.StaticEval(first_word)
word0_spid = word.LeftMostSpanForWord(first_word)
if ok and not quoted:
if val == '[':
last_word = node.words[-1]
# Check if last word is ]
ok, val, quoted = word.StaticEval(last_word)
if ok and not quoted and val == ']':
# Replace [ with 'test'
self.cursor.PrintUntil(word0_spid)
self.cursor.SkipUntil(word0_spid + 1)
self.f.write('test')
for w in node.words[1:-1]:
self.DoWordInCommand(w, local_symbols)
# Now omit ]
last_spid = word.LeftMostSpanForWord(last_word)
self.cursor.PrintUntil(last_spid - 1) # Get the space before
self.cursor.SkipUntil(last_spid + 1) # ] takes one spid
return
else:
raise RuntimeError('Got [ without ]')
elif val == '.':
self.cursor.PrintUntil(word0_spid)
self.cursor.SkipUntil(word0_spid + 1)
self.f.write('source')
return
for w in node.words:
self.DoWordInCommand(w, local_symbols)
# NOTE: This will change to "phrase"? Word or redirect.
for r in node.redirects:
self.DoRedirect(r, local_symbols)
# TODO: Print the terminator. Could be \n or ;
# Need to print env like PYTHONPATH = 'foo' && ls
# Need to print redirects:
# < > are the same. << is here string, and >> is assignment.
# append is >+
# TODO: static_eval of simple command
# - [ -> "test". Eliminate trailing ].
# - . -> source, etc.
elif node.tag == command_e.Assignment:
self.DoAssignment(node, at_top_level, local_symbols)
elif node.tag == command_e.Pipeline:
# Obscure: |& turns into |- or |+ for stderr.
# TODO:
# if ! true; then -> if not true {
# if ! echo | grep; then -> if not { echo | grep } {
# }
# not is like do {}, but it negates the return value I guess.
for child in node.children:
self.DoCommand(child, local_symbols)
elif node.tag == command_e.AndOr:
for child in node.children:
self.DoCommand(child, local_symbols)
elif node.tag == command_e.Sentence:
# 'ls &' to 'fork ls'
# Keep ; the same.
self.DoCommand(node.child, local_symbols)
# This has to be different in the function case.
elif node.tag == command_e.BraceGroup:
# { echo hi; } -> do { echo hi }
# For now it might be OK to keep 'do { echo hi; }
#left_spid, right_spid = node.spids
(left_spid,) = node.spids
self.cursor.PrintUntil(left_spid)
self.cursor.SkipUntil(left_spid + 1)
self.f.write('do {')
for child in node.children:
self.DoCommand(child, local_symbols)
elif node.tag == command_e.Subshell:
# (echo hi) -> shell echo hi
# (echo hi; echo bye) -> shell {echo hi; echo bye}
(left_spid, right_spid) = node.spids
self.cursor.PrintUntil(left_spid)
self.cursor.SkipUntil(left_spid + 1)
self.f.write('shell {')
self.DoCommand(node.child, local_symbols)
#self._DebugSpid(right_spid)
#self._DebugSpid(right_spid + 1)
#print('RIGHT SPID', right_spid)
self.cursor.PrintUntil(right_spid)
self.cursor.SkipUntil(right_spid + 1)
self.f.write('}')
elif node.tag == command_e.DParen:
# (( a == 0 )) is sh-expr ' a == 0 '
#
# NOTE: (( n++ )) is auto-translated to sh-expr 'n++', but could be set
# n++.
left_spid, right_spid = node.spids
self.cursor.PrintUntil(left_spid)
self.cursor.SkipUntil(left_spid + 1)
self.f.write("sh-expr '")
self.cursor.PrintUntil(right_spid - 1) # before ))
self.cursor.SkipUntil(right_spid + 1) # after )) -- each one is a token
self.f.write("'")
elif node.tag == command_e.DBracket:
# [[ 1 -eq 2 ]] to (1 == 2)
self.DoBoolExpr(node.expr)
elif node.tag == command_e.FuncDef:
# TODO: skip name
#self.f.write('proc %s' % node.name)
# New symbol table for every function.
new_local_symbols = {}
# Should be the left most span, including 'function'
self.cursor.PrintUntil(node.spids[0])
self.f.write('proc ')
self.f.write(node.name)
self.cursor.SkipUntil(node.spids[1])
if node.body.tag == command_e.BraceGroup:
# Don't add "do" like a standalone brace group. Just use {}.
for child in node.body.children:
self.DoCommand(child, new_local_symbols)
else:
pass
# Add {}.
# proc foo {
# shell {echo hi; echo bye}
# }
#self.DoCommand(node.body)
elif node.tag == command_e.BraceGroup:
for child in node.children:
self.DoCommand(child, local_symbols)
elif node.tag == command_e.DoGroup:
do_spid, done_spid = node.spids
self.cursor.PrintUntil(do_spid)
self.cursor.SkipUntil(do_spid + 1)
self.f.write('{')
for child in node.children:
self.DoCommand(child, local_symbols)
self.cursor.PrintUntil(done_spid)
self.cursor.SkipUntil(done_spid + 1)
self.f.write('}')
elif node.tag == command_e.ForEach:
