osh/cmd_parse.py

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

2787 lines, 1420 significant

1	# Copyright 2016 Andy Chu. All rights reserved.
2	# Licensed under the Apache License, Version 2.0 (the "License");
3	# you may not use this file except in compliance with the License.
4	# You may obtain a copy of the License at
5	#
6	# http://www.apache.org/licenses/LICENSE-2.0
7	"""
8	cmd_parse.py - Parse high level shell commands.
9	"""
10	from __future__ import print_function
11
12	from _devbuild.gen import grammar_nt
13	from _devbuild.gen.id_kind_asdl import Id, Id_t, Id_str, Kind, Kind_str
14	from _devbuild.gen.types_asdl import lex_mode_e, cmd_mode_e, cmd_mode_t
15	from _devbuild.gen.syntax_asdl import (
16	loc,
17	SourceLine,
18	source,
19	parse_result,
20	parse_result_t,
21	command,
22	command_t,
23	condition,
24	condition_t,
25	for_iter,
26	ArgList,
27	BraceGroup,
28	LiteralBlock,
29	CaseArm,
30	case_arg,
31	IfArm,
32	pat,
33	pat_t,
34	Redir,
35	redir_param,
36	redir_loc,
37	redir_loc_t,
38	word_e,
39	word_t,
40	CompoundWord,
41	Token,
42	word_part_e,
43	word_part_t,
44	rhs_word,
45	rhs_word_t,
46	sh_lhs,
47	sh_lhs_t,
48	AssignPair,
49	EnvPair,
50	ParsedAssignment,
51	assign_op_e,
52	NameType,
53	proc_sig,
54	proc_sig_e,
55	Proc,
56	Func,
57	)
58	from core import alloc
59	from core import error
60	from core.error import p_die
61	from core import ui
62	from frontend import consts
63	from frontend import lexer
64	from frontend import location
65	from frontend import match
66	from frontend import reader
67	from mycpp.mylib import log
68	from osh import braces
69	from osh import bool_parse
70	from osh import word_
71
72	from typing import Optional, List, Dict, Any, Tuple, cast, TYPE_CHECKING
73	if TYPE_CHECKING:
74	from core.alloc import Arena
75	from core import optview
76	from frontend.lexer import Lexer
77	from frontend.parse_lib import ParseContext, AliasesInFlight
78	from frontend.reader import _Reader
79	from osh.word_parse import WordParser
80
81	_ = Kind_str # for debug prints
82
83	TAB_CH = 9 # ord('\t')
84	SPACE_CH = 32 # ord(' ')
85
86
87	def _ReadHereLines(
88	line_reader, # type: _Reader
89	h, # type: Redir
90	delimiter, # type: str
91	):
92	# type: (...) -> Tuple[List[Tuple[SourceLine, int]], Tuple[SourceLine, int]]
93	# NOTE: We read all lines at once, instead of parsing line-by-line,
94	# because of cases like this:
95	# cat <<EOF
96	# 1 $(echo 2
97	# echo 3) 4
98	# EOF
99	here_lines = [] # type: List[Tuple[SourceLine, int]]
100	last_line = None # type: Tuple[SourceLine, int]
101	strip_leading_tabs = (h.op.id == Id.Redir_DLessDash)
102
103	while True:
104	src_line, unused_offset = line_reader.GetLine()
105
106	if src_line is None: # EOF
107	# An unterminated here doc is just a warning in bash. We make it
108	# fatal because we want to be strict, and because it causes problems
109	# reporting other errors.
110	# Attribute it to the << in <<EOF for now.
111	p_die("Couldn't find terminator for here doc that starts here",
112	h.op)
113
114	assert len(src_line.content) != 0 # None should be the empty line
115
116	line = src_line.content
117
118	# If op is <<-, strip off ALL leading tabs -- not spaces, and not just
119	# the first tab.
120	start_offset = 0
121	if strip_leading_tabs:
122	n = len(line)
123	i = 0 # used after loop exit
124	while i < n:
125	if line[i] != '\t':
126	break
127	i += 1
128	start_offset = i
129
130	if line[start_offset:].rstrip() == delimiter:
131	last_line = (src_line, start_offset)
132	break
133
134	here_lines.append((src_line, start_offset))
135
136	return here_lines, last_line
137
138
139	def _MakeLiteralHereLines(
140	here_lines, # type: List[Tuple[SourceLine, int]]
141	arena, # type: Arena
142	do_lossless, # type: bool
143	):
144	# type: (...) -> List[word_part_t]
145	"""Create a Token for each line.
146
147	For <<'EOF' and <<-'EOF' - single quoted rule
148
149	<<- has non-zero start_offset
150	"""
151	# less precise type, because List[T] is an invariant type
152	tokens = [] # type: List[word_part_t]
153	for src_line, start_offset in here_lines:
154
155	# Maintain lossless invariant for STRIPPED tabs: add a Token to the
156	# arena invariant, but don't refer to it.
157	#
158	# Note: We could use Lit_CharsWithoutPrefix for 'single quoted' EOF
159	# here docs, but it's more complex with double quoted EOF docs.
160
161	if do_lossless: # avoid garbage, doesn't affect correctness
162	arena.NewToken(Id.Lit_CharsWithoutPrefix, start_offset, 0,
163	src_line)
164
165	t = arena.NewToken(Id.Lit_Chars, start_offset, len(src_line.content),
166	src_line)
167	tokens.append(t)
168	return tokens
169
170
171	def _ParseHereDocBody(parse_ctx, r, line_reader, arena):
172	# type: (ParseContext, Redir, _Reader, Arena) -> None
173	"""Fill in attributes of a pending here doc node."""
174	h = cast(redir_param.HereDoc, r.arg)
175	# "If any character in word is quoted, the delimiter shall be formed by
176	# performing quote removal on word, and the here-document lines shall not
177	# be expanded. Otherwise, the delimiter shall be the word itself."
178	# NOTE: \EOF counts, or even E\OF
179	ok, delimiter, delim_quoted = word_.StaticEval(h.here_begin)
180	if not ok:
181	p_die('Invalid here doc delimiter', loc.Word(h.here_begin))
182
183	here_lines, last_line = _ReadHereLines(line_reader, r, delimiter)
184
185	if delim_quoted:
186	# <<'EOF' and <<-'EOF' - Literal for each line.
187	h.stdin_parts = _MakeLiteralHereLines(here_lines, arena,
188	parse_ctx.do_lossless)
189	else:
190	# <<EOF and <<-EOF - Parse as word
191	line_reader = reader.VirtualLineReader(arena, here_lines,
192	parse_ctx.do_lossless)
193	w_parser = parse_ctx.MakeWordParserForHereDoc(line_reader)
194	w_parser.ReadHereDocBody(h.stdin_parts) # fills this in
195
196	end_line, start_offset = last_line
197
198	# Maintain lossless invariant for STRIPPED tabs: add a Token to the
199	# arena invariant, but don't refer to it.
200	if parse_ctx.do_lossless: # avoid garbage, doesn't affect correctness
201	arena.NewToken(Id.Lit_CharsWithoutPrefix, start_offset, 0, end_line)
202
203	# Create a Token with the end terminator. Maintains the invariant that the
204	# tokens "add up".
205	h.here_end_tok = arena.NewToken(Id.Undefined_Tok, start_offset,
206	len(end_line.content), end_line)
207
208
209	def _MakeAssignPair(parse_ctx, preparsed, arena):
210	# type: (ParseContext, ParsedAssignment, Arena) -> AssignPair
211	"""Create an AssignPair from a 4-tuples from DetectShAssignment."""
212
213	left_token = preparsed.left
214	close_token = preparsed.close
215
216	lhs = None # type: sh_lhs_t
217
218	if left_token.id == Id.Lit_VarLike: # s=1
219	if lexer.IsPlusEquals(left_token):
220	var_name = lexer.TokenSliceRight(left_token, -2)
221	op = assign_op_e.PlusEqual
222	else:
223	var_name = lexer.TokenSliceRight(left_token, -1)
224	op = assign_op_e.Equal
225
226	lhs = sh_lhs.Name(left_token, var_name)
227
228	elif left_token.id == Id.Lit_ArrayLhsOpen and parse_ctx.do_lossless:
229	var_name = lexer.TokenSliceRight(left_token, -1)
230	if lexer.IsPlusEquals(close_token):
231	op = assign_op_e.PlusEqual
232	else:
233	op = assign_op_e.Equal
234
235	assert left_token.line == close_token.line, \
236	'%s and %s not on same line' % (left_token, close_token)
237
238	left_pos = left_token.col + left_token.length
239	index_str = left_token.line.content[left_pos:close_token.col]
240	lhs = sh_lhs.UnparsedIndex(left_token, var_name, index_str)
241
242	elif left_token.id == Id.Lit_ArrayLhsOpen: # a[x++]=1
243	var_name = lexer.TokenSliceRight(left_token, -1)
244	if lexer.IsPlusEquals(close_token):
245	op = assign_op_e.PlusEqual
246	else:
247	op = assign_op_e.Equal
248
249	# Similar to SnipCodeString / SnipCodeBlock
250	if left_token.line == close_token.line:
251	# extract what's between brackets
252	s = left_token.col + left_token.length
253	code_str = left_token.line.content[s:close_token.col]
254	else:
255	raise NotImplementedError('%s != %s' %
256	(left_token.line, close_token.line))
257	a_parser = parse_ctx.MakeArithParser(code_str)
258
259	# a[i+1]= is a LHS
260	src = source.Reparsed('array LHS', left_token, close_token)
261	with alloc.ctx_SourceCode(arena, src):
262	index_node = a_parser.Parse() # may raise error.Parse
263
264	lhs = sh_lhs.IndexedName(left_token, var_name, index_node)
265
266	else:
267	raise AssertionError()
268
269	# TODO: Should we also create a rhs_expr.ArrayLiteral here?
270	parts = preparsed.w.parts
271	offset = preparsed.part_offset
272
273	n = len(parts)
274	if offset == n:
275	rhs = rhs_word.Empty # type: rhs_word_t
276	else:
277	w = CompoundWord(parts[offset:])
278	word_.TildeDetectAssign(w)
279	rhs = w
280
281	return AssignPair(left_token, lhs, op, rhs)
282
283
284	def _AppendMoreEnv(preparsed_list, more_env):
285	# type: (List[ParsedAssignment], List[EnvPair]) -> None
286	"""Helper to modify a SimpleCommand node.
287
288	Args:
289	preparsed: a list of 4-tuples from DetectShAssignment
290	more_env: a list to append env_pairs to
291	"""
292	for preparsed in preparsed_list:
293	left_token = preparsed.left
294
295	if left_token.id != Id.Lit_VarLike: # can't be a[x]=1
296	p_die(
297	"Environment binding shouldn't look like an array assignment",
298	left_token)
299
300	if lexer.IsPlusEquals(left_token):
301	p_die('Expected = in environment binding, got +=', left_token)
302
303	var_name = lexer.TokenSliceRight(left_token, -1)
304
305	parts = preparsed.w.parts
306	n = len(parts)
307	offset = preparsed.part_offset
308	if offset == n:
309	rhs = rhs_word.Empty # type: rhs_word_t
310	else:
311	w = CompoundWord(parts[offset:])
312	word_.TildeDetectAssign(w)
313	rhs = w
314
315	more_env.append(EnvPair(left_token, var_name, rhs))
316
317
318	def _SplitSimpleCommandPrefix(words):
319	# type: (List[CompoundWord]) -> Tuple[List[ParsedAssignment], List[CompoundWord]]
320	"""Second pass of SimpleCommand parsing: look for assignment words."""
321	preparsed_list = [] # type: List[ParsedAssignment]
322	suffix_words = [] # type: List[CompoundWord]
323
324	done_prefix = False
325	for w in words:
326	if done_prefix:
327	suffix_words.append(w)
328	continue
329
330	left_token, close_token, part_offset = word_.DetectShAssignment(w)
331	if left_token:
332	preparsed_list.append(
333	ParsedAssignment(left_token, close_token, part_offset, w))
334	else:
335	done_prefix = True
336	suffix_words.append(w)
337
338	return preparsed_list, suffix_words
339
340
341	def _MakeSimpleCommand(
342	preparsed_list, # type: List[ParsedAssignment]
343	suffix_words, # type: List[CompoundWord]
344	typed_args, # type: Optional[ArgList]
345	block, # type: Optional[LiteralBlock]
346	):
347	# type: (...) -> command.Simple
348	"""Create a command.Simple"""
349
350	# FOO=(1 2 3) ls is not allowed.
351	for preparsed in preparsed_list:
352	if word_.HasArrayPart(preparsed.w):
353	p_die("Environment bindings can't contain array literals",
354	loc.Word(preparsed.w))
355
356	# NOTE: It would be possible to add this check back. But it already happens
357	# at runtime in EvalWordSequence2.
