def _PushOilTokens(parse_ctx, gr, p, lex):
# type: (ParseContext, Grammar, parse.Parser, Lexer) -> Token
"""Push tokens onto pgen2's parser.
Returns the last token so it can be reused/seen by the CommandParser.
"""
#log('keywords = %s', gr.keywords)
#log('tokens = %s', gr.tokens)
last_token = None # type: Optional[Token]
prev_was_newline = False
balance = 0 # to ignore newlines
while True:
if last_token: # e.g. left over from WordParser
tok = last_token
#log('last_token = %s', last_token)
last_token = None
else:
tok = lex.Read(lex_mode_e.Expr)
#log('tok = %s', tok)
# Comments and whitespace. Newlines aren't ignored.
if consts.GetKind(tok.id) == Kind.Ignored:
continue
# For multiline lists, maps, etc.
if tok.id == Id.Op_Newline:
if balance > 0:
#log('*** SKIPPING NEWLINE')
continue
# Eliminate duplicate newline tokens. It makes the grammar simpler, and
# it's consistent with CPython's lexer and our own WordParser.
if prev_was_newline:
continue
prev_was_newline = True
else:
prev_was_newline = False
balance += _OTHER_BALANCE.get(tok.id, 0)
#log('BALANCE after seeing %s = %d', tok.id, balance)
#if tok.id == Id.Expr_Name and tok.val in KEYWORDS:
# tok.id = KEYWORDS[tok.val]
# log('Replaced with %s', tok.id)
assert tok.id < 256, Id_str(tok.id)
ilabel = _Classify(gr, tok)
#log('tok = %s, ilabel = %d', tok, ilabel)
if p.addtoken(tok.id, tok, ilabel):
return tok
#
# Mututally recursive calls into the command/word parsers.
#
if mylib.PYTHON:
if tok.id == Id.Left_PercentParen: # %(
left_tok = tok
lex.PushHint(Id.Op_RParen, Id.Right_ShArrayLiteral)
# Blame the opening token
line_reader = reader.DisallowedLineReader(parse_ctx.arena, tok)
w_parser = parse_ctx.MakeWordParser(lex, line_reader)
words = []
close_tok = None # type: Optional[Token]
while True:
w = w_parser.ReadWord(lex_mode_e.ShCommand)
if 0:
log('w = %s', w)
if w.tag_() == word_e.Token:
tok = cast(Token, w)
if tok.id == Id.Right_ShArrayLiteral:
close_tok = tok
break
elif tok.id == Id.Op_Newline: # internal newlines allowed
continue
else:
# Token
p_die('Unexpected token in array literal: %r',
tok.val,
word=w)
assert isinstance(w, compound_word) # for MyPy
words.append(w)
words2 = braces.BraceDetectAll(words)
words3 = word_.TildeDetectAll(words2)
typ = Id.Expr_CastedDummy
lit_part = sh_array_literal(left_tok, words3)
opaque = cast(Token, lit_part) # HACK for expr_to_ast
done = p.addtoken(typ, opaque, gr.tokens[typ])
assert not done # can't end the expression
# Now push the closing )
ilabel = _Classify(gr, close_tok)
done = p.addtoken(tok.id, close_tok, ilabel)
assert not done # can't end the expression
continue
if tok.id in (Id.Left_DollarParen, Id.Left_AtParen): # $( @(
left_token = tok
lex.PushHint(Id.Op_RParen, Id.Eof_RParen)
line_reader = reader.DisallowedLineReader(parse_ctx.arena, tok)
c_parser = parse_ctx.MakeParserForCommandSub(
line_reader, lex, Id.Eof_RParen)
node = c_parser.ParseCommandSub()
# A little gross: Copied from osh/word_parse.py
right_token = c_parser.w_parser.cur_token
cs_part = command_sub(left_token, node)
cs_part.spids.append(left_token.span_id)
cs_part.spids.append(right_token.span_id)
