/// \brief Consume tokens and store them in the passed token container until
/// we've passed the try keyword and constructor initializers and have consumed
/// the opening brace of the function body. The opening brace will be consumed
/// if and only if there was no error.
///
/// \return True on error.
bool Parser::ConsumeAndStoreFunctionPrologue(CachedTokens &Toks) {
if (Tok.is(tok::kw_try)) {
Toks.push_back(Tok);
ConsumeToken();
}
if (Tok.is(tok::colon)) {
// Initializers can contain braces too.
Toks.push_back(Tok);
ConsumeToken();
while (Tok.is(tok::identifier) || Tok.is(tok::coloncolon)) {
if (Tok.is(tok::eof) || Tok.is(tok::semi))
return true;
// Grab the identifier.
if (!ConsumeAndStoreUntil(tok::l_paren, tok::l_brace, Toks,
/*StopAtSemi=*/true,
/*ConsumeFinalToken=*/false))
return true;
tok::TokenKind kind = Tok.getKind();
Toks.push_back(Tok);
if (kind == tok::l_paren)
ConsumeParen();
else {
assert(kind == tok::l_brace && "Must be left paren or brace here.");
ConsumeBrace();
// In C++03, this has to be the start of the function body, which
// means the initializer is malformed.
if (!getLang().CPlusPlus0x)
return false;
}
// Grab the initializer
if (!ConsumeAndStoreUntil(kind == tok::l_paren ? tok::r_paren :
tok::r_brace,
Toks, /*StopAtSemi=*/true))
return true;
// Grab the separating comma, if any.
if (Tok.is(tok::comma)) {
Toks.push_back(Tok);
ConsumeToken();
}
}
}
// Grab any remaining garbage to be diagnosed later. We stop when we reach a
// brace: an opening one is the function body, while a closing one probably
// means we've reached the end of the class.
if (!ConsumeAndStoreUntil(tok::l_brace, tok::r_brace, Toks,
/*StopAtSemi=*/true, /*ConsumeFinalToken=*/false))
return true;
if(Tok.isNot(tok::l_brace))
return true;
Toks.push_back(Tok);
ConsumeBrace();
return false;
}
/// \brief Consume and store tokens from the '?' to the ':' in a conditional
/// expression.
bool Parser::ConsumeAndStoreConditional(CachedTokens &Toks) {
// Consume '?'.
assert(Tok.is(tok::question));
Toks.push_back(Tok);
ConsumeToken();
while (Tok.isNot(tok::colon)) {
if (!ConsumeAndStoreUntil(tok::question, tok::colon, Toks, /*StopAtSemi*/true,
/*ConsumeFinalToken*/false))
return false;
// If we found a nested conditional, consume it.
if (Tok.is(tok::question) && !ConsumeAndStoreConditional(Toks))
return false;
}
// Consume ':'.
Toks.push_back(Tok);
ConsumeToken();
return true;
}
/// \brief Lex a delayed template function for late parsing.
void Parser::LexTemplateFunctionForLateParsing(CachedTokens &Toks) {
tok::TokenKind kind = Tok.getKind();
// We may have a constructor initializer or function-try-block here.
if (kind == tok::colon || kind == tok::kw_try)
ConsumeAndStoreUntil(tok::l_brace, Toks);
else {
Toks.push_back(Tok);
ConsumeBrace();
}
// Consume everything up to (and including) the matching right brace.
ConsumeAndStoreUntil(tok::r_brace, Toks, /*StopAtSemi=*/false);
// If we're in a function-try-block, we need to store all the catch blocks.
if (kind == tok::kw_try) {
while (Tok.is(tok::kw_catch)) {
ConsumeAndStoreUntil(tok::l_brace, Toks, /*StopAtSemi=*/false);
ConsumeAndStoreUntil(tok::r_brace, Toks, /*StopAtSemi=*/false);
}
}
}
/// \brief Consume tokens and store them in the passed token container until
/// we've passed the try keyword and constructor initializers and have consumed
/// the opening brace of the function body. The opening brace will be consumed
/// if and only if there was no error.