358	# echo FOO=(1 2 3) is not allowed (but we should NOT fail on echo FOO[x]=1).
359	if 0:
360	for w in suffix_words:
361	if word_.HasArrayPart(w):
362	p_die("Commands can't contain array literals", loc.Word(w))
363
364	assert len(suffix_words) != 0
365	# {a,b,c} # Use { before brace detection
366	# ~/bin/ls # Use ~ before tilde detection
367	part0 = suffix_words[0].parts[0]
368	blame_tok = location.LeftTokenForWordPart(part0)
369
370	# NOTE: We only do brace DETECTION here, not brace EXPANSION. Therefore we
371	# can't implement bash's behavior of having say {~bob,~jane}/src work,
372	# because we only have a BracedTree.
373	# This is documented in spec/brace-expansion.
374	# NOTE: Technically we could do expansion outside of 'oshc translate', but it
375	# doesn't seem worth it.
376	words2 = braces.BraceDetectAll(suffix_words)
377	words3 = word_.TildeDetectAll(words2)
378
379	more_env = [] # type: List[EnvPair]
380	_AppendMoreEnv(preparsed_list, more_env)
381
382	# do_fork by default
383	return command.Simple(blame_tok, more_env, words3, typed_args, block, True)
384
385
386	class VarChecker(object):
387	"""Statically check for proc and variable usage errors."""
388
389	def __init__(self):
390	# type: () -> None
391	"""
392	Args:
393	oil_proc: Whether to disallow nested proc/function declarations
394	"""
395	# self.tokens for location info: 'proc' or another token
396	self.tokens = [] # type: List[Token]
397	self.names = [] # type: List[Dict[str, Id_t]]
398
399	def Push(self, blame_tok):
400	# type: (Token) -> None
401	"""Called when we enter a shell function, proc, or func.
402
403	Bash allows this, but it's confusing because it's the same as two
404	functions at the top level.
405
406	f() {
407	g() {
408	echo 'top level function defined in another one'
409	}
410	}
411
412	YSH disallows nested procs and funcs.
413	"""
414	if len(self.tokens) != 0:
415	if blame_tok.id == Id.KW_Proc:
416	p_die("procs must be defined at the top level", blame_tok)
417	if blame_tok.id == Id.KW_Func:
418	p_die("funcs must be defined at the top level", blame_tok)
419	if self.tokens[0].id in (Id.KW_Proc, Id.KW_Func):
420	p_die("shell functions can't be defined inside proc or func",
421	blame_tok)
422
423	self.tokens.append(blame_tok)
424	entry = {} # type: Dict[str, Id_t]
425	self.names.append(entry)
426
427	def Pop(self):
428	# type: () -> None
429	self.names.pop()
430	self.tokens.pop()
431
432	def Check(self, keyword_id, var_name, blame_tok):
433	# type: (Id_t, str, Token) -> None
434	"""Check for declaration / mutation errors in proc and func.
435
436	var x
437	x already declared
438	setvar x:
439	x is not declared
440	setglobal x:
441	No errors are possible; we would need all these many conditions to
442	statically know the names:
443	- no 'source'
444	- shopt -u copy_env.
445	- AND use lib has to be static
446
447	What about bare assignment in Hay? I think these are dynamic checks --
448	there is no static check. Hay is for building up data imperatively,
449	and then LATER, right before main(), it can be type checked.
450
451	Package {
452	version = '3.11'
453	version = '3.12'
454	}
455	"""
456	# No static checks are the global level! Because of 'source', var and
457	# setvar are essentially the same.
458	if len(self.names) == 0:
459	return
460
461	top = self.names[-1]
462	if keyword_id == Id.KW_Var:
463	if var_name in top:
464	p_die('%r was already declared' % var_name, blame_tok)
465	else:
466	top[var_name] = keyword_id
467
468	if keyword_id == Id.KW_SetVar:
469	if var_name not in top:
470	# Note: the solution could be setglobal, etc.
471	p_die(
472	"setvar couldn't find matching 'var %s' (OILS-ERR-10)" %
473	var_name, blame_tok)
474
475
476	class ctx_VarChecker(object):
477
478	def __init__(self, var_checker, blame_tok):
479	# type: (VarChecker, Token) -> None
480	var_checker.Push(blame_tok)
481	self.var_checker = var_checker
482
483	def __enter__(self):
484	# type: () -> None
485	pass
486
487	def __exit__(self, type, value, traceback):
488	# type: (Any, Any, Any) -> None
489	self.var_checker.Pop()
490
491
492	class ctx_CmdMode(object):
493
494	def __init__(self, cmd_parse, new_cmd_mode):
495	# type: (CommandParser, cmd_mode_t) -> None
496	self.cmd_parse = cmd_parse
497	self.prev_cmd_mode = cmd_parse.cmd_mode
498	cmd_parse.cmd_mode = new_cmd_mode
499
500	def __enter__(self):
501	# type: () -> None
502	pass
503
504	def __exit__(self, type, value, traceback):
505	# type: (Any, Any, Any) -> None
506	self.cmd_parse.cmd_mode = self.prev_cmd_mode
507
508
509	SECONDARY_KEYWORDS = [
510	Id.KW_Do, Id.KW_Done, Id.KW_Then, Id.KW_Fi, Id.KW_Elif, Id.KW_Else,
511	Id.KW_Esac
512	]
513
514
515	class CommandParser(object):
516	"""Recursive descent parser derived from POSIX shell grammar.
517
518	This is a BNF grammar:
519	https://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#tag_18_10
520
521	- Augmented with both bash/OSH and YSH constructs.
522
523	- We use regex-like iteration rather than recursive references
524	? means optional (0 or 1)
525	* means 0 or more
526	+ means 1 or more
527
528	- Keywords are spelled in Caps:
529	If Elif Case
530
531	- Operator tokens are quoted:
532	'(' '\|'
533
534	or can be spelled directly if it matters:
535
536	Op_LParen Op_Pipe
537
538	- Non-terminals are snake_case:
539	brace_group subshell
540
541	Methods in this class should ROUGHLY CORRESPOND to grammar productions, and
542	the production should be in the method docstrings, e.g.
543
544	def ParseSubshell():
545	"
546	subshell : '(' compound_list ')'
547
548	Looking at Op_LParen # Comment to say how this method is called
549	"
550
551	The grammar may be factored to make parsing easier.
552	"""
553
554	def __init__(self,
555	parse_ctx,
556	parse_opts,
557	w_parser,
558	lexer,
559	line_reader,
560	eof_id=Id.Eof_Real):
561	# type: (ParseContext, optview.Parse, WordParser, Lexer, _Reader, Id_t) -> None
562	self.parse_ctx = parse_ctx
563	self.aliases = parse_ctx.aliases # aliases to expand at parse time
564
565	self.parse_opts = parse_opts
566	self.w_parser = w_parser # type: WordParser # for normal parsing
567	self.lexer = lexer # for pushing hints, lookahead to (
568	self.line_reader = line_reader # for here docs
569	self.eof_id = eof_id
570
571	self.arena = line_reader.arena # for adding here doc and alias spans
572	self.aliases_in_flight = [] # type: AliasesInFlight
573
574	# A hacky boolean to remove 'if cd / {' ambiguity.
575	self.allow_block = True
576
577	# Stack of booleans for nested Attr and SHELL nodes.
578	# Attr nodes allow bare assignment x = 42, but not shell x=42.
579	# SHELL nodes are the inverse. 'var x = 42' is preferred in shell
580	# nodes, but x42 is still allowed.
581	#
582	# Note: this stack could be optimized by turning it into an integer and
583	# binary encoding.
584	self.hay_attrs_stack = [] # type: List[bool]
585
586	# Note: VarChecker is instantiated with each CommandParser, which means
587	# that two 'proc foo' -- inside a command sub and outside -- don't
588	# conflict, because they use different CommandParser instances. I think
589	# this OK but you can imagine different behaviors.
590	self.var_checker = VarChecker()
591
592	self.cmd_mode = cmd_mode_e.Shell # type: cmd_mode_t
593
594	self.Reset()
595
596	# Init_() function for "keyword arg"
597	def Init_AliasesInFlight(self, aliases_in_flight):
598	# type: (AliasesInFlight) -> None
599	self.aliases_in_flight = aliases_in_flight
600
601	def Reset(self):
602	# type: () -> None
603	"""Reset our own internal state.
604
605	Called by the interactive loop.
606	"""
607	# Cursor state set by _GetWord()
608	self.next_lex_mode = lex_mode_e.ShCommand
609	self.cur_word = None # type: word_t # current word
610	self.c_kind = Kind.Undefined
611	self.c_id = Id.Undefined_Tok
612
613	self.pending_here_docs = [] # type: List[Redir]
614
615	def ResetInputObjects(self):
616	# type: () -> None
617	"""Reset the internal state of our inputs.
618
619	Called by the interactive loop.
620	"""
621	self.w_parser.Reset()
622	self.lexer.ResetInputObjects()
623	self.line_reader.Reset()
624
625	def _SetNext(self):
626	# type: () -> None
627	"""Call this when you no longer need the current token.
628
629	This method is lazy. A subsequent call to _GetWord() will
630	actually read the next Token.
631	"""
632	self.next_lex_mode = lex_mode_e.ShCommand
633
634	def _SetNextBrack(self):
635	# type: () -> None
636	self.next_lex_mode = lex_mode_e.ShCommandFakeBrack
637
638	def _GetWord(self):
639	# type: () -> None
640	"""Call this when you need to make a decision based on Id or Kind.
641
642	If there was an "unfulfilled" call to _SetNext(), it reads a word and sets
643	self.c_id and self.c_kind.
644
645	Otherwise it does nothing.
646	"""
647	if self.next_lex_mode != lex_mode_e.Undefined:
648	w = self.w_parser.ReadWord(self.next_lex_mode)
649	#log("w %s", w)
650
651	# Here docs only happen in command mode, so other kinds of newlines don't
652	# count.
653	if w.tag() == word_e.Operator:
654	tok = cast(Token, w)
655	if tok.id == Id.Op_Newline:
656	for h in self.pending_here_docs:
657	_ParseHereDocBody(self.parse_ctx, h, self.line_reader,
658	self.arena)
659	del self.pending_here_docs[:] # No .clear() until Python 3.3.
660
661	self.cur_word = w
662
663	self.c_kind = word_.CommandKind(self.cur_word)
664	# Has special case for Id.Lit_{LBrace,RBrace,Equals}
665	self.c_id = word_.CommandId(self.cur_word)
666	self.next_lex_mode = lex_mode_e.Undefined
667
668	def _Eat(self, c_id, msg=None):
669	# type: (Id_t, Optional[str]) -> word_t
670	"""Consume a word of a type, maybe showing a custom error message.
671
672	Args:
673	c_id: the Id we expected
674	msg: improved error message
675	"""
676	self._GetWord()
677	if self.c_id != c_id:
678	if msg is None:
679	msg = 'Expected word type %s, got %s' % (
680	ui.PrettyId(c_id), ui.PrettyId(self.c_id))
681	p_die(msg, loc.Word(self.cur_word))
682
683	skipped = self.cur_word
684	self._SetNext()
685	return skipped
686
687	def _NewlineOk(self):
688	# type: () -> None
689	"""Check for optional newline and consume it."""
690	self._GetWord()
691	if self.c_id == Id.Op_Newline:
692	self._SetNext()
693
694	def _AtSecondaryKeyword(self):
695	# type: () -> bool
696	self._GetWord()
697	if self.c_id in SECONDARY_KEYWORDS:
698	return True
699	return False
700
701	def ParseRedirect(self):
702	# type: () -> Redir
703	self._GetWord()
704	assert self.c_kind == Kind.Redir, self.cur_word
705	op_tok = cast(Token, self.cur_word) # for MyPy
706
707	# Note: the lexer could take distinguish between
708	# >out
709	# 3>out
710	# {fd}>out
711	#
712	# which would make the code below faster. But small string optimization
713	# would also speed it up, since redirects are small.
714
715	# One way to do this is with Kind.Redir and Kind.RedirNamed, and then
716	# possibly "unify" the IDs by subtracting a constant like 8 or 16?