typ = Id.Expr_CastedDummy
opaque = cast(Token, cs_part) # HACK for expr_to_ast
done = p.addtoken(typ, opaque, gr.tokens[typ])
assert not done # can't end the expression
# Now push the closing )
ilabel = _Classify(gr, right_token)
done = p.addtoken(right_token.id, right_token, ilabel)
assert not done # can't end the expression
continue
if tok.id == Id.Left_DoubleQuote:
left_token = tok
line_reader = reader.DisallowedLineReader(parse_ctx.arena, tok)
w_parser = parse_ctx.MakeWordParser(lex, line_reader)
parts = [] # type: List[word_part_t]
last_token = w_parser.ReadDoubleQuoted(left_token, parts)
expr_dq_part = double_quoted(left_token, parts)
typ = Id.Expr_CastedDummy
opaque = cast(Token, expr_dq_part) # HACK for expr_to_ast
done = p.addtoken(typ, opaque, gr.tokens[typ])
assert not done # can't end the expression
continue
if tok.id == Id.Left_DollarBrace:
left_token = tok
line_reader = reader.DisallowedLineReader(parse_ctx.arena, tok)
w_parser = parse_ctx.MakeWordParser(lex, line_reader)
part, last_token = w_parser.ReadBracedVarSub(left_token)
# It's casted word_part__BracedVarSub -> dummy -> expr__BracedVarSub!
typ = Id.Expr_CastedDummy
opaque = cast(Token, part) # HACK for expr_to_ast
done = p.addtoken(typ, opaque, gr.tokens[typ])
assert not done # can't end the expression
continue
# '' and c''
if tok.id in (Id.Left_SingleQuoteRaw, Id.Left_SingleQuoteC):
if tok.id == Id.Left_SingleQuoteRaw:
sq_mode = lex_mode_e.SQ_Raw
else:
sq_mode = lex_mode_e.SQ_C
left_token = tok
line_reader = reader.DisallowedLineReader(parse_ctx.arena, tok)
w_parser = parse_ctx.MakeWordParser(lex, line_reader)
tokens = [] # type: List[Token]
no_backslashes = (left_token.val == "'")
last_token = w_parser.ReadSingleQuoted(sq_mode, left_token,
tokens, no_backslashes)
sq_part = single_quoted(left_token, tokens)
typ = Id.Expr_CastedDummy
opaque = cast(Token, sq_part) # HACK for expr_to_ast
done = p.addtoken(typ, opaque, gr.tokens[typ])
assert not done # can't end the expression
continue
else:
# We never broke out -- EOF is too soon (how can this happen???)
raise parse.ParseError("incomplete input", tok.id, tok)
def Matches(self, comp):
# type: (Api) -> Iterator[Union[Iterator, Iterator[str]]]
"""
Args:
comp: Callback args from readline. Readline uses set_completer_delims to
tokenize the string.
Returns a list of matches relative to readline's completion_delims.
We have to post-process the output of various completers.
"""
arena = self.parse_ctx.arena # Used by inner functions
# Pass the original line "out of band" to the completion callback.
line_until_tab = comp.line[:comp.end]
self.comp_ui_state.line_until_tab = line_until_tab
self.parse_ctx.trail.Clear()
line_reader = reader.StringLineReader(line_until_tab,
self.parse_ctx.arena)
c_parser = self.parse_ctx.MakeOshParser(line_reader,
emit_comp_dummy=True)
# We want the output from parse_ctx, so we don't use the return value.
try:
c_parser.ParseLogicalLine()
except error.Parse as e:
# e.g. 'ls | ' will not parse. Now inspect the parser state!
pass
debug_f = self.debug_f
trail = self.parse_ctx.trail
if 1:
trail.PrintDebugString(debug_f)