///
/// \return True on error.
bool Parser::ConsumeAndStoreFunctionPrologue(CachedTokens &Toks) {
if (Tok.is(tok::kw_try)) {
Toks.push_back(Tok);
ConsumeToken();
}
bool ReadInitializer = false;
if (Tok.is(tok::colon)) {
// Initializers can contain braces too.
Toks.push_back(Tok);
ConsumeToken();
while (Tok.is(tok::identifier) || Tok.is(tok::coloncolon)) {
if (Tok.is(tok::eof) || Tok.is(tok::semi))
return Diag(Tok.getLocation(), diag::err_expected_lbrace);
// Grab the identifier.
if (!ConsumeAndStoreUntil(tok::l_paren, tok::l_brace, Toks,
/*StopAtSemi=*/true,
/*ConsumeFinalToken=*/false))
return Diag(Tok.getLocation(), diag::err_expected_lparen);
tok::TokenKind kind = Tok.getKind();
Toks.push_back(Tok);
bool IsLParen = (kind == tok::l_paren);
SourceLocation LOpen = Tok.getLocation();
if (IsLParen) {
ConsumeParen();
} else {
assert(kind == tok::l_brace && "Must be left paren or brace here.");
ConsumeBrace();
// In C++03, this has to be the start of the function body, which
// means the initializer is malformed; we'll diagnose it later.
if (!getLangOpts().CPlusPlus11)
return false;
}
// Grab the initializer
if (!ConsumeAndStoreUntil(IsLParen ? tok::r_paren : tok::r_brace,
Toks, /*StopAtSemi=*/true)) {
Diag(Tok, IsLParen ? diag::err_expected_rparen :
diag::err_expected_rbrace);
Diag(LOpen, diag::note_matching) << (IsLParen ? "(" : "{");
return true;
}
// Grab pack ellipsis, if present
if (Tok.is(tok::ellipsis)) {
Toks.push_back(Tok);
ConsumeToken();
}
// Grab the separating comma, if any.
if (Tok.is(tok::comma)) {
Toks.push_back(Tok);
ConsumeToken();
} else if (Tok.isNot(tok::l_brace)) {
ReadInitializer = true;
break;
}
}
}
// Grab any remaining garbage to be diagnosed later. We stop when we reach a
// brace: an opening one is the function body, while a closing one probably
// means we've reached the end of the class.
ConsumeAndStoreUntil(tok::l_brace, tok::r_brace, Toks,
/*StopAtSemi=*/true,
/*ConsumeFinalToken=*/false);
if (Tok.isNot(tok::l_brace)) {
if (ReadInitializer)
return Diag(Tok.getLocation(), diag::err_expected_lbrace_or_comma);
return Diag(Tok.getLocation(), diag::err_expected_lbrace);
}
Toks.push_back(Tok);
ConsumeBrace();
return false;
}
/// ConsumeAndStoreUntil - Consume and store the token at the passed token
/// container until the token 'T' is reached (which gets
/// consumed/stored too, if ConsumeFinalToken).
/// If StopAtSemi is true, then we will stop early at a ';' character.
/// Returns true if token 'T1' or 'T2' was found.
/// NOTE: This is a specialized version of Parser::SkipUntil.
bool Parser::ConsumeAndStoreUntil(tok::TokenKind T1, tok::TokenKind T2,
CachedTokens &Toks,
bool StopAtSemi, bool ConsumeFinalToken) {
// We always want this function to consume at least one token if the first
// token isn't T and if not at EOF.
bool isFirstTokenConsumed = true;
while (1) {
// If we found one of the tokens, stop and return true.
if (Tok.is(T1) || Tok.is(T2)) {
if (ConsumeFinalToken) {
Toks.push_back(Tok);
ConsumeAnyToken();
}
return true;
}
switch (Tok.getKind()) {
case tok::eof:
// Ran out of tokens.
return false;
case tok::l_paren:
// Recursively consume properly-nested parens.