717
718	op_val = lexer.TokenVal(op_tok)
719	if op_val[0] == '{':
720	pos = op_val.find('}')
721	assert pos != -1 # lexer ensures this
722	where = redir_loc.VarName(op_val[1:pos]) # type: redir_loc_t
723
724	elif op_val[0].isdigit():
725	pos = 1
726	if op_val[1].isdigit():
727	pos = 2
728	where = redir_loc.Fd(int(op_val[:pos]))
729
730	else:
731	where = redir_loc.Fd(consts.RedirDefaultFd(op_tok.id))
732
733	self._SetNext()
734
735	self._GetWord()
736	# Other redirect
737	if self.c_kind != Kind.Word:
738	p_die('Invalid token after redirect operator',
739	loc.Word(self.cur_word))
740
741	# Here doc
742	if op_tok.id in (Id.Redir_DLess, Id.Redir_DLessDash):
743	arg = redir_param.HereDoc.CreateNull()
744	arg.here_begin = self.cur_word
745	arg.stdin_parts = []
746
747	r = Redir(op_tok, where, arg)
748
749	self.pending_here_docs.append(r) # will be filled on next newline.
750
751	self._SetNext()
752	return r
753
754	arg_word = self.cur_word
755	tilde = word_.TildeDetect(arg_word)
756	if tilde:
757	arg_word = tilde
758	self._SetNext()
759
760	# We should never get Empty, Token, etc.
761	assert arg_word.tag() == word_e.Compound, arg_word
762	return Redir(op_tok, where, cast(CompoundWord, arg_word))
763
764	def _ParseRedirectList(self):
765	# type: () -> List[Redir]
766	"""Try parsing any redirects at the cursor.
767
768	This is used for blocks only, not commands.
769	"""
770	redirects = [] # type: List[Redir]
771	while True:
772	# This prediction needs to ONLY accept redirect operators. Should we
773	# make them a separate Kind?
774	self._GetWord()
775	if self.c_kind != Kind.Redir:
776	break
777
778	node = self.ParseRedirect()
779	redirects.append(node)
780	self._SetNext()
781
782	return redirects
783
784	def _MaybeParseRedirectList(self, node):
785	# type: (command_t) -> command_t
786	"""Try parsing redirects at the current position.
787
788	If there are any, then wrap the command_t argument with a
789	command.Redirect node. Otherwise, return argument unchanged.
790	"""
791	self._GetWord()
792	if self.c_kind != Kind.Redir:
793	return node
794
795	redirects = [self.ParseRedirect()]
796
797	while True:
798	# This prediction needs to ONLY accept redirect operators. Should we
799	# make them a separate Kind?
800	self._GetWord()
801	if self.c_kind != Kind.Redir:
802	break
803
804	redirects.append(self.ParseRedirect())
805	self._SetNext()
806
807	return command.Redirect(node, redirects)
808
809	def _ScanSimpleCommand(self):
810	# type: () -> Tuple[List[Redir], List[CompoundWord], Optional[ArgList], Optional[LiteralBlock]]
811	"""YSH extends simple commands with typed args and blocks.
812
813	Shell has a recursive grammar, which awkwardly expresses
814	non-grammatical rules:
815
816	simple_command : cmd_prefix cmd_word cmd_suffix
817	\| cmd_prefix cmd_word
818	\| cmd_prefix
819	\| cmd_name cmd_suffix
820	\| cmd_name
821	;
822	cmd_name : WORD /* Apply rule 7a */
823	;
824	cmd_word : WORD /* Apply rule 7b */
825	;
826	cmd_prefix : io_redirect
827	\| cmd_prefix io_redirect
828	\| ASSIGNMENT_WORD
829	\| cmd_prefix ASSIGNMENT_WORD
830	;
831	cmd_suffix : io_redirect
832	\| cmd_suffix io_redirect
833	\| WORD
834	\| cmd_suffix WORD
835
836	YSH grammar:
837
838	redirect = redir_op WORD
839	item = WORD \| redirect
840
841	typed_args =
842	'(' arglist ')'
843	\| '[' arglist ']'
844
845	simple_command =
846	cmd_prefix* item+ typed_args? BraceGroup? cmd_suffix*
847
848	Notably, redirects shouldn't appear after typed args, or after
849	BraceGroup.
850
851	Examples:
852
853	This is an assignment:
854	foo=1 >out
855
856	This is a command.Simple
857	>out
858
859	What about
860	>out (42)
861	"""
862	redirects = [] # type: List[Redir]
863	words = [] # type: List[CompoundWord]
864	typed_args = None # type: Optional[ArgList]
865	block = None # type: Optional[LiteralBlock]
866
867	first_word_caps = False # does first word look like Caps, but not CAPS
868
869	i = 0
870	while True:
871	self._GetWord()
872
873	# If we got { }, change it to something that's not Kind.Word
874	kind2 = self.c_kind
875	if (kind2 == Kind.Word and self.parse_opts.parse_brace() and
876	self.c_id in (Id.Lit_LBrace, Id.Lit_RBrace)):
877	kind2 = Kind.Op
878
879	if kind2 == Kind.Redir:
880	node = self.ParseRedirect()
881	redirects.append(node)
882
883	elif kind2 == Kind.Word:
884	w = cast(CompoundWord, self.cur_word) # Kind.Word ensures this
885
886	if i == 0:
887	# Disallow leading =a because it's confusing
888	part0 = w.parts[0]
889	if part0.tag() == word_part_e.Literal:
890	tok = cast(Token, part0)
891	if tok.id == Id.Lit_Equals:
892	p_die(
893	"=word isn't allowed. Hint: add a space after =, or quote it",
894	tok)
895
896	# Is the first word a Hay Attr word?
897	#
898	# Can we remove this StaticEval() call, and just look
899	# inside Token? I think once we get rid of SHELL nodes,
900	# this will be simpler.
901
902	ok, word_str, quoted = word_.StaticEval(w)
903	# Foo { a = 1 } is OK, but not foo { a = 1 } or FOO { a = 1 }
904	if (ok and len(word_str) and word_str[0].isupper() and
905	not word_str.isupper()):
906	first_word_caps = True
907	#log('W %s', word_str)
908
909	words.append(w)
910
911	else:
912	break
913
914	self._SetNextBrack() # Allow bracket for SECOND word on
915	i += 1
916
917	# my-cmd (x) or my-cmd [x]
918	self._GetWord()
919	if self.c_id == Id.Op_LParen:
920	# 1. Check that there's a preceding space
921	prev_byte = self.lexer.ByteLookBack()
922	if prev_byte not in (SPACE_CH, TAB_CH):
923	if self.parse_opts.parse_at():
924	p_die('Space required before (', loc.Word(self.cur_word))
925	else:
926	# inline func call like @sorted(x) is invalid in OSH, but the
927	# solution isn't a space
928	p_die(
929	'Unexpected left paren (might need a space before it)',
930	loc.Word(self.cur_word))
931
932	# 2. Check that it's not (). We disallow this because it's a no-op and
933	# there could be confusion with shell func defs.
934	# For some reason we need to call lexer.LookPastSpace, not
935	# w_parser.LookPastSpace. I think this is because we're at (, which is
936	# an operator token. All the other cases are like 'x=', which is PART
937	# of a word, and we don't know if it will end.
938	next_id = self.lexer.LookPastSpace(lex_mode_e.ShCommand)
939	if next_id == Id.Op_RParen:
940	p_die('Empty arg list not allowed', loc.Word(self.cur_word))
941
942	typed_args = self.w_parser.ParseProcCallArgs(
943	grammar_nt.ysh_eager_arglist)
944
945	self._SetNext()
946
947	elif self.c_id == Id.Op_LBracket: # only when parse_bracket set
948	typed_args = self.w_parser.ParseProcCallArgs(
949	grammar_nt.ysh_lazy_arglist)
950
951	self._SetNext()
952
953	self._GetWord()
954
955	# Allow redirects after typed args, e.g.
956	# json write (x) > out.txt
957	if self.c_kind == Kind.Redir:
958	redirects.extend(self._ParseRedirectList())
959
960	# my-cmd { echo hi } my-cmd (x) { echo hi } ...
961	if (self.parse_opts.parse_brace() and self.c_id == Id.Lit_LBrace and
962	# Disabled for if/while condition, etc.
963	self.allow_block):
964
965	# allow x = 42
966	self.hay_attrs_stack.append(first_word_caps)
967	brace_group = self.ParseBraceGroup()
968
969	# So we can get the source code back later
970	lines = self.arena.SaveLinesAndDiscard(brace_group.left,
971	brace_group.right)
972	block = LiteralBlock(brace_group, lines)
973
974	self.hay_attrs_stack.pop()
975
976	self._GetWord()
977
978	# Allow redirects after block, e.g.
979	# cd /tmp { echo $PWD } > out.txt
980	if self.c_kind == Kind.Redir:
981	redirects.extend(self._ParseRedirectList())
982
983	return redirects, words, typed_args, block
984
985	def _MaybeExpandAliases(self, words):
986	# type: (List[CompoundWord]) -> Optional[command_t]
987	"""Try to expand aliases.
988
989	Args:
990	words: A list of Compound
991
992	Returns:
993	A new LST node, or None.
994
995	Our implementation of alias has two design choices:
996	- Where to insert it in parsing. We do it at the end of ParseSimpleCommand.
997	- What grammar rule to parse the expanded alias buffer with. In our case
998	it's ParseCommand().
999
1000	This doesn't quite match what other shells do, but I can't figure out a
1001	better places.
1002
1003	Most test cases pass, except for ones like:
1004
1005	alias LBRACE='{'
1006	LBRACE echo one; echo two; }
1007
1008	alias MULTILINE='echo 1
1009	echo 2
1010	echo 3'
1011	MULTILINE
1012
1013	NOTE: dash handles aliases in a totally different way. It has a global
1014	variable checkkwd in parser.c. It assigns it all over the grammar, like
1015	this:
1016
1017	checkkwd = CHKNL \| CHKKWD \| CHKALIAS;
1018
1019	The readtoken() function checks (checkkwd & CHKALIAS) and then calls
1020	lookupalias(). This seems to provide a consistent behavior among shells,
1021	but it's less modular and testable.
1022
1023	Bash also uses a global 'parser_state & PST_ALEXPNEXT'.
1024
1025	Returns:
1026	A command node if any aliases were expanded, or None otherwise.
1027	"""
1028	# Start a new list if there aren't any. This will be passed recursively
1029	# through CommandParser instances.
1030	aliases_in_flight = (self.aliases_in_flight
1031	if len(self.aliases_in_flight) else [])
1032
1033	# for error message
1034	first_word_str = None # type: Optional[str]
1035	argv0_loc = loc.Word(words[0])
1036
1037	expanded = [] # type: List[str]
1038	i = 0
1039	n = len(words)
1040
1041	while i < n:
1042	w = words[i]
1043
1044	ok, word_str, quoted = word_.StaticEval(w)
1045	if not ok or quoted:
1046	break
1047
1048	alias_exp = self.aliases.get(word_str)
1049	if alias_exp is None:
1050	break
1051
1052	# Prevent infinite loops. This is subtle: we want to prevent infinite
1053	# expansion of alias echo='echo x'. But we don't want to prevent
1054	# expansion of the second word in 'echo echo', so we add 'i' to
1055	# "aliases_in_flight".
1056	if (word_str, i) in aliases_in_flight:
1057	break
1058
1059	if i == 0:
1060	first_word_str = word_str # for error message
1061
1062	#log('%r -> %r', word_str, alias_exp)
1063	aliases_in_flight.append((word_str, i))
1064	expanded.append(alias_exp)
1065	i += 1
1066
1067	if not alias_exp.endswith(' '):
1068	# alias e='echo [ ' is the same expansion as
1069	# alias e='echo ['
1070	# The trailing space indicates whether we should continue to expand
1071	# aliases; it's not part of it.
1072	expanded.append(' ')
1073	break # No more expansions
1074
1075	if len(expanded) == 0: # No expansions; caller does parsing.
1076	return None
1077
1078	# We are expanding an alias, so copy the rest of the words and re-parse.
1079	if i < n:
1080	left_tok = location.LeftTokenForWord(words[i])
1081	right_tok = location.RightTokenForWord(words[-1])
1082
1083	# OLD CONSTRAINT
1084	#assert left_tok.line_id == right_tok.line_id
1085
1086	words_str = self.arena.SnipCodeString(left_tok, right_tok)
1087	expanded.append(words_str)
1088
1089	code_str = ''.join(expanded)
1090
1091	# TODO:
1092	# Aliases break static parsing (like backticks), so use our own Arena.
1093	# This matters for Hay, which calls SaveLinesAndDiscard().
1094	# arena = alloc.Arena()
1095	arena = self.arena
1096
1097	line_reader = reader.StringLineReader(code_str, arena)
1098	cp = self.parse_ctx.MakeOshParser(line_reader)
1099	cp.Init_AliasesInFlight(aliases_in_flight)
1100
1101	# break circular dep
1102	from frontend import parse_lib
1103
1104	# The interaction between COMPLETION and ALIASES requires special care.
1105	# See docstring of BeginAliasExpansion() in parse_lib.py.