#
# First try completing the shell language itself.
#
# NOTE: We get Eof_Real in the command state, but not in the middle of a
# BracedVarSub. This is due to the difference between the CommandParser
# and WordParser.
tokens = trail.tokens
last = -1
if tokens[-1].id == Id.Eof_Real:
last -= 1 # ignore it
try:
t1 = tokens[last]
except IndexError:
t1 = None
try:
t2 = tokens[last - 1]
except IndexError:
t2 = None
debug_f.log('line: %r', comp.line)
debug_f.log('rl_slice from byte %d to %d: %r', comp.begin, comp.end,
comp.line[comp.begin:comp.end])
debug_f.log('t1 %s', t1)
debug_f.log('t2 %s', t2)
# Each of the 'yield' statements below returns a fully-completed line, to
# appease the readline library. The root cause of this dance: If there's
# one candidate, readline is responsible for redrawing the input line. OSH
# only displays candidates and never redraws the input line.
def _TokenStart(tok):
# type: (Token) -> int
span = arena.GetLineSpan(tok.span_id)
return span.col
if t2: # We always have t1?
# echo $
if IsDollar(t2) and IsDummy(t1):
self.comp_ui_state.display_pos = _TokenStart(t2) + 1 # 1 for $
for name in self.mem.VarNames():
yield line_until_tab + name # no need to quote var names
return
# echo ${
if t2.id == Id.Left_DollarBrace and IsDummy(t1):
self.comp_ui_state.display_pos = _TokenStart(
t2) + 2 # 2 for ${
for name in self.mem.VarNames():
yield line_until_tab + name # no need to quote var names
return
# echo $P
if t2.id == Id.VSub_DollarName and IsDummy(t1):
# Example: ${undef:-$P
# readline splits at ':' so we have to prepend '-$' to every completed
# variable name.
self.comp_ui_state.display_pos = _TokenStart(t2) + 1 # 1 for $
to_complete = t2.val[1:]
n = len(to_complete)
for name in self.mem.VarNames():
if name.startswith(to_complete):
yield line_until_tab + name[
n:] # no need to quote var names
return
# echo ${P
if t2.id == Id.VSub_Name and IsDummy(t1):
self.comp_ui_state.display_pos = _TokenStart(t2) # no offset
to_complete = t2.val
n = len(to_complete)
for name in self.mem.VarNames():
if name.startswith(to_complete):
yield line_until_tab + name[
n:] # no need to quote var names
return
# echo $(( VAR
if t2.id == Id.Lit_ArithVarLike and IsDummy(t1):
self.comp_ui_state.display_pos = _TokenStart(t2) # no offset
to_complete = t2.val
n = len(to_complete)
for name in self.mem.VarNames():
if name.startswith(to_complete):
yield line_until_tab + name[
n:] # no need to quote var names
return
if trail.words:
# echo ~<TAB>
# echo ~a<TAB> $(home dirs)
# This must be done at a word level, and TildeDetectAll() does NOT help
# here, because they don't have trailing slashes yet! We can't do it on
# tokens, because otherwise f~a will complete. Looking at word_part is
# EXACTLY what we want.
parts = trail.words[-1].parts
if (len(parts) == 2 and parts[0].tag_() == word_part_e.Literal
and parts[1].tag_() == word_part_e.Literal
and parts[0].id == Id.Lit_TildeLike
and parts[1].id == Id.Lit_CompDummy):
t2 = parts[0]
# +1 for ~
self.comp_ui_state.display_pos = _TokenStart(parts[0]) + 1
to_complete = t2.val[1:]
n = len(to_complete)
for u in pwd.getpwall(): # catch errors?