Toks.push_back(Tok);
ConsumeParen();
ConsumeAndStoreUntil(tok::r_paren, Toks, /*StopAtSemi=*/false);
break;
case tok::l_square:
// Recursively consume properly-nested square brackets.
Toks.push_back(Tok);
ConsumeBracket();
ConsumeAndStoreUntil(tok::r_square, Toks, /*StopAtSemi=*/false);
break;
case tok::l_brace:
// Recursively consume properly-nested braces.
Toks.push_back(Tok);
ConsumeBrace();
ConsumeAndStoreUntil(tok::r_brace, Toks, /*StopAtSemi=*/false);
break;
// Okay, we found a ']' or '}' or ')', which we think should be balanced.
// Since the user wasn't looking for this token (if they were, it would
// already be handled), this isn't balanced. If there is a LHS token at a
// higher level, we will assume that this matches the unbalanced token
// and return it. Otherwise, this is a spurious RHS token, which we skip.
case tok::r_paren:
if (ParenCount && !isFirstTokenConsumed)
return false; // Matches something.
Toks.push_back(Tok);
ConsumeParen();
break;
case tok::r_square:
if (BracketCount && !isFirstTokenConsumed)
return false; // Matches something.
Toks.push_back(Tok);
ConsumeBracket();
break;
case tok::r_brace:
if (BraceCount && !isFirstTokenConsumed)
return false; // Matches something.
Toks.push_back(Tok);
ConsumeBrace();
break;
case tok::code_completion:
Toks.push_back(Tok);
ConsumeCodeCompletionToken();
break;
case tok::string_literal:
case tok::wide_string_literal:
case tok::utf8_string_literal:
case tok::utf16_string_literal:
case tok::utf32_string_literal:
Toks.push_back(Tok);
ConsumeStringToken();
break;
case tok::semi:
if (StopAtSemi)
return false;
// FALL THROUGH.
default:
// consume this token.
Toks.push_back(Tok);
ConsumeToken();
break;
}
isFirstTokenConsumed = false;
}
}
/// ConsumeAndStoreInitializer - Consume and store the token at the passed token
/// container until the end of the current initializer expression (either a
/// default argument or an in-class initializer for a non-static data member).
///
/// Returns \c true if we reached the end of something initializer-shaped,
/// \c false if we bailed out.
bool Parser::ConsumeAndStoreInitializer(CachedTokens &Toks,
CachedInitKind CIK) {
// We always want this function to consume at least one token if not at EOF.
bool IsFirstToken = true;
// Number of possible unclosed <s we've seen so far. These might be templates,
// and might not, but if there were none of them (or we know for sure that
// we're within a template), we can avoid a tentative parse.
unsigned AngleCount = 0;
unsigned KnownTemplateCount = 0;
while (1) {
switch (Tok.getKind()) {
case tok::comma:
// If we might be in a template, perform a tentative parse to check.
if (!AngleCount)
// Not a template argument: this is the end of the initializer.
return true;
if (KnownTemplateCount)
goto consume_token;
// We hit a comma inside angle brackets. This is the hard case. The
// rule we follow is:
// * For a default argument, if the tokens after the comma form a
// syntactically-valid parameter-declaration-clause, in which each
// parameter has an initializer, then this comma ends the default
// argument.
// * For a default initializer, if the tokens after the comma form a
// syntactically-valid init-declarator-list, then this comma ends
// the default initializer.