1106	src = source.Alias(first_word_str, argv0_loc)
1107	with alloc.ctx_SourceCode(arena, src):
1108	with parse_lib.ctx_Alias(self.parse_ctx.trail):
1109	try:
1110	# _ParseCommandTerm() handles multiline commands, compound
1111	# commands, etc. as opposed to ParseLogicalLine()
1112	node = cp._ParseCommandTerm()
1113	except error.Parse as e:
1114	# Failure to parse alias expansion is a fatal error
1115	# We don't need more handling here/
1116	raise
1117
1118	if 0:
1119	log('AFTER expansion:')
1120	node.PrettyPrint()
1121
1122	return node
1123
1124	def ParseSimpleCommand(self):
1125	# type: () -> command_t
1126	"""Fixed transcription of the POSIX grammar
1127
1128	io_file : '<' filename
1129	\| LESSAND filename
1130	...
1131
1132	io_here : DLESS here_end
1133	\| DLESSDASH here_end
1134
1135	redirect : IO_NUMBER (io_redirect \| io_here)
1136
1137	prefix_part : ASSIGNMENT_WORD \| redirect
1138	cmd_part : WORD \| redirect
1139
1140	assign_kw : Declare \| Export \| Local \| Readonly
1141
1142	# Without any words it is parsed as a command, not an assignment
1143	assign_listing : assign_kw
1144
1145	# Now we have something to do (might be changing assignment flags too)
1146	# NOTE: any prefixes should be a warning, but they are allowed in shell.
1147	assignment : prefix_part* assign_kw (WORD \| ASSIGNMENT_WORD)+
1148
1149	# an external command, a function call, or a builtin -- a "word_command"
1150	word_command : prefix_part* cmd_part+
1151
1152	simple_command : assign_listing
1153	\| assignment
1154	\| proc_command
1155
1156	Simple imperative algorithm:
1157
1158	1) Read a list of words and redirects. Append them to separate lists.
1159	2) Look for the first non-assignment word. If it's declare, etc., then
1160	keep parsing words AND assign words. Otherwise, just parse words.
1161	3) If there are no non-assignment words, then it's a global assignment.
1162
1163	{ redirects, global assignments } OR
1164	{ redirects, prefix_bindings, words } OR
1165	{ redirects, ERROR_prefix_bindings, keyword, assignments, words }
1166
1167	THEN CHECK that prefix bindings don't have any array literal parts!
1168	global assignment and keyword assignments can have the of course.
1169	well actually EXPORT shouldn't have them either -- WARNING
1170
1171	3 cases we want to warn: prefix_bindings for assignment, and array literal
1172	in prefix bindings, or export
1173
1174	A command can be an assignment word, word, or redirect on its own.
1175
1176	ls
1177	>out.txt
1178
1179	>out.txt FOO=bar # this touches the file
1180
1181	Or any sequence:
1182	ls foo bar
1183	<in.txt ls foo bar >out.txt
1184	<in.txt ls >out.txt foo bar
1185
1186	Or add one or more environment bindings:
1187	VAR=val env
1188	>out.txt VAR=val env
1189
1190	here_end vs filename is a matter of whether we test that it's quoted. e.g.
1191	<<EOF vs <<'EOF'.
1192	"""
1193	redirects, words, typed_args, block = self._ScanSimpleCommand()
1194
1195	typed_loc = None # type: Optional[Token]
1196	if block:
1197	typed_loc = block.brace_group.left
1198	if typed_args:
1199	typed_loc = typed_args.left # preferred over block location
1200
1201	if len(words) == 0: # e.g. >out.txt # redirect without words
1202	assert len(redirects) != 0
1203	if typed_loc is not None:
1204	p_die("Unexpected typed args", typed_loc)
1205	return command.Redirect(command.NoOp, redirects)
1206
1207	preparsed_list, suffix_words = _SplitSimpleCommandPrefix(words)
1208	if len(preparsed_list):
1209	# Disallow X=Y inside proc and func
1210	# and inside Hay Attr blocks
1211	# But allow X=Y at the top level
1212	# for interactive use foo=bar
1213	# for global constants GLOBAL=~/src
1214	# because YSH assignment doesn't have tilde sub
1215	if len(suffix_words) == 0:
1216	if (self.cmd_mode != cmd_mode_e.Shell or
1217	(len(self.hay_attrs_stack) and self.hay_attrs_stack[-1])):
1218	p_die('Use var/setvar to assign in YSH',
1219	preparsed_list[0].left)
1220
1221	# Set a reference to words and redirects for completion. We want to
1222	# inspect this state after a failed parse.
1223	self.parse_ctx.trail.SetLatestWords(suffix_words, redirects)
1224
1225	if len(suffix_words) == 0:
1226	if typed_loc is not None:
1227	p_die("Unexpected typed args", typed_loc)
1228
1229	# ShAssignment: No suffix words like ONE=1 a[x]=1 TWO=2
1230	pairs = [] # type: List[AssignPair]
1231	for preparsed in preparsed_list:
1232	pairs.append(
1233	_MakeAssignPair(self.parse_ctx, preparsed, self.arena))
1234
1235	left_tok = location.LeftTokenForCompoundWord(words[0])
1236	assign_node = command.ShAssignment(left_tok, pairs)
1237	if len(redirects):
1238	return command.Redirect(assign_node, redirects)
1239	else:
1240	return assign_node
1241
1242	kind, kw_token = word_.IsControlFlow(suffix_words[0])
1243
1244	if kind == Kind.ControlFlow:
1245	if not self.parse_opts.parse_ignored() and len(redirects):
1246	p_die("Control flow shouldn't have redirects", kw_token)
1247	if len(preparsed_list): # FOO=bar local spam=eggs not allowed
1248	p_die("Control flow shouldn't have environment bindings",
1249	preparsed_list[0].left)
1250
1251	if kw_token.id == Id.ControlFlow_Return:
1252	# return x - inside procs and shell functions
1253	# return (x) - inside funcs
1254	if typed_args is None:
1255	if self.cmd_mode not in (cmd_mode_e.Shell,
1256	cmd_mode_e.Proc):
1257	p_die('Shell-style returns not allowed here', kw_token)
1258	else:
1259	if self.cmd_mode != cmd_mode_e.Func:
1260	p_die('Typed return is only allowed inside func',
1261	typed_loc)
1262	if len(typed_args.pos_args) != 1:
1263	p_die("Typed return expects one argument", typed_loc)
1264	if len(typed_args.named_args) != 0:
1265	p_die("Typed return doesn't take named arguments",
1266	typed_loc)
1267	return command.Retval(kw_token, typed_args.pos_args[0])
1268
1269	# Except for return (x), we shouldn't have typed args
1270	if typed_loc is not None:
1271	p_die("Unexpected typed args", typed_loc)
1272
1273	# Attach the token for errors. (ShAssignment may not need it.)
1274	if len(suffix_words) == 1:
1275	arg_word = None # type: Optional[word_t]
1276	elif len(suffix_words) == 2:
1277	arg_word = suffix_words[1]
1278	else:
1279	p_die('Unexpected argument to %r' % lexer.TokenVal(kw_token),
1280	loc.Word(suffix_words[2]))
1281
1282	return command.ControlFlow(kw_token, arg_word)
1283
1284	# Alias expansion only understands words, not typed args ( ) or block { }
1285	if not typed_args and not block and self.parse_opts.expand_aliases():
1286	# If any expansions were detected, then parse again.
1287	expanded_node = self._MaybeExpandAliases(suffix_words)
1288	if expanded_node:
1289	# Attach env bindings and redirects to the expanded node.
1290	more_env = [] # type: List[EnvPair]
1291	_AppendMoreEnv(preparsed_list, more_env)
1292	exp = command.ExpandedAlias(expanded_node, more_env)
1293	if len(redirects):
1294	return command.Redirect(exp, redirects)
1295	else:
1296	return exp
1297
1298	# TODO: check that we don't have env1=x x[1]=y env2=z here.
1299
1300	# FOO=bar printenv.py FOO
1301	node = _MakeSimpleCommand(preparsed_list, suffix_words, typed_args,
1302	block)
1303	if len(redirects):
1304	return command.Redirect(node, redirects)
1305	else:
1306	return node
1307
1308	def ParseBraceGroup(self):
1309	# type: () -> BraceGroup
1310	"""
1311	Original:
1312	brace_group : LBrace command_list RBrace ;
1313
1314	YSH:
1315	brace_group : LBrace (Op_Newline IgnoredComment?)? command_list RBrace ;
1316
1317	The doc comment can only occur if there's a newline.
1318	"""
1319	ate = self._Eat(Id.Lit_LBrace)
1320	left = word_.BraceToken(ate)
1321
1322	doc_word = None # type: word_t
1323	self._GetWord()
1324	if self.c_id == Id.Op_Newline:
1325	self._SetNext()
1326	# Set a flag so we don't skip over ###
1327	with word_.ctx_EmitDocToken(self.w_parser):
1328	self._GetWord()
1329
1330	if self.c_id == Id.Ignored_Comment:
1331	doc_word = self.cur_word
1332	self._SetNext()
1333
1334	# Id.Ignored_Comment means it's a Token, or None
1335	doc_token = cast(Token, doc_word)
1336
1337	c_list = self._ParseCommandList()
1338
1339	ate = self._Eat(Id.Lit_RBrace)
1340	right = word_.BraceToken(ate)
1341
1342	# Note(andychu): Related ASDL bug #1216. Choosing the Python [] behavior
1343	# would allow us to revert this back to None, which was changed in
1344	# https://github.com/oilshell/oil/pull/1211. Choosing the C++ nullptr
1345	# behavior saves allocations, but is less type safe.
1346	return BraceGroup(left, doc_token, c_list.children, right)
1347
1348	def ParseDoGroup(self):
1349	# type: () -> command.DoGroup
1350	"""Used by ForEach, ForExpr, While, Until. Should this be a Do node?
1351
1352	do_group : Do command_list Done ; /* Apply rule 6 */
1353	"""
1354	ate = self._Eat(Id.KW_Do)
1355	do_kw = word_.AsKeywordToken(ate)
1356
1357	c_list = self._ParseCommandList() # could be anything
1358
1359	ate = self._Eat(Id.KW_Done)
1360	done_kw = word_.AsKeywordToken(ate)
1361
1362	return command.DoGroup(do_kw, c_list.children, done_kw)
1363
1364	def ParseForWords(self):
1365	# type: () -> Tuple[List[CompoundWord], Optional[Token]]
1366	"""
1367	for_words : WORD* for_sep
1368	;
1369	for_sep : ';' newline_ok
1370	\| NEWLINES
1371	;
1372	"""
1373	words = [] # type: List[CompoundWord]
1374	# The token of any semi-colon, so we can remove it.
1375	semi_tok = None # type: Optional[Token]
1376
1377	while True:
1378	self._GetWord()
1379	if self.c_id == Id.Op_Semi:
1380	tok = cast(Token, self.cur_word)
1381	semi_tok = tok
1382	self._SetNext()
1383	self._NewlineOk()
1384	break
1385	elif self.c_id == Id.Op_Newline:
1386	self._SetNext()
1387	break
1388	elif self.parse_opts.parse_brace() and self.c_id == Id.Lit_LBrace:
1389	break
1390
1391	if self.cur_word.tag() != word_e.Compound:
1392	# TODO: Can we also show a pointer to the 'for' keyword?
1393	p_die('Invalid word in for loop', loc.Word(self.cur_word))
1394
1395	w2 = cast(CompoundWord, self.cur_word)
1396	words.append(w2)
1397	self._SetNext()
1398	return words, semi_tok
1399
1400	def _ParseForExprLoop(self, for_kw):
1401	# type: (Token) -> command.ForExpr
1402	"""
1403	Shell:
1404	for '((' init ';' cond ';' update '))' for_sep? do_group
1405
1406	YSH:
1407	for '((' init ';' cond ';' update '))' for_sep? brace_group
1408	"""
1409	node = self.w_parser.ReadForExpression()
1410	node.keyword = for_kw
1411
1412	self._SetNext()
1413
1414	self._GetWord()
1415	if self.c_id == Id.Op_Semi:
1416	self._SetNext()
1417	self._NewlineOk()
1418	elif self.c_id == Id.Op_Newline:
1419	self._SetNext()
1420	elif self.c_id == Id.KW_Do: # missing semicolon/newline allowed
1421	pass
1422	elif self.c_id == Id.Lit_LBrace: # does NOT require parse_brace
1423	pass
1424	else:
1425	p_die('Invalid word after for expression', loc.Word(self.cur_word))
1426
1427	if self.c_id == Id.Lit_LBrace:
1428	node.body = self.ParseBraceGroup()
1429	else:
1430	node.body = self.ParseDoGroup()
1431	return node
1432
1433	def _ParseForEachLoop(self, for_kw):
1434	# type: (Token) -> command.ForEach
1435	node = command.ForEach.CreateNull(alloc_lists=True)
1436	node.keyword = for_kw
1437
1438	num_iter_names = 0
1439	while True:
1440	w = self.cur_word
1441
1442	# Hack that makes the language more familiar:
1443	# - 'x, y' is accepted, but not 'x,y' or 'x ,y'
1444	# - 'x y' is also accepted but not idiomatic.