name = u.pw_name
if name.startswith(to_complete):
yield line_until_tab + ShellQuoteB(name[n:]) + '/'
return
# echo hi > f<TAB> (complete redirect arg)
if trail.redirects:
r = trail.redirects[-1]
# Only complete 'echo >', but not 'echo >&' or 'cat <<'
# TODO: Don't complete <<< 'h'
if (r.arg.tag_() == redir_param_e.Word
and consts.RedirArgType(r.op.id) == redir_arg_type_e.Path):
arg_word = r.arg
if WordEndsWithCompDummy(arg_word):
debug_f.log('Completing redirect arg')
try:
val = self.word_ev.EvalWordToString(r.arg)
except error.FatalRuntime as e:
debug_f.log('Error evaluating redirect word: %s', e)
return
if val.tag_() != value_e.Str:
debug_f.log("Didn't get a string from redir arg")
return
span_id = word_.LeftMostSpanForWord(arg_word)
span = arena.GetLineSpan(span_id)
self.comp_ui_state.display_pos = span.col
comp.Update(
to_complete=val.s) # FileSystemAction uses only this
n = len(val.s)
action = FileSystemAction(add_slash=True)
for name in action.Matches(comp):
yield line_until_tab + ShellQuoteB(name[n:])
return
#
# We're not completing the shell language. Delegate to user-defined
# completion for external tools.
#
# Set below, and set on retries.
base_opts = None
user_spec = None
# Used on retries.
partial_argv = []
num_partial = -1
first = None
if trail.words:
# Now check if we're completing a word!
if WordEndsWithCompDummy(trail.words[-1]):
debug_f.log('Completing words')
#
# It didn't look like we need to complete var names, tilde, redirects,
# etc. Now try partial_argv, which may involve invoking PLUGINS.
# needed to complete paths with ~
words2 = word_.TildeDetectAll(trail.words)
if 0:
debug_f.log('After tilde detection')
for w in words2:
print(w, file=debug_f)
if 0:
debug_f.log('words2:')
for w2 in words2:
debug_f.log(' %s', w2)
for w in words2:
try:
# TODO:
# - Should we call EvalWordSequence? But turn globbing off? It
# can do splitting and such.
# - We could have a variant to eval TildeSub to ~ ?
val = self.word_ev.EvalWordToString(w)
except error.FatalRuntime:
# Why would it fail?
continue
if val.tag_() == value_e.Str:
partial_argv.append(val.s)
else:
pass
debug_f.log('partial_argv: %s', partial_argv)
num_partial = len(partial_argv)
first = partial_argv[0]
alias_first = None
debug_f.log('alias_words: %s', trail.alias_words)
if trail.alias_words:
w = trail.alias_words[0]
try:
val = self.word_ev.EvalWordToString(w)
except error.FatalRuntime:
pass
alias_first = val.s
debug_f.log('alias_first: %s', alias_first)
if num_partial == 0: # should never happen because of Lit_CompDummy
raise AssertionError()
elif num_partial == 1:
base_opts, user_spec = self.comp_lookup.GetFirstSpec()
# Display/replace since the beginning of the first word. Note: this
# is non-zero in the case of
# echo $(gr and
# echo `gr
span_id = word_.LeftMostSpanForWord(trail.words[0])
span = arena.GetLineSpan(span_id)
self.comp_ui_state.display_pos = span.col
self.debug_f.log('** DISPLAY_POS = %d',
self.comp_ui_state.display_pos)
else:
base_opts, user_spec = self.comp_lookup.GetSpecForName(
first)
if not user_spec and alias_first:
base_opts, user_spec = self.comp_lookup.GetSpecForName(
alias_first)
if user_spec:
# Pass the aliased command to the user-defined function, and use
# it for retries.
first = alias_first
if not user_spec:
base_opts, user_spec = self.comp_lookup.GetFallback()
# Display since the beginning
span_id = word_.LeftMostSpanForWord(trail.words[-1])
span = arena.GetLineSpan(span_id)
self.comp_ui_state.display_pos = span.col
self.debug_f.log('words[-1]: %r', trail.words[-1])
self.debug_f.log('display_pos %d',
self.comp_ui_state.display_pos)
