/// StringifyArgument - Implement C99 6.10.3.2p2, converting a sequence of
/// tokens into the literal string token that should be produced by the C #
/// preprocessor operator. If Charify is true, then it should be turned into
/// a character literal for the Microsoft charize (#@) extension.
///
Token MacroArgs::StringifyArgument(const Token *ArgToks,
Preprocessor &PP, bool Charify,
SourceLocation ExpansionLocStart,
SourceLocation ExpansionLocEnd) {
Token Tok;
Tok.startToken();
Tok.setKind(Charify ? tok::char_constant : tok::string_literal);
const Token *ArgTokStart = ArgToks;
// Stringify all the tokens.
SmallString<128> Result;
Result += "\"";
bool isFirst = true;
for (; ArgToks->isNot(tok::eof); ++ArgToks) {
const Token &Tok = *ArgToks;
if (!isFirst && (Tok.hasLeadingSpace() || Tok.isAtStartOfLine()))
Result += ' ';
isFirst = false;
// If this is a string or character constant, escape the token as specified
// by 6.10.3.2p2.
if (tok::isStringLiteral(Tok.getKind()) || // "foo", u8R"x(foo)x"_bar, etc.
Tok.is(tok::char_constant) || // 'x'
Tok.is(tok::wide_char_constant) || // L'x'.
Tok.is(tok::utf8_char_constant) || // u8'x'.
Tok.is(tok::utf16_char_constant) || // u'x'.
Tok.is(tok::utf32_char_constant)) { // U'x'.
bool Invalid = false;
std::string TokStr = PP.getSpelling(Tok, &Invalid);
if (!Invalid) {
std::string Str = Lexer::Stringify(TokStr);
Result.append(Str.begin(), Str.end());
}
} else if (Tok.is(tok::code_completion)) {
PP.CodeCompleteNaturalLanguage();
} else {
// Otherwise, just append the token. Do some gymnastics to get the token
// in place and avoid copies where possible.
unsigned CurStrLen = Result.size();
Result.resize(CurStrLen+Tok.getLength());
const char *BufPtr = Result.data() + CurStrLen;
bool Invalid = false;
unsigned ActualTokLen = PP.getSpelling(Tok, BufPtr, &Invalid);
if (!Invalid) {
// If getSpelling returned a pointer to an already uniqued version of
// the string instead of filling in BufPtr, memcpy it onto our string.
if (ActualTokLen && BufPtr != &Result[CurStrLen])
memcpy(&Result[CurStrLen], BufPtr, ActualTokLen);
// If the token was dirty, the spelling may be shorter than the token.
if (ActualTokLen != Tok.getLength())
Result.resize(CurStrLen+ActualTokLen);
}
}
}
// If the last character of the string is a \, and if it isn't escaped, this
// is an invalid string literal, diagnose it as specified in C99.
if (Result.back() == '\\') {
// Count the number of consecutive \ characters. If even, then they are
// just escaped backslashes, otherwise it's an error.
unsigned FirstNonSlash = Result.size()-2;
// Guaranteed to find the starting " if nothing else.
while (Result[FirstNonSlash] == '\\')
--FirstNonSlash;
if ((Result.size()-1-FirstNonSlash) & 1) {
// Diagnose errors for things like: #define F(X) #X / F(\)
PP.Diag(ArgToks[-1], diag::pp_invalid_string_literal);
Result.pop_back(); // remove one of the \'s.
}
}
Result += '"';
// If this is the charify operation and the result is not a legal character
// constant, diagnose it.
if (Charify) {
// First step, turn double quotes into single quotes:
Result[0] = '\'';
Result[Result.size()-1] = '\'';
// Check for bogus character.
bool isBad = false;
if (Result.size() == 3)
isBad = Result[1] == '\''; // ''' is not legal. '\' already fixed above.
else
isBad = (Result.size() != 4 || Result[1] != '\\'); // Not '\x'
if (isBad) {
PP.Diag(ArgTokStart[0], diag::err_invalid_character_to_charify);
Result = "' '"; // Use something arbitrary, but legal.
}
}
PP.CreateString(Result, Tok,
ExpansionLocStart, ExpansionLocEnd);
return Tok;
}