{
UnannotatedTentativeParsingAction PA(*this,
CIK == CIK_DefaultInitializer
? tok::semi : tok::r_paren);
Sema::TentativeAnalysisScope Scope(Actions);
TPResult Result = TPResult::Error;
ConsumeToken();
switch (CIK) {
case CIK_DefaultInitializer:
Result = TryParseInitDeclaratorList();
// If we parsed a complete, ambiguous init-declarator-list, this
// is only syntactically-valid if it's followed by a semicolon.
if (Result == TPResult::Ambiguous && Tok.isNot(tok::semi))
Result = TPResult::False;
break;
case CIK_DefaultArgument:
bool InvalidAsDeclaration = false;
Result = TryParseParameterDeclarationClause(
&InvalidAsDeclaration, /*VersusTemplateArgument=*/true);
// If this is an expression or a declaration with a missing
// 'typename', assume it's not a declaration.
if (Result == TPResult::Ambiguous && InvalidAsDeclaration)
Result = TPResult::False;
break;
}
// If what follows could be a declaration, it is a declaration.
if (Result != TPResult::False && Result != TPResult::Error) {
PA.Revert();
return true;
}
// In the uncommon case that we decide the following tokens are part
// of a template argument, revert any annotations we've performed in
// those tokens. We're not going to look them up until we've parsed
// the rest of the class, and that might add more declarations.
PA.RevertAnnotations();
}
// Keep going. We know we're inside a template argument list now.
++KnownTemplateCount;
goto consume_token;
case tok::eof:
case tok::annot_module_begin:
case tok::annot_module_end:
case tok::annot_module_include:
// Ran out of tokens.
return false;
case tok::less:
// FIXME: A '<' can only start a template-id if it's preceded by an
// identifier, an operator-function-id, or a literal-operator-id.
++AngleCount;
goto consume_token;
case tok::question:
// In 'a ? b : c', 'b' can contain an unparenthesized comma. If it does,
// that is *never* the end of the initializer. Skip to the ':'.
if (!ConsumeAndStoreConditional(Toks))
return false;
break;
case tok::greatergreatergreater:
if (!getLangOpts().CPlusPlus11)
goto consume_token;
if (AngleCount) --AngleCount;
if (KnownTemplateCount) --KnownTemplateCount;
// Fall through.
case tok::greatergreater:
if (!getLangOpts().CPlusPlus11)
goto consume_token;
if (AngleCount) --AngleCount;
if (KnownTemplateCount) --KnownTemplateCount;
// Fall through.
case tok::greater:
if (AngleCount) --AngleCount;
if (KnownTemplateCount) --KnownTemplateCount;
goto consume_token;
case tok::kw_template:
// 'template' identifier '<' is known to start a template argument list,
// and can be used to disambiguate the parse.
// FIXME: Support all forms of 'template' unqualified-id '<'.
Toks.push_back(Tok);
ConsumeToken();
if (Tok.is(tok::identifier)) {
Toks.push_back(Tok);
ConsumeToken();
if (Tok.is(tok::less)) {
++AngleCount;
++KnownTemplateCount;
Toks.push_back(Tok);
ConsumeToken();
}
}
break;
case tok::kw_operator:
// If 'operator' precedes other punctuation, that punctuation loses
// its special behavior.
Toks.push_back(Tok);
ConsumeToken();
switch (Tok.getKind()) {
case tok::comma:
case tok::greatergreatergreater:
case tok::greatergreater:
case tok::greater:
case tok::less:
Toks.push_back(Tok);
ConsumeToken();
break;
default:
break;
}
break;
case tok::l_paren:
// Recursively consume properly-nested parens.
Toks.push_back(Tok);
ConsumeParen();
ConsumeAndStoreUntil(tok::r_paren, Toks, /*StopAtSemi=*/false);
break;
case tok::l_square:
// Recursively consume properly-nested square brackets.
Toks.push_back(Tok);
ConsumeBracket();
ConsumeAndStoreUntil(tok::r_square, Toks, /*StopAtSemi=*/false);
break;
case tok::l_brace:
// Recursively consume properly-nested braces.
Toks.push_back(Tok);
ConsumeBrace();
ConsumeAndStoreUntil(tok::r_brace, Toks, /*StopAtSemi=*/false);
break;
// Okay, we found a ']' or '}' or ')', which we think should be balanced.