1445	UP_w = w
1446	if w.tag() == word_e.Compound:
1447	w = cast(CompoundWord, UP_w)
1448	if word_.LiteralId(w.parts[-1]) == Id.Lit_Comma:
1449	w.parts.pop()
1450
1451	ok, iter_name, quoted = word_.StaticEval(w)
1452	if not ok or quoted: # error: for $x
1453	p_die('Expected loop variable (a constant word)', loc.Word(w))
1454
1455	if not match.IsValidVarName(iter_name): # error: for -
1456	# TODO: consider commas?
1457	if ',' in iter_name:
1458	p_die('Loop variables look like x, y (fix spaces)',
1459	loc.Word(w))
1460	p_die('Invalid loop variable name %r' % iter_name, loc.Word(w))
1461
1462	node.iter_names.append(iter_name)
1463	num_iter_names += 1
1464	self._SetNext()
1465
1466	self._GetWord()
1467	# 'in' or 'do' or ';' or Op_Newline marks the end of variable names
1468	# Subtlety: 'var' is KW_Var and is a valid loop name
1469	if self.c_id in (Id.KW_In, Id.KW_Do) or self.c_kind == Kind.Op:
1470	break
1471
1472	if num_iter_names == 3:
1473	p_die('Unexpected word after 3 loop variables',
1474	loc.Word(self.cur_word))
1475
1476	self._NewlineOk()
1477
1478	self._GetWord()
1479	if self.c_id == Id.KW_In:
1480	# Ideally we would want ( not 'in'. But we still have to fix the bug
1481	# where we require a SPACE between in and (
1482	# for x in(y) # should be accepted, but isn't
1483
1484	expr_blame = word_.AsKeywordToken(self.cur_word)
1485
1486	self._SetNext() # skip in
1487	if self.w_parser.LookPastSpace() == Id.Op_LParen:
1488	enode = self.w_parser.ParseYshExprForCommand()
1489	node.iterable = for_iter.YshExpr(enode, expr_blame)
1490
1491	# For simplicity, we don't accept for x in (obj); do ...
1492	self._GetWord()
1493	if self.c_id != Id.Lit_LBrace:
1494	p_die('Expected { after iterable expression',
1495	loc.Word(self.cur_word))
1496	else:
1497	semi_tok = None # type: Optional[Token]
1498	iter_words, semi_tok = self.ParseForWords()
1499	node.semi_tok = semi_tok
1500
1501	if not self.parse_opts.parse_bare_word() and len(
1502	iter_words) == 1:
1503	ok, s, quoted = word_.StaticEval(iter_words[0])
1504	if ok and match.IsValidVarName(s) and not quoted:
1505	p_die(
1506	'Surround this word with either parens or quotes (parse_bare_word)',
1507	loc.Word(iter_words[0]))
1508
1509	words2 = braces.BraceDetectAll(iter_words)
1510	words3 = word_.TildeDetectAll(words2)
1511	node.iterable = for_iter.Words(words3)
1512
1513	# Now that we know there are words, do an extra check
1514	if num_iter_names > 2:
1515	p_die('Expected at most 2 loop variables', for_kw)
1516
1517	elif self.c_id == Id.KW_Do:
1518	node.iterable = for_iter.Args # implicitly loop over "$@"
1519	# do not advance
1520
1521	elif self.c_id == Id.Op_Semi: # for x; do
1522	node.iterable = for_iter.Args # implicitly loop over "$@"
1523	self._SetNext()
1524
1525	else: # for foo BAD
1526	p_die('Unexpected word after for loop variable',
1527	loc.Word(self.cur_word))
1528
1529	self._GetWord()
1530	if self.c_id == Id.Lit_LBrace: # parse_opts.parse_brace() must be on
1531	node.body = self.ParseBraceGroup()
1532	else:
1533	node.body = self.ParseDoGroup()
1534
1535	return node
1536
1537	def ParseFor(self):
1538	# type: () -> command_t
1539	"""
1540	TODO: Update the grammar
1541
1542	for_clause : For for_name newline_ok (in for_words? for_sep)? do_group ;
1543	\| For '((' ... TODO
1544	"""
1545	ate = self._Eat(Id.KW_For)
1546	for_kw = word_.AsKeywordToken(ate)
1547
1548	self._GetWord()
1549	if self.c_id == Id.Op_DLeftParen:
1550	if not self.parse_opts.parse_dparen():
1551	p_die("Bash for loops aren't allowed (parse_dparen)",
1552	loc.Word(self.cur_word))
1553
1554	# for (( i = 0; i < 10; i++)
1555	n1 = self._ParseForExprLoop(for_kw)
1556	return self._MaybeParseRedirectList(n1)
1557	else:
1558	# for x in a b; do echo hi; done
1559	n2 = self._ParseForEachLoop(for_kw)
1560	return self._MaybeParseRedirectList(n2)
1561
1562	def _ParseConditionList(self):
1563	# type: () -> condition_t
1564	"""
1565	condition_list: command_list
1566
1567	This is a helper to parse a condition list for if commands and while/until
1568	loops. It will throw a parse error if there are no conditions in the list.
1569	"""
1570	self.allow_block = False
1571	commands = self._ParseCommandList()
1572	self.allow_block = True
1573
1574	if len(commands.children) == 0:
1575	p_die("Expected a condition", loc.Word(self.cur_word))
1576
1577	return condition.Shell(commands.children)
1578
1579	def ParseWhileUntil(self, keyword):
1580	# type: (Token) -> command.WhileUntil
1581	"""
1582	while_clause : While command_list do_group ;
1583	until_clause : Until command_list do_group ;
1584	"""
1585	self._SetNext() # skip keyword
1586
1587	if (self.parse_opts.parse_paren() and
1588	self.w_parser.LookPastSpace() == Id.Op_LParen):
1589	enode = self.w_parser.ParseYshExprForCommand()
1590	cond = condition.YshExpr(enode) # type: condition_t
1591	else:
1592	cond = self._ParseConditionList()
1593
1594	# NOTE: The LSTs will be different for OSH and YSH, but the execution
1595	# should be unchanged. To be sure we should desugar.
1596	self._GetWord()
1597	if self.parse_opts.parse_brace() and self.c_id == Id.Lit_LBrace:
1598	# while test -f foo {
1599	body_node = self.ParseBraceGroup() # type: command_t
1600	else:
1601	body_node = self.ParseDoGroup()
1602
1603	return command.WhileUntil(keyword, cond, body_node)
1604
1605	def ParseCaseArm(self):
1606	# type: () -> CaseArm
1607	"""
1608	case_item: '('? pattern ('\|' pattern)* ')'
1609	newline_ok command_term? trailer? ;
1610
1611	Looking at '(' or pattern
1612	"""
1613	self.lexer.PushHint(Id.Op_RParen, Id.Right_CasePat)
1614
1615	left_tok = location.LeftTokenForWord(self.cur_word) # ( or pat
1616
1617	if self.c_id == Id.Op_LParen: # Optional (
1618	self._SetNext()
1619
1620	pat_words = [] # type: List[word_t]
1621	while True:
1622	self._GetWord()
1623	if self.c_kind != Kind.Word:
1624	p_die('Expected case pattern', loc.Word(self.cur_word))
1625	pat_words.append(self.cur_word)
1626	self._SetNext()
1627
1628	self._GetWord()
1629	if self.c_id == Id.Op_Pipe:
1630	self._SetNext()
1631	else:
1632	break
1633
1634	ate = self._Eat(Id.Right_CasePat)
1635	middle_tok = word_.AsOperatorToken(ate)
1636
1637	self._NewlineOk()
1638
1639	self._GetWord()
1640	if self.c_id not in (Id.Op_DSemi, Id.Op_SemiAmp, Id.Op_DSemiAmp,
1641	Id.KW_Esac):
1642	c_list = self._ParseCommandTerm()
1643	action_children = c_list.children
1644	else:
1645	action_children = []
1646
1647	dsemi_tok = None # type: Token
1648	self._GetWord()
1649	if self.c_id == Id.KW_Esac: # missing last ;;
1650	pass
1651	elif self.c_id in (Id.Op_DSemi, Id.Op_SemiAmp, Id.Op_DSemiAmp):
1652	dsemi_tok = word_.AsOperatorToken(self.cur_word)
1653	self._SetNext()
1654	else:
1655	# Happens on EOF
1656	p_die('Expected ;; or esac', loc.Word(self.cur_word))
1657
1658	self._NewlineOk()
1659
1660	return CaseArm(left_tok, pat.Words(pat_words), middle_tok,
1661	action_children, dsemi_tok)
1662
1663	def ParseYshCaseArm(self, discriminant):
1664	# type: (Id_t) -> CaseArm
1665	"""
1666	case_item : pattern newline_ok brace_group newline_ok
1667	pattern : pat_words
1668	\| pat_exprs
1669	\| pat_eggex
1670	\| pat_else
1671	pat_words : pat_word (newline_ok '\|' newline_ok pat_word)*
1672	pat_exprs : pat_expr (newline_ok '\|' newline_ok pat_expr)*
1673	pat_word : WORD
1674	pat_eggex : '/' oil_eggex '/'
1675	pat_expr : '(' oil_expr ')'
1676	pat_else : '(' Id.KW_Else ')'
1677
1678	Looking at: 'pattern'
1679
1680	Note that the trailing `newline_ok` in `case_item` is handled by
1681	`ParseYshCase`. We do this because parsing that `newline_ok` returns
1682	the next "discriminant" for the next token, so it makes more sense to
1683	handle it there.
1684	"""
1685	left_tok = None # type: Token
1686	pattern = None # type: pat_t
1687
1688	if discriminant in (Id.Op_LParen, Id.Arith_Slash):
1689	# pat_exprs, pat_else or pat_eggex
1690	pattern, left_tok = self.w_parser.ParseYshCasePattern()
1691	else:
1692	# pat_words
1693	pat_words = [] # type: List[word_t]
1694	while True:
1695	self._GetWord()
1696	if self.c_kind != Kind.Word:
1697	p_die('Expected case pattern', loc.Word(self.cur_word))
1698	pat_words.append(self.cur_word)
1699	self._SetNext()
1700
1701	if not left_tok:
1702	left_tok = location.LeftTokenForWord(self.cur_word)
1703
1704	self._NewlineOk()
1705
1706	self._GetWord()
1707	if self.c_id == Id.Op_Pipe:
1708	self._SetNext()
1709	self._NewlineOk()
1710	else:
1711	break
1712	pattern = pat.Words(pat_words)
1713
1714	self._NewlineOk()
1715	action = self.ParseBraceGroup()
1716
1717	# The left token of the action is our "middle" token
1718	return CaseArm(left_tok, pattern, action.left, action.children,
1719	action.right)
1720
1721	def ParseYshCase(self, case_kw):
1722	# type: (Token) -> command.Case
1723	"""
1724	ysh_case : Case '(' expr ')' LBrace newline_ok ysh_case_arm* RBrace ;
1725
1726	Looking at: token after 'case'
1727	"""
1728	enode = self.w_parser.ParseYshExprForCommand()
1729	to_match = case_arg.YshExpr(enode)
1730
1731	ate = self._Eat(Id.Lit_LBrace)
1732	arms_start = word_.BraceToken(ate)
1733
1734	discriminant = self.w_parser.NewlineOkForYshCase()
1735
1736	# Note: for now, zero arms are accepted, just like POSIX case $x in esac
1737	arms = [] # type: List[CaseArm]
1738	while discriminant != Id.Op_RBrace:
1739	arm = self.ParseYshCaseArm(discriminant)
1740	arms.append(arm)
1741
1742	discriminant = self.w_parser.NewlineOkForYshCase()
1743
1744	# NewlineOkForYshCase leaves the lexer in lex_mode_e.Expr. So the '}'
1745	# token is read as an Id.Op_RBrace, but we need to store this as a
1746	# Id.Lit_RBrace.
