def consume(cls, ctx, tokens, breakset=None):
"""
Consume tokens and return a tree of nodes. Top-level consumer parsers
comments, whitespace, statements, and flow control blocks.
"""
if breakset is None:
breakset = ()
tree = cls()
blocks = tree.children
while tokens:
token = tokens[0]
if token.type in WHITESPACE_TOKENS:
node = WhitespaceNode.consume(ctx, tokens)
blocks.append(node)
elif token.type in COMMENT_TOKENS:
node = CommentNode.consume(ctx, tokens)
blocks.append(node)
elif token.type in ONOFF_TOKENS:
node = OnOffNode.consume(ctx, tokens)
blocks.append(node)
elif token.type == lex.TokenType.BRACKET_COMMENT:
node = CommentNode.consume(ctx, tokens)
blocks.append(node)
elif token.type == lex.TokenType.WORD:
upper = token.spelling.upper()
if upper in breakset:
return tree
if FlowType.get(upper) is not None:
subtree = FlowControlNode.consume(ctx, tokens)
blocks.append(subtree)
else:
subtree = StatementNode.consume(ctx, tokens)
blocks.append(subtree)
elif token.type == lex.TokenType.ATWORD:
subtree = AtWordStatementNode.consume(ctx, tokens)
blocks.append(subtree)
elif token.type == lex.TokenType.BYTEORDER_MARK:
tokens.pop(0)
else:
raise InternalError(
"Unexpected {} token at {}:{}"
.format(tokens[0].type.name,
tokens[0].begin.line, tokens[0].begin.col))
return tree
def parse(cls, ctx, tokens, npargs, ntup, flags, breakstack):
"""Parse a continuous sequence of `npargs` positional argument pairs.
If npargs is an integer we will consume exactly that many arguments.
If it is not an integer then it is a string meaning:
* "?": zero or one
* "*": zero or more
* "+": one or more
"""
tree = cls()
tree.spec = PositionalSpec(ntup, flags=flags, legacy=True)
subtree = None
active_depth = tree
npargs_consumed = 0
ntup_consumed = 0
while tokens:
# Break if we have consumed enough positional arguments
if pargs_are_full(npargs, npargs_consumed):
break
# Break if the next token belongs to a parent parser, i.e. if it
# matches a keyword argument of something higher in the stack, or if
# it closes a parent group.
if should_break(tokens[0], breakstack):
break
# Otherwise we will consume the token
token = tokens.pop(0)
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
if token.type in WHITESPACE_TOKENS:
active_depth.children.append(token)
continue
# If it's a comment token not associated with an argument, then put it
# directly into the parse tree at the current depth
if token.type in (lex.TokenType.COMMENT,
lex.TokenType.BRACKET_COMMENT):
child = CommentNode()
tree.children.append(child)
child.children.append(token)
continue
# If it's a sentinel comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.FORMAT_OFF,
lex.TokenType.FORMAT_ON):
tree.children.append(OnOffNode.consume(ctx, tokens))
continue
if subtree is None:
subtree = PositionalGroupNode()
subtree.spec = PositionalSpec(2, False, [], flags)
tree.children.append(subtree)
ntup_consumed = 0
# Otherwise is it is a positional argument, so add it to the tree as such
if get_normalized_kwarg(token) in flags:
child = TreeNode(NodeType.FLAG)
else:
child = TreeNode(NodeType.ARGUMENT)
child.children.append(token)
CommentNode.consume_trailing(ctx, tokens, child)
subtree.children.append(child)
ntup_consumed += 1
if ntup_consumed >= ntup:
npargs_consumed += 1
subtree = None
return tree
def parse2(cls, ctx, tokens, cmdspec, kwargs, breakstack):
"""
Standard parser for the commands in the form of::
command_name(parg1 parg2 parg3...
KEYWORD1 kwarg1 kwarg2...
KEYWORD2 kwarg3 kwarg4...
FLAG1 FLAG2 FLAG3)
The parser starts off as a positional parser. If a keyword or flag is
encountered the positional parser is popped off the parse stack. If it was
a keyword then the keyword parser is pushed on the parse stack. If it was
a flag than a new flag parser is pushed onto the stack.