# Update the API for user-defined functions.
index = len(
partial_argv) - 1 # COMP_CWORD is -1 when it's empty
prev = '' if index == 0 else partial_argv[index - 1]
comp.Update(first=first,
to_complete=partial_argv[-1],
prev=prev,
index=index,
partial_argv=partial_argv)
# This happens in the case of [[ and ((, or a syntax error like 'echo < >'.
if not user_spec:
debug_f.log("Didn't find anything to complete")
return
# Reset it back to what was registered. User-defined functions can mutate
# it.
dynamic_opts = {}
self.compopt_state.dynamic_opts = dynamic_opts
self.compopt_state.currently_completing = True
try:
done = False
while not done:
try:
for candidate in self._PostProcess(base_opts, dynamic_opts,
user_spec, comp):
yield candidate
except _RetryCompletion as e:
debug_f.log('Got 124, trying again ...')
# Get another user_spec. The ShellFuncAction may have 'sourced' code
# and run 'complete' to mutate comp_lookup, and we want to get that
# new entry.
if num_partial == 0:
raise AssertionError()
elif num_partial == 1:
base_opts, user_spec = self.comp_lookup.GetFirstSpec()
else:
# (already processed alias_first)
base_opts, user_spec = self.comp_lookup.GetSpecForName(
first)
if not user_spec:
base_opts, user_spec = self.comp_lookup.GetFallback(
)
else:
done = True # exhausted candidates without getting a retry
finally:
self.compopt_state.currently_completing = False
def _ReadArrayLiteral(self):
# type: () -> word_part_t
"""
a=(1 2 3)
TODO: See osh/cmd_parse.py:164 for Id.Lit_ArrayLhsOpen, for a[x++]=1
We want:
A=(['x']=1 ["x"]=2 [$x$y]=3)
Maybe allow this as a literal string? Because I think I've seen it before?
Or maybe force people to patch to learn the rule.
A=([x]=4)
Starts with Lit_Other '[', and then it has Lit_ArrayLhsClose
Maybe enforce that ALL have keys or NONE of have keys.
"""
self._Next(lex_mode_e.ShCommand) # advance past (
self._Peek()
if self.cur_token.id != Id.Op_LParen:
p_die('Expected ( after =, got %r', self.cur_token.val,
token=self.cur_token)
left_token = self.cur_token
paren_spid = self.cur_token.span_id
# MUST use a new word parser (with same lexer).
w_parser = self.parse_ctx.MakeWordParser(self.lexer, self.line_reader)
words = []
while True:
w = w_parser.ReadWord(lex_mode_e.ShCommand)
if isinstance(w, word__Token):
word_id = word_.CommandId(w)
if word_id == Id.Right_ShArrayLiteral:
break
# Unlike command parsing, array parsing allows embedded \n.
elif word_id == Id.Op_Newline:
continue
else:
# Token
p_die('Unexpected token in array literal: %r', w.token.val, word=w)
assert isinstance(w, word__Compound) # for MyPy
words.append(w)
if not words: # a=() is empty indexed array
# ignore for invariant List?
node = sh_array_literal(left_token, words) # type: ignore
node.spids.append(left_token.span_id)
return node
# If the first one is a key/value pair, then the rest are assumed to be.
pair = word_.DetectAssocPair(words[0])
if pair:
pairs = [pair[0], pair[1]] # flat representation
n = len(words)
for i in xrange(1, n):
w = words[i]
pair = word_.DetectAssocPair(w)
if not pair:
p_die("Expected associative array pair", word=w)
pairs.append(pair[0]) # flat representation
pairs.append(pair[1])
# invariant List?
node = word_part.AssocArrayLiteral(left_token, pairs) # type: ignore
node.spids.append(paren_spid)
return node
words2 = braces.BraceDetectAll(words)
words3 = word_.TildeDetectAll(words2)
node = sh_array_literal(left_token, words3)
node.spids.append(paren_spid)
return node