1747	ate = self._Eat(Id.Op_RBrace)
1748	arms_end = word_.AsOperatorToken(ate)
1749	arms_end.id = Id.Lit_RBrace
1750
1751	return command.Case(case_kw, to_match, arms_start, arms, arms_end)
1752
1753	def ParseOldCase(self, case_kw):
1754	# type: (Token) -> command.Case
1755	"""
1756	case_clause : Case WORD newline_ok In newline_ok case_arm* Esac ;
1757
1758	-> Looking at WORD
1759
1760	FYI original POSIX case list, which takes pains for DSEMI
1761
1762	case_list: case_item (DSEMI newline_ok case_item)* DSEMI? newline_ok;
1763	"""
1764	self._GetWord()
1765	w = self.cur_word
1766	if not self.parse_opts.parse_bare_word():
1767	ok, s, quoted = word_.StaticEval(w)
1768	if ok and not quoted:
1769	p_die(
1770	"This is a constant string. You may want a variable like $x (parse_bare_word)",
1771	loc.Word(w))
1772
1773	if w.tag() != word_e.Compound:
1774	p_die("Expected a word to match against", loc.Word(w))
1775
1776	to_match = case_arg.Word(w)
1777	self._SetNext() # past WORD
1778
1779	self._NewlineOk()
1780
1781	ate = self._Eat(Id.KW_In)
1782	arms_start = word_.AsKeywordToken(ate)
1783
1784	self._NewlineOk()
1785
1786	arms = [] # type: List[CaseArm]
1787	while True:
1788	self._GetWord()
1789	if self.c_id == Id.KW_Esac:
1790	break
1791	# case arm should begin with a pattern word or (
1792	if self.c_kind != Kind.Word and self.c_id != Id.Op_LParen:
1793	break
1794
1795	arm = self.ParseCaseArm()
1796	arms.append(arm)
1797
1798	ate = self._Eat(Id.KW_Esac)
1799	arms_end = word_.AsKeywordToken(ate)
1800
1801	# no redirects yet
1802	return command.Case(case_kw, to_match, arms_start, arms, arms_end)
1803
1804	def ParseCase(self):
1805	# type: () -> command.Case
1806	"""
1807	case_clause : old_case # from POSIX
1808	\| ysh_case
1809	;
1810
1811	Looking at 'Case'
1812	"""
1813	case_kw = word_.AsKeywordToken(self.cur_word)
1814	self._SetNext() # past 'case'
1815
1816	if self.w_parser.LookPastSpace() == Id.Op_LParen:
1817	return self.ParseYshCase(case_kw)
1818	else:
1819	return self.ParseOldCase(case_kw)
1820
1821	def _ParseYshElifElse(self, if_node):
1822	# type: (command.If) -> None
1823	"""If test -f foo { echo foo.
1824
1825	} elif test -f bar; test -f spam { ^ we parsed up to here echo
1826	bar } else { echo none }
1827	"""
1828	arms = if_node.arms
1829
1830	while self.c_id == Id.KW_Elif:
1831	elif_kw = word_.AsKeywordToken(self.cur_word)
1832	self._SetNext() # skip elif
1833	if (self.parse_opts.parse_paren() and
1834	self.w_parser.LookPastSpace() == Id.Op_LParen):
1835	enode = self.w_parser.ParseYshExprForCommand()
1836	cond = condition.YshExpr(enode) # type: condition_t
1837	else:
1838	self.allow_block = False
1839	commands = self._ParseCommandList()
1840	self.allow_block = True
1841	cond = condition.Shell(commands.children)
1842
1843	body = self.ParseBraceGroup()
1844	self._GetWord()
1845
1846	arm = IfArm(elif_kw, cond, None, body.children, None)
1847	arms.append(arm)
1848
1849	self._GetWord()
1850	if self.c_id == Id.KW_Else:
1851	self._SetNext()
1852	body = self.ParseBraceGroup()
1853	if_node.else_action = body.children
1854
1855	def _ParseYshIf(self, if_kw, cond):
1856	# type: (Token, condition_t) -> command.If
1857	"""
1858	if test -f foo {
1859	# ^ we parsed up to here
1860	echo foo
1861	} elif test -f bar; test -f spam {
1862	echo bar
1863	} else {
1864	echo none
1865	}
1866	NOTE: If you do something like if test -n foo{, the parser keeps going, and
1867	the error is confusing because it doesn't point to the right place.
1868
1869	I think we might need strict_brace so that foo{ is disallowed. It has to
1870	be foo\{ or foo{a,b}. Or just turn that on with parse_brace? After you
1871	form ANY CompoundWord, make sure it's balanced for Lit_LBrace and
1872	Lit_RBrace? Maybe this is pre-parsing step in the WordParser?
1873	"""
1874	if_node = command.If.CreateNull(alloc_lists=True)
1875	if_node.if_kw = if_kw
1876
1877	body1 = self.ParseBraceGroup()
1878	# Every arm has 1 spid, unlike shell-style
1879	# TODO: We could get the spids from the brace group.
1880	arm = IfArm(if_kw, cond, None, body1.children, None)
1881
1882	if_node.arms.append(arm)
1883
1884	self._GetWord()
1885	if self.c_id in (Id.KW_Elif, Id.KW_Else):
1886	self._ParseYshElifElse(if_node)
1887	# the whole if node has the 'else' spid, unlike shell-style there's no 'fi'
1888	# spid because that's in the BraceGroup.
1889	return if_node
1890
1891	def _ParseElifElse(self, if_node):
1892	# type: (command.If) -> None
1893	"""
1894	else_part: (Elif command_list Then command_list)* Else command_list ;
1895	"""
1896	arms = if_node.arms
1897
1898	self._GetWord()
1899	while self.c_id == Id.KW_Elif:
1900	elif_kw = word_.AsKeywordToken(self.cur_word)
1901	self._SetNext() # past 'elif'
1902
1903	cond = self._ParseConditionList()
1904
1905	ate = self._Eat(Id.KW_Then)
1906	then_kw = word_.AsKeywordToken(ate)
1907
1908	body = self._ParseCommandList()
1909	arm = IfArm(elif_kw, cond, then_kw, body.children, then_kw)
1910
1911	arms.append(arm)
1912
1913	self._GetWord()
1914	if self.c_id == Id.KW_Else:
1915	else_kw = word_.AsKeywordToken(self.cur_word)
1916	self._SetNext() # past 'else'
1917	body = self._ParseCommandList()
1918	if_node.else_action = body.children
1919	else:
1920	else_kw = None
1921
1922	if_node.else_kw = else_kw
1923
1924	def ParseIf(self):
1925	# type: () -> command.If
1926	"""
1927	if_clause : If command_list Then command_list else_part? Fi ;
1928
1929	open : '{' \| Then
1930	close : '}' \| Fi
1931
1932	ysh_if : If ( command_list \| '(' expr ')' )
1933	open command_list else_part? close;
1934
1935	There are 2 conditionals here: parse_paren, then parse_brace
1936	"""
1937	if_node = command.If.CreateNull(alloc_lists=True)
1938	if_kw = word_.AsKeywordToken(self.cur_word)
1939	if_node.if_kw = if_kw
1940	self._SetNext() # past 'if'
1941
1942	if (self.parse_opts.parse_paren() and
1943	self.w_parser.LookPastSpace() == Id.Op_LParen):
1944	# if (x + 1)
1945	enode = self.w_parser.ParseYshExprForCommand()
1946	cond = condition.YshExpr(enode) # type: condition_t
1947	else:
1948	# if echo 1; echo 2; then
1949	# Remove ambiguity with if cd / {
1950	cond = self._ParseConditionList()
1951
1952	self._GetWord()
1953	if self.parse_opts.parse_brace() and self.c_id == Id.Lit_LBrace:
1954	return self._ParseYshIf(if_kw, cond)
1955
1956	ate = self._Eat(Id.KW_Then)
1957	then_kw = word_.AsKeywordToken(ate)
1958
1959	body = self._ParseCommandList()
1960
1961	# First arm
1962	arm = IfArm(if_kw, cond, then_kw, body.children, then_kw)
1963	if_node.arms.append(arm)
1964
1965	# 2nd to Nth arm
1966	if self.c_id in (Id.KW_Elif, Id.KW_Else):
1967	self._ParseElifElse(if_node)
1968
1969	ate = self._Eat(Id.KW_Fi)
1970	if_node.fi_kw = word_.AsKeywordToken(ate)
1971
1972	return if_node
1973
1974	def ParseTime(self):
1975	# type: () -> command_t
1976	"""Time [-p] pipeline.
1977
1978	According to bash help.
1979	"""
1980	time_kw = word_.AsKeywordToken(self.cur_word)
1981	self._SetNext() # skip time
1982	pipeline = self.ParsePipeline()
1983	return command.TimeBlock(time_kw, pipeline)
1984
1985	def ParseCompoundCommand(self):
1986	# type: () -> command_t
1987	"""
1988	Refactoring: we put io_redirect* here instead of in function_body and
1989	command.
1990
1991	compound_command : brace_group io_redirect*
1992	\| subshell io_redirect*
1993	\| for_clause io_redirect*
1994	\| while_clause io_redirect*
1995	\| until_clause io_redirect*
1996	\| if_clause io_redirect*
1997	\| case_clause io_redirect*
1998
1999	# bash extensions
2000	\| time_clause
2001	\| [[ BoolExpr ]]
2002	\| (( ArithExpr ))
2003	"""
2004	self._GetWord()
2005	if self.c_id == Id.Lit_LBrace:
2006	n1 = self.ParseBraceGroup()
2007	return self._MaybeParseRedirectList(n1)
2008	if self.c_id == Id.Op_LParen:
2009	n2 = self.ParseSubshell()
2010	return self._MaybeParseRedirectList(n2)
2011
2012	if self.c_id == Id.KW_For:
2013	# Note: Redirects parsed in this call. POSIX for and bash for ((
2014	# have different nodetypes.
2015	return self.ParseFor()
2016	if self.c_id in (Id.KW_While, Id.KW_Until):
2017	keyword = word_.AsKeywordToken(self.cur_word)
2018	n3 = self.ParseWhileUntil(keyword)
2019	return self._MaybeParseRedirectList(n3)
2020
2021	if self.c_id == Id.KW_If:
2022	n4 = self.ParseIf()
2023	return self._MaybeParseRedirectList(n4)
2024
2025	if self.c_id == Id.KW_Case:
2026	n5 = self.ParseCase()
2027	return self._MaybeParseRedirectList(n5)
2028
2029	if self.c_id == Id.KW_DLeftBracket:
2030	if not self.parse_opts.parse_dbracket():
2031	p_die('Bash [[ not allowed in YSH (parse_dbracket)',
2032	loc.Word(self.cur_word))
2033	n6 = self.ParseDBracket()
2034	return self._MaybeParseRedirectList(n6)
2035	if self.c_id == Id.Op_DLeftParen:
2036	if not self.parse_opts.parse_dparen():
2037	p_die(
2038	'Bash (( not allowed in YSH (parse_dparen, see OILS-ERR-14 for wart)',
2039	loc.Word(self.cur_word))
2040	n7 = self.ParseDParen()
2041	return self._MaybeParseRedirectList(n7)
2042
2043	# bash extensions: no redirects
2044	if self.c_id == Id.KW_Time:
2045	return self.ParseTime()
2046
2047	# Happens in function body, e.g. myfunc() oops
2048	p_die(
2049	'Unexpected word while parsing compound command (%s)' %
2050	Id_str(self.c_id), loc.Word(self.cur_word))
2051	assert False # for MyPy
2052
2053	def ParseFunctionDef(self):
2054	# type: () -> command.ShFunction
2055	"""
2056	function_header : fname '(' ')'
2057	function_def : function_header newline_ok function_body ;
2058
2059	Precondition: Looking at the function name.
2060
2061	NOTE: There is an ambiguity with:
2062
2063	function foo ( echo hi ) and
2064	function foo () ( echo hi )
2065
2066	Bash only accepts the latter, though it doesn't really follow a grammar.
2067	"""
2068	word0 = cast(CompoundWord, self.cur_word) # caller ensures validity
2069	name = word_.ShFunctionName(word0)
2070	if len(name) == 0: # example: foo$x is invalid
2071	p_die('Invalid function name', loc.Word(word0))
2072
2073	part0 = word0.parts[0]
2074	# If we got a non-empty string from ShFunctionName, this should be true.