"""
# NOTE(josh): we will pop things off this list, so let's make a copy
pargspecs = list(cmdspec.pargs)
tree = cls()
tree.cmdspec = cmdspec
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
while tokens and tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
continue
# NOTE(josh): if there is only one non-exact legacy specification then we
# reuse that specification for any additional positional arguments that we
# pick up. This is to maintain the current/legacy behavior of simple
# positional argument specifications
# TODO(josh): double check the reasoning for this. I think it might be
# mistaken and unnecessary
default_spec = DEFAULT_PSPEC
if (len(pargspecs) == 1 and pargspecs[0].legacy
and not npargs_is_exact(pargspecs[0].nargs)):
default_spec = pargspecs.pop(0)
all_flags = list(default_spec.flags)
for pspec in pargspecs:
all_flags.extend(pspec.flags)
kwarg_breakstack = breakstack + [
KwargBreaker(list(kwargs.keys()) + all_flags)
]
while tokens:
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
if tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
continue
# If it's a comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.COMMENT,
lex.TokenType.BRACKET_COMMENT):
if comment_belongs_up_tree(ctx, tokens, tree, breakstack):
break
tree.children.append(CommentNode.consume(ctx, tokens))
continue
# If it's a sentinel comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.FORMAT_OFF,
lex.TokenType.FORMAT_ON):
tree.children.append(OnOffNode.consume(ctx, tokens))
continue
# Break if the next token belongs to a parent parser, i.e. if it
# matches a keyword argument of something higher in the stack, or if
# it closes a parent group.
if should_break(tokens[0], breakstack):
# NOTE(josh): if spec.nargs is an exact number of arguments, then we
# shouldn't break on kwarg match from a parent parser. Instead, we
# should consume that many tokens. This is a hack to deal with
# ```install(RUNTIME COMPONENT runtime)``. In this case the second
# occurance of "runtime" should not match the ``RUNTIME`` keyword
# and should not break the positional parser.
# TODO(josh): this is kind of hacky because it will force the positional
# parser to consume a right parenthesis and will lead to parse errors
# in the event of a missing positional argument. Such errors will be
# difficult to debug for the user.
if pargspecs:
pspec = pargspecs[0]
else:
pspec = default_spec
if not npargs_is_exact(pspec.nargs) or pspec.nargs == 0:
break
ntokens = len(tokens)
word = get_normalized_kwarg(tokens[0])
if word in kwargs:
with ctx.pusharg(tree):
subtree = KeywordGroupNode.parse(ctx, tokens, word,
kwargs[word],
kwarg_breakstack)
tree.kwarg_groups.append(subtree)
else:
if pargspecs:
pspec = pargspecs.pop(0)
else:
pspec = default_spec
other_flags = []
for otherspec in pargspecs:
for flag in otherspec.flags:
if flag in pspec.flags:
continue
other_flags.append(flag)
positional_breakstack = breakstack + [
KwargBreaker(list(kwargs.keys()) + other_flags)
]
with ctx.pusharg(tree):
subtree = PositionalGroupNode.parse2(
ctx, tokens, pspec, positional_breakstack)
tree.parg_groups.append(subtree)
if len(tokens) >= ntokens:
raise InternalError(
"parsed an empty subtree at {}:\n {}\n pspec: {}".format(
tokens[0], dump_tree_tostr([tree]), pspec))
tree.children.append(subtree)
return tree
def parse(cls, ctx, tokens, breakstack):
"""
Parser for the commands that take conditional arguments. Similar to the
standard parser but it understands parentheses and can generate
parenthentical groups::
while(CONDITION1 AND (CONDITION2 OR CONDITION3)
OR (CONDITION3 AND (CONDITION4 AND CONDITION5)
OR CONDITION6)
"""
kwargs = {'AND': cls.parse, 'OR': cls.parse}
flags = list(CONDITIONAL_FLAGS)
tree = cls()
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
while tokens and tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
continue
flags = [flag.upper() for flag in flags]
breaker = KwargBreaker(list(kwargs.keys()))
child_breakstack = breakstack + [breaker]
while tokens:
# Break if the next token belongs to a parent parser, i.e. if it
# matches a keyword argument of something higher in the stack, or if
# it closes a parent group.
if should_break(tokens[0], breakstack):
break
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
if tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
continue
# If it's a comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.COMMENT,
lex.TokenType.BRACKET_COMMENT):
# TODO(josh): not sure if we should check comment_belongs_up_tree
# here
child = CommentNode()
tree.children.append(child)
child.children.append(tokens.pop(0))
continue
# If it's a sentinel comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.FORMAT_OFF,
lex.TokenType.FORMAT_ON):
tree.children.append(OnOffNode.consume(ctx, tokens))
continue
# If this is the start of a parenthetical group, then parse the group
if tokens[0].type == lex.TokenType.LEFT_PAREN:
with ctx.pusharg(tree):
subtree = ParenGroupNode.parse(ctx, tokens, breakstack)
tree.children.append(subtree)
continue
ntokens = len(tokens)
word = get_normalized_kwarg(tokens[0])
if word in kwargs:
with ctx.pusharg(tree):
subtree = KeywordGroupNode.parse(ctx, tokens, word,
kwargs[word],
child_breakstack)
assert len(tokens) < ntokens, "parsed an empty subtree"
tree.children.append(subtree)
continue
# Otherwise is it is a positional argument, so add it to the tree as such
with ctx.pusharg(tree):
child = PositionalGroupNode.parse(ctx, tokens, '+', flags,
child_breakstack)
# token = tokens.pop(0)
# if get_normalized_kwarg(token) in flags:
# child = TreeNode(NodeType.FLAG)
# else:
# child = TreeNode(NodeType.ARGUMENT)
# child.children.append(token)
# consume_trailing_comment(child, tokens)
tree.children.append(child)
return tree
def parse2(cls, ctx, tokens, spec, breakstack):
"""
Parse a continuous sequence of `npargs` positional arguments. If npargs is
an integer we will consume exactly that many arguments. If it is not an
integer then it is a string meaning:
* "?": zero or one
* "*": zero or more
* "+": one or more
"""
tree = cls(sortable=spec.sortable, tags=spec.tags)
tree.spec = spec
nconsumed = 0
# Strip off any preceeding whitespace (note that in most cases this has
# already been done but in some cases (such ask kwarg subparser) where
# it hasn't
while tokens and tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
# If the first non-whitespace token is a cmake-format tag annotating
# sortability, then parse it out here and record the annotation
if tokens and get_tag(tokens[0]) in ("sortable", "sort"):
tree.sortable = True
elif tokens and get_tag(tokens[0]) in ("unsortable", "unsort"):
tree.sortable = False
while tokens:
# Break if we have consumed enough positional arguments
if pargs_are_full(spec.nargs, nconsumed):
break
# Break if the next token belongs to a parent parser, i.e. if it
# matches a keyword argument of something higher in the stack, or if
# it closes a parent group.
if should_break(tokens[0], breakstack):
# NOTE(josh): if spec.nargs is an exact number of arguments, then we
# shouldn't break on kwarg match from a parent parser. Instead, we
# should consume the token. This is a hack to deal with
# ```install(RUNTIME COMPONENT runtime)``. In this case the second
# occurance of "runtime" should not match the ``RUNTIME`` keyword
# and should not break the positional parser.
# TODO(josh): this is kind of hacky because it will force the positional
# parser to consume a right parenthesis and will lead to parse errors
# in the event of a missing positional argument. Such errors will be
# difficult to debug for the user.
if not npargs_is_exact(spec.nargs):
break
if tokens[0].type == lex.TokenType.RIGHT_PAREN:
break
# If this is the start of a parenthetical group, then parse the group
# NOTE(josh): syntatically this probably shouldn't be allowed here, but
# cmake seems to accept it so we probably should too.
if tokens[0].type == lex.TokenType.LEFT_PAREN:
with ctx.pusharg(tree):
subtree = ParenGroupNode.parse(ctx, tokens, breakstack)
tree.children.append(subtree)
continue
# If it is a whitespace token then put it directly in the parse tree at
# the current depth
if tokens[0].type in WHITESPACE_TOKENS:
tree.children.append(tokens.pop(0))
continue
# If it's a comment token not associated with an argument, then put it
# directly into the parse tree at the current depth
if tokens[0].type in (lex.TokenType.COMMENT,
lex.TokenType.BRACKET_COMMENT):
if comment_belongs_up_tree(ctx, tokens, tree, breakstack):
break
child = CommentNode.consume(ctx, tokens)
tree.children.append(child)
continue
# If it's a sentinel comment, then add it at the current depth
if tokens[0].type in (lex.TokenType.FORMAT_OFF,
lex.TokenType.FORMAT_ON):
tree.children.append(OnOffNode.consume(ctx, tokens))
continue
# Otherwise is it is a positional argument, so add it to the tree as such
if get_normalized_kwarg(tokens[0]) in spec.flags:
child = TreeNode(NodeType.FLAG)
else:
child = TreeNode(NodeType.ARGUMENT)
child.children.append(tokens.pop(0))
CommentNode.consume_trailing(ctx, tokens, child)
tree.children.append(child)
nconsumed += 1
return tree