// Since the user wasn't looking for this token (if they were, it would
// already be handled), this isn't balanced. If there is a LHS token at a
// higher level, we will assume that this matches the unbalanced token
// and return it. Otherwise, this is a spurious RHS token, which we
// consume and pass on to downstream code to diagnose.
case tok::r_paren:
if (CIK == CIK_DefaultArgument)
return true; // End of the default argument.
if (ParenCount && !IsFirstToken)
return false;
Toks.push_back(Tok);
ConsumeParen();
continue;
case tok::r_square:
if (BracketCount && !IsFirstToken)
return false;
Toks.push_back(Tok);
ConsumeBracket();
continue;
case tok::r_brace:
if (BraceCount && !IsFirstToken)
return false;
Toks.push_back(Tok);
ConsumeBrace();
continue;
case tok::code_completion:
Toks.push_back(Tok);
ConsumeCodeCompletionToken();
break;
case tok::string_literal:
case tok::wide_string_literal:
case tok::utf8_string_literal:
case tok::utf16_string_literal:
case tok::utf32_string_literal:
Toks.push_back(Tok);
ConsumeStringToken();
break;
case tok::semi:
if (CIK == CIK_DefaultInitializer)
return true; // End of the default initializer.
// FALL THROUGH.
default:
consume_token:
Toks.push_back(Tok);
ConsumeToken();
break;
}
IsFirstToken = false;
}
}
/// \brief Consume tokens and store them in the passed token container until
/// we've passed the try keyword and constructor initializers and have consumed
/// the opening brace of the function body. The opening brace will be consumed
/// if and only if there was no error.
///
/// \return True on error.
bool Parser::ConsumeAndStoreFunctionPrologue(CachedTokens &Toks) {
if (Tok.is(tok::kw_try)) {
Toks.push_back(Tok);
ConsumeToken();
}
if (Tok.isNot(tok::colon)) {
// Easy case, just a function body.
// Grab any remaining garbage to be diagnosed later. We stop when we reach a
// brace: an opening one is the function body, while a closing one probably
// means we've reached the end of the class.
ConsumeAndStoreUntil(tok::l_brace, tok::r_brace, Toks,
/*StopAtSemi=*/true,
/*ConsumeFinalToken=*/false);
if (Tok.isNot(tok::l_brace))
return Diag(Tok.getLocation(), diag::err_expected) << tok::l_brace;
Toks.push_back(Tok);
ConsumeBrace();
return false;
}
Toks.push_back(Tok);
ConsumeToken();
// We can't reliably skip over a mem-initializer-id, because it could be
// a template-id involving not-yet-declared names. Given:
//
// S ( ) : a < b < c > ( e )
//
// 'e' might be an initializer or part of a template argument, depending
// on whether 'b' is a template.
// Track whether we might be inside a template argument. We can give
// significantly better diagnostics if we know that we're not.
bool MightBeTemplateArgument = false;
while (true) {
// Skip over the mem-initializer-id, if possible.
if (Tok.is(tok::kw_decltype)) {
Toks.push_back(Tok);
SourceLocation OpenLoc = ConsumeToken();
if (Tok.isNot(tok::l_paren))
return Diag(Tok.getLocation(), diag::err_expected_lparen_after)
<< "decltype";
Toks.push_back(Tok);
ConsumeParen();
if (!ConsumeAndStoreUntil(tok::r_paren, Toks, /*StopAtSemi=*/true)) {
Diag(Tok.getLocation(), diag::err_expected) << tok::r_paren;
Diag(OpenLoc, diag::note_matching) << tok::l_paren;
return true;
}
}
do {
// Walk over a component of a nested-name-specifier.
if (Tok.is(tok::coloncolon)) {
Toks.push_back(Tok);
ConsumeToken();
if (Tok.is(tok::kw_template)) {
Toks.push_back(Tok);
ConsumeToken();
}
}
if (Tok.is(tok::identifier) || Tok.is(tok::kw_template)) {
Toks.push_back(Tok);
ConsumeToken();
} else if (Tok.is(tok::code_completion)) {
Toks.push_back(Tok);
ConsumeCodeCompletionToken();
// Consume the rest of the initializers permissively.