# Need to preserve spaces between words, because there can be line
# wrapping.
# for x in a b c \
# d e f; do
in_spid, semi_spid = node.spids
if in_spid == const.NO_INTEGER:
#self.cursor.PrintUntil() # 'for x' and then space
self.f.write('for %s in @Argv ' % node.iter_name)
self.cursor.SkipUntil(node.body.spids[0])
else:
self.cursor.PrintUntil(in_spid + 1) # 'for x in' and then space
self.f.write('[')
for w in node.iter_words:
self.DoWordInCommand(w, local_symbols)
self.f.write(']')
#print("SKIPPING SEMI %d" % semi_spid, file=sys.stderr)
if semi_spid != const.NO_INTEGER:
self.cursor.PrintUntil(semi_spid)
self.cursor.SkipUntil(semi_spid + 1)
self.DoCommand(node.body, local_symbols)
elif node.tag == command_e.ForExpr:
# Change (( )) to ( ), and then _FixDoGroup
pass
elif node.tag == command_e.WhileUntil:
# Skip 'until', and replace it with 'while not'
if node.keyword.id == Id.KW_Until:
kw_spid = node.keyword.span_id
self.cursor.PrintUntil(kw_spid)
self.f.write('while not')
self.cursor.SkipUntil(kw_spid + 1)
cond = node.cond
# Skip the semi-colon in the condition, which is ususally a Sentence
if len(cond) == 1 and cond[0].tag == command_e.Sentence:
self.DoCommand(cond[0].child, local_symbols)
semi_spid = cond[0].terminator.span_id
self.cursor.SkipUntil(semi_spid + 1)
self.DoCommand(node.body, local_symbols)
elif node.tag == command_e.If:
else_spid, fi_spid = node.spids
# if foo; then -> if foo {
# elif foo; then -> } elif foo {
for arm in node.arms:
elif_spid, then_spid = arm.spids
if elif_spid != const.NO_INTEGER:
self.cursor.PrintUntil(elif_spid)
self.f.write('} ')
cond = arm.cond
if len(cond) == 1 and cond[0].tag == command_e.Sentence:
sentence = cond[0]
self.DoCommand(sentence, local_symbols)
# Remove semi-colon
semi_spid = sentence.terminator.span_id
self.cursor.PrintUntil(semi_spid)
self.cursor.SkipUntil(semi_spid + 1)
else:
for child in arm.cond:
self.DoCommand(child, local_symbols)
self.cursor.PrintUntil(then_spid)
self.cursor.SkipUntil(then_spid + 1)
self.f.write('{')
for child in arm.action:
self.DoCommand(child, local_symbols)
# else -> } else {
if node.else_action:
self.cursor.PrintUntil(else_spid)
self.f.write('} ')
self.cursor.PrintUntil(else_spid + 1)
self.f.write(' {')
for child in node.else_action:
self.DoCommand(child, local_symbols)
# fi -> }
self.cursor.PrintUntil(fi_spid)
self.cursor.SkipUntil(fi_spid + 1)
self.f.write('}')
elif node.tag == command_e.Case:
case_spid, in_spid, esac_spid = node.spids
self.cursor.PrintUntil(case_spid)
self.cursor.SkipUntil(case_spid + 1)
self.f.write('match')
# Reformat "$1" to $1
self.DoWordInCommand(node.to_match, local_symbols)
self.cursor.PrintUntil(in_spid)
self.cursor.SkipUntil(in_spid + 1)
self.f.write('{') # matchstr $var {
# each arm needs the ) and the ;; node to skip over?
for arm in node.arms:
left_spid, rparen_spid, dsemi_spid, last_spid = arm.spids
#print(left_spid, rparen_spid, dsemi_spid)
self.cursor.PrintUntil(left_spid)
# Hm maybe keep | because it's semi-deprecated? You acn use
# reload|force-relaod {
# }
# e/reload|force-reload/ {
# }
# / 'reload' or 'force-reload' / {
# }
#
# Yeah it's the more abbreviated syntax.
# change | to 'or'
for pat in arm.pat_list:
pass
self.f.write('with ')
# Remove the )
self.cursor.PrintUntil(rparen_spid)
self.cursor.SkipUntil(rparen_spid + 1)
for child in arm.action:
self.DoCommand(child, local_symbols)
if dsemi_spid != const.NO_INTEGER:
# Remove ;;
self.cursor.PrintUntil(dsemi_spid)
self.cursor.SkipUntil(dsemi_spid + 1)
elif last_spid != const.NO_INTEGER:
self.cursor.PrintUntil(last_spid)
else:
raise AssertionError(
"Expected with dsemi_spid or last_spid in case arm")
self.cursor.PrintUntil(esac_spid)
self.cursor.SkipUntil(esac_spid + 1)
self.f.write('}') # strmatch $var {
elif node.tag == command_e.NoOp:
pass
elif node.tag == command_e.ControlFlow:
# No change for break / return / continue
pass
elif node.tag == command_e.TimeBlock:
self.DoCommand(node.pipeline, local_symbols)
else:
#log('Command not handled: %s', node)
raise AssertionError(node.__class__.__name__)