2075	assert part0.tag() == word_part_e.Literal
2076	blame_tok = cast(Token, part0) # for ctx_VarChecker
2077
2078	self._SetNext() # move past function name
2079
2080	# Must be true because of lookahead
2081	self._GetWord()
2082	assert self.c_id == Id.Op_LParen, self.cur_word
2083
2084	self.lexer.PushHint(Id.Op_RParen, Id.Right_ShFunction)
2085	self._SetNext()
2086
2087	self._GetWord()
2088	if self.c_id == Id.Right_ShFunction:
2089	# 'f ()' implies a function definition, since invoking it with no args
2090	# would just be 'f'
2091	self._SetNext()
2092
2093	self._NewlineOk()
2094
2095	func = command.ShFunction.CreateNull()
2096	func.name = name
2097	with ctx_VarChecker(self.var_checker, blame_tok):
2098	func.body = self.ParseCompoundCommand()
2099
2100	func.name_tok = location.LeftTokenForCompoundWord(word0)
2101	return func
2102	else:
2103	p_die('Expected ) in function definition', loc.Word(self.cur_word))
2104	return None
2105
2106	def ParseKshFunctionDef(self):
2107	# type: () -> command.ShFunction
2108	"""
2109	ksh_function_def : 'function' fname ( '(' ')' )? newline_ok function_body
2110	"""
2111	keyword_tok = word_.AsKeywordToken(self.cur_word)
2112
2113	self._SetNext() # skip past 'function'
2114	self._GetWord()
2115
2116	cur_word = cast(CompoundWord, self.cur_word) # caller ensures validity
2117	name = word_.ShFunctionName(cur_word)
2118	if len(name) == 0: # example: foo$x is invalid
2119	p_die('Invalid KSH-style function name', loc.Word(cur_word))
2120
2121	name_word = self.cur_word
2122	self._SetNext() # skip past 'function name
2123
2124	self._GetWord()
2125	if self.c_id == Id.Op_LParen:
2126	self.lexer.PushHint(Id.Op_RParen, Id.Right_ShFunction)
2127	self._SetNext()
2128	self._Eat(Id.Right_ShFunction)
2129
2130	self._NewlineOk()
2131
2132	func = command.ShFunction.CreateNull()
2133	func.name = name
2134	with ctx_VarChecker(self.var_checker, keyword_tok):
2135	func.body = self.ParseCompoundCommand()
2136
2137	func.keyword = keyword_tok
2138	func.name_tok = location.LeftTokenForWord(name_word)
2139	return func
2140
2141	def ParseYshProc(self):
2142	# type: () -> Proc
2143	node = Proc.CreateNull(alloc_lists=True)
2144
2145	keyword_tok = word_.AsKeywordToken(self.cur_word)
2146	node.keyword = keyword_tok
2147
2148	with ctx_VarChecker(self.var_checker, keyword_tok):
2149	with ctx_CmdMode(self, cmd_mode_e.Proc):
2150	self.w_parser.ParseProc(node)
2151	if node.sig.tag() == proc_sig_e.Closed: # Register params
2152	sig = cast(proc_sig.Closed, node.sig)
2153
2154	# Treat 3 kinds of params as variables.
2155	wp = sig.word
2156	if wp:
2157	for param in wp.params:
2158	self.var_checker.Check(Id.KW_Var, param.name,
2159	param.blame_tok)
2160	if wp.rest_of:
2161	r = wp.rest_of
2162	self.var_checker.Check(Id.KW_Var, r.name,
2163	r.blame_tok)
2164	# We COULD register __out here but it would require a different API.
2165	#if param.prefix and param.prefix.id == Id.Arith_Colon:
2166	# self.var_checker.Check(Id.KW_Var, '__' + param.name)
2167
2168	posit = sig.positional
2169	if posit:
2170	for param in posit.params:
2171	self.var_checker.Check(Id.KW_Var, param.name,
2172	param.blame_tok)
2173	if posit.rest_of:
2174	r = posit.rest_of
2175	self.var_checker.Check(Id.KW_Var, r.name,
2176	r.blame_tok)
2177
2178	named = sig.named
2179	if named:
2180	for param in named.params:
2181	self.var_checker.Check(Id.KW_Var, param.name,
2182	param.blame_tok)
2183	if named.rest_of:
2184	r = named.rest_of
2185	self.var_checker.Check(Id.KW_Var, r.name,
2186	r.blame_tok)
2187
2188	if sig.block_param:
2189	b = sig.block_param
2190	self.var_checker.Check(Id.KW_Var, b.name, b.blame_tok)
2191
2192	self._SetNext()
2193	node.body = self.ParseBraceGroup()
2194	# No redirects for YSH procs (only at call site)
2195
2196	return node
2197
2198	def ParseYshFunc(self):
2199	# type: () -> Func
2200	"""
2201	ysh_func: (
2202	Expr_Name '(' [func_params] [';' func_params] ')' ['=>' type_expr] '{'
2203	)
2204	Looking at KW_Func
2205	"""
2206	node = Func.CreateNull(alloc_lists=True)
2207
2208	keyword_tok = word_.AsKeywordToken(self.cur_word)
2209	node.keyword = keyword_tok
2210
2211	with ctx_VarChecker(self.var_checker, keyword_tok):
2212	self.w_parser.ParseFunc(node)
2213
2214	posit = node.positional
2215	if posit:
2216	for param in posit.params:
2217	self.var_checker.Check(Id.KW_Var, param.name,
2218	param.blame_tok)
2219	if posit.rest_of:
2220	r = posit.rest_of
2221	self.var_checker.Check(Id.KW_Var, r.name, r.blame_tok)
2222
2223	named = node.named
2224	if named:
2225	for param in named.params:
2226	self.var_checker.Check(Id.KW_Var, param.name,
2227	param.blame_tok)
2228	if named.rest_of:
2229	r = named.rest_of
2230	self.var_checker.Check(Id.KW_Var, r.name, r.blame_tok)
2231
2232	self._SetNext()
2233	with ctx_CmdMode(self, cmd_mode_e.Func):
2234	node.body = self.ParseBraceGroup()
2235
2236	return node
2237
2238	def ParseCoproc(self):
2239	# type: () -> command_t
2240	"""
2241	TODO: command.Coproc?
2242	"""
2243	raise NotImplementedError()
2244
2245	def ParseSubshell(self):
2246	# type: () -> command.Subshell
2247	"""
2248	subshell : '(' compound_list ')'
2249
2250	Looking at Op_LParen
2251	"""
2252	left = word_.AsOperatorToken(self.cur_word)
2253	self._SetNext() # skip past (
2254
2255	# Ensure that something $( (cd / && pwd) ) works. If ) is already on the
2256	# translation stack, we want to delay it.
2257
2258	self.lexer.PushHint(Id.Op_RParen, Id.Right_Subshell)
2259
2260	c_list = self._ParseCommandList()
2261	if len(c_list.children) == 1:
2262	child = c_list.children[0]
2263	else:
2264	child = c_list
2265
2266	ate = self._Eat(Id.Right_Subshell)
2267	right = word_.AsOperatorToken(ate)
2268
2269	return command.Subshell(left, child, right)
2270
2271	def ParseDBracket(self):
2272	# type: () -> command.DBracket
2273	"""Pass the underlying word parser off to the boolean expression
2274	parser."""
2275	left = word_.AsKeywordToken(self.cur_word)
2276	# TODO: Test interactive. Without closing ]], you should get > prompt
2277	# (PS2)
2278
2279	self._SetNext() # skip [[
2280	b_parser = bool_parse.BoolParser(self.w_parser)
2281	bnode, right = b_parser.Parse() # May raise
2282	return command.DBracket(left, bnode, right)
2283
2284	def ParseDParen(self):
2285	# type: () -> command.DParen
2286	left = word_.AsOperatorToken(self.cur_word)
2287
2288	self._SetNext() # skip ((
2289	anode, right = self.w_parser.ReadDParen()
2290	assert anode is not None
2291
2292	return command.DParen(left, anode, right)
2293
2294	def ParseCommand(self):
2295	# type: () -> command_t
2296	"""
2297	command : simple_command
2298	\| compound_command # OSH edit: io_redirect* folded in
2299	\| function_def
2300	\| ksh_function_def
2301
2302	# YSH extensions
2303	\| proc NAME ...
2304	\| typed proc NAME ...
2305	\| func NAME ...
2306	\| const ...
2307	\| var ...
2308	\| setglobal ...
2309	\| setref ...
2310	\| setvar ...
2311	\| call EXPR
2312	\| = EXPR
2313	;
2314
2315	Note: the reason const / var are not part of compound_command is because
2316	they can't be alone in a shell function body.
2317
2318	Example:
2319	This is valid shell f() if true; then echo hi; fi
2320	This is invalid f() var x = 1
2321	"""
2322	if self._AtSecondaryKeyword():
2323	p_die('Unexpected word when parsing command',
2324	loc.Word(self.cur_word))
2325
2326	# YSH Extensions
2327
2328	if self.c_id == Id.KW_Proc: # proc p { ... }
2329	# proc is hidden because of the 'local reasoning' principle. Code
2330	# inside procs should be YSH, full stop. That means ysh:upgrade is
2331	# on.
2332	if self.parse_opts.parse_proc():
2333	return self.ParseYshProc()
2334	else:
2335	# 2024-02: This avoids bad syntax errors if you type YSH code
2336	# into OSH
2337	# proc p (x) { echo hi } would actually be parsed as a
2338	# command.Simple! Shell compatibility: quote 'proc'
2339	p_die("proc is a YSH keyword, but this is OSH.",
2340	loc.Word(self.cur_word))
2341
2342	if self.c_id == Id.KW_Typed: # typed proc p () { ... }
2343	self._SetNext()
2344	self._GetWord()
2345	if self.c_id != Id.KW_Proc:
2346	p_die("Expected 'proc' after 'typed'", loc.Word(self.cur_word))
2347
2348	if self.parse_opts.parse_proc():
2349	return self.ParseYshProc()
2350	else:
2351	p_die("typed is a YSH keyword, but this is OSH.",
2352	loc.Word(self.cur_word))
2353
2354	if self.c_id == Id.KW_Func: # func f(x) { ... }
2355	if self.parse_opts.parse_func():
2356	return self.ParseYshFunc()
2357	else:
2358	# Same reasoning as above, for 'proc'
2359	p_die("func is a YSH keyword, but this is OSH.",
2360	loc.Word(self.cur_word))
2361
2362	if self.c_id == Id.KW_Const and self.cmd_mode != cmd_mode_e.Shell:
2363	p_die("const can't be inside proc or func. Use var instead.",
2364	loc.Word(self.cur_word))
2365
2366	if self.c_id in (Id.KW_Var, Id.KW_Const): # var x = 1
2367	keyword_id = self.c_id
2368	kw_token = word_.LiteralToken(self.cur_word)
2369	self._SetNext()
2370	n8 = self.w_parser.ParseVarDecl(kw_token)
2371	for lhs in n8.lhs:
2372	self.var_checker.Check(keyword_id, lhs.name, lhs.left)
2373	return n8
2374
2375	if self.c_id in (Id.KW_SetVar, Id.KW_SetGlobal):
2376	kw_token = word_.LiteralToken(self.cur_word)
2377	self._SetNext()
2378	n9 = self.w_parser.ParseMutation(kw_token, self.var_checker)
2379	return n9
2380
2381	if self.c_id in (Id.KW_Call, Id.Lit_Equals):
2382	# = 42 + a[i]
2383	# call mylist->append('x')
2384
2385	keyword = word_.LiteralToken(self.cur_word)
2386	assert keyword is not None
2387	self._SetNext()
2388	enode = self.w_parser.ParseCommandExpr()
2389	return command.Expr(keyword, enode)
2390
2391	if self.c_id == Id.KW_Function:
2392	return self.ParseKshFunctionDef()
2393
2394	if self.c_id in (Id.KW_DLeftBracket, Id.Op_DLeftParen, Id.Op_LParen,
2395	Id.Lit_LBrace, Id.KW_For, Id.KW_While, Id.KW_Until,
2396	Id.KW_If, Id.KW_Case, Id.KW_Time):
2397	return self.ParseCompoundCommand()
2398
2399	# Syntax error for '}' starting a line, which all shells disallow.
2400	if self.c_id == Id.Lit_RBrace:
2401	p_die('Unexpected right brace', loc.Word(self.cur_word))
2402
2403	if self.c_kind == Kind.Redir: # Leading redirect
2404	return self.ParseSimpleCommand()
2405
2406	if self.c_kind == Kind.Word:
2407	# ensured by Kind.Word
2408	cur_word = cast(CompoundWord, self.cur_word)
2409
2410	# NOTE: At the top level, only Token and Compound are possible.
2411	# Can this be modelled better in the type system, removing asserts?