// FIXME: We should be able to perform code-completion here even if
// there isn't a subsequent '{' token.
MightBeTemplateArgument = true;
break;
} else {
break;
}
} while (Tok.is(tok::coloncolon));
if (Tok.is(tok::less))
MightBeTemplateArgument = true;
if (MightBeTemplateArgument) {
// We may be inside a template argument list. Grab up to the start of the
// next parenthesized initializer or braced-init-list. This *might* be the
// initializer, or it might be a subexpression in the template argument
// list.
// FIXME: Count angle brackets, and clear MightBeTemplateArgument
// if all angles are closed.
if (!ConsumeAndStoreUntil(tok::l_paren, tok::l_brace, Toks,
/*StopAtSemi=*/true,
/*ConsumeFinalToken=*/false)) {
// We're not just missing the initializer, we're also missing the
// function body!
return Diag(Tok.getLocation(), diag::err_expected) << tok::l_brace;
}
} else if (Tok.isNot(tok::l_paren) && Tok.isNot(tok::l_brace)) {
// We found something weird in a mem-initializer-id.
if (getLangOpts().CPlusPlus11)
return Diag(Tok.getLocation(), diag::err_expected_either)
<< tok::l_paren << tok::l_brace;
else
return Diag(Tok.getLocation(), diag::err_expected) << tok::l_paren;
}
tok::TokenKind kind = Tok.getKind();
Toks.push_back(Tok);
bool IsLParen = (kind == tok::l_paren);
SourceLocation OpenLoc = Tok.getLocation();
if (IsLParen) {
ConsumeParen();
} else {
assert(kind == tok::l_brace && "Must be left paren or brace here.");
ConsumeBrace();
// In C++03, this has to be the start of the function body, which
// means the initializer is malformed; we'll diagnose it later.
if (!getLangOpts().CPlusPlus11)
return false;
}
// Grab the initializer (or the subexpression of the template argument).
// FIXME: If we support lambdas here, we'll need to set StopAtSemi to false
// if we might be inside the braces of a lambda-expression.
tok::TokenKind CloseKind = IsLParen ? tok::r_paren : tok::r_brace;
if (!ConsumeAndStoreUntil(CloseKind, Toks, /*StopAtSemi=*/true)) {
Diag(Tok, diag::err_expected) << CloseKind;
Diag(OpenLoc, diag::note_matching) << kind;
return true;
}
// Grab pack ellipsis, if present.
if (Tok.is(tok::ellipsis)) {
Toks.push_back(Tok);
ConsumeToken();
}
// If we know we just consumed a mem-initializer, we must have ',' or '{'
// next.
if (Tok.is(tok::comma)) {
Toks.push_back(Tok);
ConsumeToken();
} else if (Tok.is(tok::l_brace)) {
// This is the function body if the ')' or '}' is immediately followed by
// a '{'. That cannot happen within a template argument, apart from the
// case where a template argument contains a compound literal:
//
// S ( ) : a < b < c > ( d ) { }
// // End of declaration, or still inside the template argument?
//
// ... and the case where the template argument contains a lambda:
//
// S ( ) : a < 0 && b < c > ( d ) + [ ] ( ) { return 0; }
// ( ) > ( ) { }
//
// FIXME: Disambiguate these cases. Note that the latter case is probably
// going to be made ill-formed by core issue 1607.
Toks.push_back(Tok);
ConsumeBrace();
return false;
} else if (!MightBeTemplateArgument) {
return Diag(Tok.getLocation(), diag::err_expected_either) << tok::l_brace
<< tok::comma;
}
}
}