2412	#
2413	# TODO: This can be a proc INVOCATION! (Doesn't even need parse_paren)
2414	# Problem: We have to distinguish f( ) { echo ; } and myproc (x, y)
2415	# That requires 2 tokens of lookahead, which we don't have
2416	#
2417	# Or maybe we don't just have ParseSimpleCommand -- we will have
2418	# ParseYshCommand or something
2419
2420	if (self.w_parser.LookAheadFuncParens() and
2421	not word_.IsVarLike(cur_word)):
2422	return self.ParseFunctionDef() # f() { echo; } # function
2423
2424	# Parse x = 1+2*3 when inside HayNode { } blocks
2425	parts = cur_word.parts
2426	if self.parse_opts.parse_equals() and len(parts) == 1:
2427	part0 = parts[0]
2428	if part0.tag() == word_part_e.Literal:
2429	tok = cast(Token, part0)
2430	if (tok.id == Id.Lit_Chars and
2431	self.w_parser.LookPastSpace() == Id.Lit_Equals and
2432	match.IsValidVarName(lexer.LazyStr(tok))):
2433
2434	if (len(self.hay_attrs_stack) and
2435	self.hay_attrs_stack[-1]):
2436	# Note: no static var_checker.Check() for bare assignment
2437	enode = self.w_parser.ParseBareDecl()
2438	self._SetNext() # Somehow this is necessary
2439	# TODO: Use BareDecl here. Well, do that when we
2440	# treat it as const or lazy.
2441	return command.VarDecl(
2442	None,
2443	[NameType(tok, lexer.TokenVal(tok), None)],
2444	enode)
2445	else:
2446	self._SetNext()
2447	self._GetWord()
2448	p_die(
2449	'Unexpected = (Hint: use var/setvar, or quote it)',
2450	loc.Word(self.cur_word))
2451
2452	# echo foo
2453	# f=(a b c) # array
2454	# array[1+2]+=1
2455	return self.ParseSimpleCommand()
2456
2457	if self.c_kind == Kind.Eof:
2458	p_die("Unexpected EOF while parsing command",
2459	loc.Word(self.cur_word))
2460
2461	# NOTE: This only happens in batch mode in the second turn of the loop!
2462	# e.g. )
2463	p_die("Invalid word while parsing command", loc.Word(self.cur_word))
2464
2465	assert False # for MyPy
2466
2467	def ParsePipeline(self):
2468	# type: () -> command_t
2469	"""
2470	pipeline : Bang? command ( '\|' newline_ok command )* ;
2471	"""
2472	negated = None # type: Optional[Token]
2473
2474	self._GetWord()
2475	if self.c_id == Id.KW_Bang:
2476	negated = word_.AsKeywordToken(self.cur_word)
2477	self._SetNext()
2478
2479	child = self.ParseCommand()
2480	assert child is not None
2481
2482	children = [child]
2483
2484	self._GetWord()
2485	if self.c_id not in (Id.Op_Pipe, Id.Op_PipeAmp):
2486	if negated is not None:
2487	node = command.Pipeline(negated, children, [])
2488	return node
2489	else:
2490	return child # no pipeline
2491
2492	# \| or \|&
2493	ops = [] # type: List[Token]
2494	while True:
2495	op = word_.AsOperatorToken(self.cur_word)
2496	ops.append(op)
2497
2498	self._SetNext() # skip past Id.Op_Pipe or Id.Op_PipeAmp
2499	self._NewlineOk()
2500
2501	child = self.ParseCommand()
2502	children.append(child)
2503
2504	self._GetWord()
2505	if self.c_id not in (Id.Op_Pipe, Id.Op_PipeAmp):
2506	break
2507
2508	return command.Pipeline(negated, children, ops)
2509
2510	def ParseAndOr(self):
2511	# type: () -> command_t
2512	self._GetWord()
2513	if self.c_id == Id.Lit_TDot:
2514	# We got '...', so parse in multiline mode
2515	self._SetNext()
2516	with word_.ctx_Multiline(self.w_parser):
2517	return self._ParseAndOr()
2518
2519	# Parse in normal mode, not multiline
2520	return self._ParseAndOr()
2521
2522	def _ParseAndOr(self):
2523	# type: () -> command_t
2524	"""
2525	and_or : and_or ( AND_IF \| OR_IF ) newline_ok pipeline
2526	\| pipeline
2527
2528	Note that it is left recursive and left associative. We parse it
2529	iteratively with a token of lookahead.
2530	"""
2531	child = self.ParsePipeline()
2532	assert child is not None
2533
2534	self._GetWord()
2535	if self.c_id not in (Id.Op_DPipe, Id.Op_DAmp):
2536	return child
2537
2538	ops = [] # type: List[Token]
2539	children = [child]
2540
2541	while True:
2542	ops.append(word_.AsOperatorToken(self.cur_word))
2543
2544	self._SetNext() # skip past \|\| &&
2545	self._NewlineOk()
2546
2547	child = self.ParsePipeline()
2548	children.append(child)
2549
2550	self._GetWord()
2551	if self.c_id not in (Id.Op_DPipe, Id.Op_DAmp):
2552	break
2553
2554	return command.AndOr(children, ops)
2555
2556	# NOTE: _ParseCommandLine and _ParseCommandTerm are similar, but different.
2557
2558	# At the top level, we execute after every line, e.g. to
2559	# - process alias (a form of dynamic parsing)
2560	# - process 'exit', because invalid syntax might appear after it
2561
2562	# On the other hand, for a while loop body, we parse the whole thing at once,
2563	# and then execute it. We don't want to parse it over and over again!
2564
2565	# COMPARE
2566	# command_line : and_or (sync_op and_or)* trailer? ; # TOP LEVEL
2567	# command_term : and_or (trailer and_or)* ; # CHILDREN
2568
2569	def _ParseCommandLine(self):
2570	# type: () -> command_t
2571	"""
2572	command_line : and_or (sync_op and_or)* trailer? ;
2573	trailer : sync_op newline_ok
2574	\| NEWLINES;
2575	sync_op : '&' \| ';';
2576
2577	NOTE: This rule causes LL(k > 1) behavior. We would have to peek to see if
2578	there is another command word after the sync op.
2579
2580	But it's easier to express imperatively. Do the following in a loop:
2581	1. ParseAndOr
2582	2. Peek.
2583	a. If there's a newline, then return. (We're only parsing a single
2584	line.)
2585	b. If there's a sync_op, process it. Then look for a newline and
2586	return. Otherwise, parse another AndOr.
2587	"""
2588	# This END_LIST is slightly different than END_LIST in _ParseCommandTerm.
2589	# I don't think we should add anything else here; otherwise it will be
2590	# ignored at the end of ParseInteractiveLine(), e.g. leading to bug #301.
2591	END_LIST = [Id.Op_Newline, Id.Eof_Real]
2592
2593	children = [] # type: List[command_t]
2594	done = False
2595	while not done:
2596	child = self.ParseAndOr()
2597
2598	self._GetWord()
2599	if self.c_id in (Id.Op_Semi, Id.Op_Amp):
2600	tok = cast(Token, self.cur_word) # for MyPy
2601	child = command.Sentence(child, tok)
2602	self._SetNext()
2603
2604	self._GetWord()
2605	if self.c_id in END_LIST:
2606	done = True
2607
2608	elif self.c_id in END_LIST:
2609	done = True
2610
2611	else:
2612	# e.g. echo a(b)
2613	p_die(
2614	'Invalid word while parsing command line (%s)' %
2615	Id_str(self.c_id), loc.Word(self.cur_word))
2616
2617	children.append(child)
2618
2619	# Simplify the AST.
2620	if len(children) > 1:
2621	return command.CommandList(children)
2622	else:
2623	return children[0]
2624
2625	def _ParseCommandTerm(self):
2626	# type: () -> command.CommandList
2627	""""
2628	command_term : and_or (trailer and_or)* ;
2629	trailer : sync_op newline_ok
2630	\| NEWLINES;
2631	sync_op : '&' \| ';';
2632
2633	This is handled in imperative style, like _ParseCommandLine.
2634	Called by _ParseCommandList for all blocks, and also for ParseCaseArm,
2635	which is slightly different. (HOW? Is it the DSEMI?)
2636
2637	Returns:
2638	syntax_asdl.command
2639	"""
2640	# Token types that will end the command term.
2641	END_LIST = [
2642	self.eof_id, Id.Right_Subshell, Id.Lit_RBrace, Id.Op_DSemi,
2643	Id.Op_SemiAmp, Id.Op_DSemiAmp
2644	]
2645
2646	# NOTE: This is similar to _ParseCommandLine.
2647	#
2648	# - Why aren't we doing END_LIST in _ParseCommandLine?
2649	# - Because you will never be inside $() at the top level.
2650	# - We also know it will end in a newline. It can't end in "fi"!
2651	# - example: if true; then { echo hi; } fi
2652
2653	children = [] # type: List[command_t]
2654	done = False
2655	while not done:
2656	# Most keywords are valid "first words". But do/done/then do not BEGIN
2657	# commands, so they are not valid.
2658	if self._AtSecondaryKeyword():
2659	break
2660
2661	child = self.ParseAndOr()
2662
2663	self._GetWord()
2664	if self.c_id == Id.Op_Newline:
2665	self._SetNext()
2666
2667	self._GetWord()
2668	if self.c_id in END_LIST:
2669	done = True
2670
2671	elif self.c_id in (Id.Op_Semi, Id.Op_Amp):
2672	tok = cast(Token, self.cur_word) # for MyPy
2673	child = command.Sentence(child, tok)
2674	self._SetNext()
2675
2676	self._GetWord()
2677	if self.c_id == Id.Op_Newline:
2678	self._SetNext() # skip over newline
2679
2680	# Test if we should keep going. There might be another command after
2681	# the semi and newline.
2682	self._GetWord()
2683	if self.c_id in END_LIST: # \n EOF
2684	done = True
2685
2686	elif self.c_id in END_LIST: # ; EOF
2687	done = True
2688
2689	elif self.c_id in END_LIST: # EOF
2690	done = True
2691
2692	# For if test -f foo; test -f bar {
2693	elif self.parse_opts.parse_brace() and self.c_id == Id.Lit_LBrace:
2694	done = True
2695
2696	elif self.c_kind != Kind.Word:
2697	# e.g. f() { echo (( x )) ; }
2698	# but can't fail on 'fi fi', see osh/cmd_parse_test.py
2699
2700	#log("Invalid %s", self.cur_word)
2701	p_die("Invalid word while parsing command list",
2702	loc.Word(self.cur_word))
2703
2704	children.append(child)
2705
2706	return command.CommandList(children)
2707
2708	def _ParseCommandList(self):
2709	# type: () -> command.CommandList
2710	"""
2711	command_list : newline_ok command_term trailer? ;
2712
2713	This one is called by all the compound commands. It's basically a command
2714	block.
2715
2716	NOTE: Rather than translating the CFG directly, the code follows a style
2717	more like this: more like this: (and_or trailer)+. It makes capture
2718	easier.
2719	"""
2720	self._NewlineOk()
2721	return self._ParseCommandTerm()
2722
2723	def ParseLogicalLine(self):
2724	# type: () -> command_t
2725	"""Parse a single line for main_loop.
2726
2727	A wrapper around _ParseCommandLine(). Similar but not identical to
2728	_ParseCommandList() and ParseCommandSub().
2729
2730	Raises:
2731	ParseError
2732	"""
2733	self._NewlineOk()
2734	self._GetWord()
2735	if self.c_id == Id.Eof_Real:
2736	return None # main loop checks for here docs
2737	node = self._ParseCommandLine()
2738	return node
2739
2740	def ParseInteractiveLine(self):
2741	# type: () -> parse_result_t
2742	"""Parse a single line for Interactive main_loop.
2743
2744	Different from ParseLogicalLine because newlines are handled differently.
2745
2746	Raises:
2747	ParseError
2748	"""
2749	self._GetWord()
2750	if self.c_id == Id.Op_Newline:
2751	return parse_result.EmptyLine
2752	if self.c_id == Id.Eof_Real:
2753	return parse_result.Eof
2754
2755	node = self._ParseCommandLine()
2756	return parse_result.Node(node)
2757
2758	def ParseCommandSub(self):
2759	# type: () -> command_t
2760	"""Parse $(echo hi) and `echo hi` for word_parse.py.
2761
2762	They can have multiple lines, like this: echo $( echo one echo
2763	two )
2764	"""
2765	self._NewlineOk()
2766
2767	self._GetWord()
2768	if self.c_kind == Kind.Eof: # e.g. $()
2769	return command.NoOp
2770
2771	c_list = self._ParseCommandTerm()
2772	if len(c_list.children) == 1:
2773	return c_list.children[0]
2774	else:
2775	return c_list
2776
2777	def CheckForPendingHereDocs(self):
2778	# type: () -> None
2779	# NOTE: This happens when there is no newline at the end of a file, like
2780	# osh -c 'cat <<EOF'
2781	if len(self.pending_here_docs):
2782	node = self.pending_here_docs[0] # Just show the first one?
2783	h = cast(redir_param.HereDoc, node.arg)
2784	p_die('Unterminated here doc began here', loc.Word(h.here_begin))
2785
2786
2787	# vim: sw=4