
This is a major change on how we represent nested name qualifications in the AST. * The nested name specifier itself and how it's stored is changed. The prefixes for types are handled within the type hierarchy, which makes canonicalization for them super cheap, no memory allocation required. Also translating a type into nested name specifier form becomes a no-op. An identifier is stored as a DependentNameType. The nested name specifier gains a lightweight handle class, to be used instead of passing around pointers, which is similar to what is implemented for TemplateName. There is still one free bit available, and this handle can be used within a PointerUnion and PointerIntPair, which should keep bit-packing aficionados happy. * The ElaboratedType node is removed, all type nodes in which it could previously apply to can now store the elaborated keyword and name qualifier, tail allocating when present. * TagTypes can now point to the exact declaration found when producing these, as opposed to the previous situation of there only existing one TagType per entity. This increases the amount of type sugar retained, and can have several applications, for example in tracking module ownership, and other tools which care about source file origins, such as IWYU. These TagTypes are lazily allocated, in order to limit the increase in AST size. This patch offers a great performance benefit. It greatly improves compilation time for [stdexec](https://github.com/NVIDIA/stdexec). For one datapoint, for `test_on2.cpp` in that project, which is the slowest compiling test, this patch improves `-c` compilation time by about 7.2%, with the `-fsyntax-only` improvement being at ~12%. This has great results on compile-time-tracker as well:  This patch also further enables other optimziations in the future, and will reduce the performance impact of template specialization resugaring when that lands. It has some other miscelaneous drive-by fixes. About the review: Yes the patch is huge, sorry about that. Part of the reason is that I started by the nested name specifier part, before the ElaboratedType part, but that had a huge performance downside, as ElaboratedType is a big performance hog. I didn't have the steam to go back and change the patch after the fact. There is also a lot of internal API changes, and it made sense to remove ElaboratedType in one go, versus removing it from one type at a time, as that would present much more churn to the users. Also, the nested name specifier having a different API avoids missing changes related to how prefixes work now, which could make existing code compile but not work. How to review: The important changes are all in `clang/include/clang/AST` and `clang/lib/AST`, with also important changes in `clang/lib/Sema/TreeTransform.h`. The rest and bulk of the changes are mostly consequences of the changes in API. PS: TagType::getDecl is renamed to `getOriginalDecl` in this patch, just for easier to rebasing. I plan to rename it back after this lands. Fixes #136624 Fixes https://github.com/llvm/llvm-project/issues/43179 Fixes https://github.com/llvm/llvm-project/issues/68670 Fixes https://github.com/llvm/llvm-project/issues/92757
819 lines
31 KiB
C++
819 lines
31 KiB
C++
//===--- FormatStringConverter.cpp - clang-tidy----------------------------===//
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
///
|
|
/// \file
|
|
/// Implementation of the FormatStringConverter class which is used to convert
|
|
/// printf format strings to C++ std::formatter format strings.
|
|
///
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "FormatStringConverter.h"
|
|
#include "../utils/FixItHintUtils.h"
|
|
#include "clang/AST/Expr.h"
|
|
#include "clang/ASTMatchers/ASTMatchFinder.h"
|
|
#include "clang/Basic/LangOptions.h"
|
|
#include "clang/Lex/Lexer.h"
|
|
#include "clang/Lex/Preprocessor.h"
|
|
#include "clang/Tooling/FixIt.h"
|
|
#include "llvm/ADT/StringExtras.h"
|
|
#include "llvm/Support/Debug.h"
|
|
|
|
using namespace clang::ast_matchers;
|
|
using namespace clang::analyze_printf;
|
|
|
|
namespace clang::tidy::utils {
|
|
using clang::analyze_format_string::ConversionSpecifier;
|
|
|
|
/// Is the passed type the actual "char" type, whether that be signed or
|
|
/// unsigned, rather than explicit signed char or unsigned char types.
|
|
static bool isRealCharType(const clang::QualType &Ty) {
|
|
using namespace clang;
|
|
const Type *DesugaredType = Ty->getUnqualifiedDesugaredType();
|
|
if (const auto *BT = llvm::dyn_cast<BuiltinType>(DesugaredType))
|
|
return (BT->getKind() == BuiltinType::Char_U ||
|
|
BT->getKind() == BuiltinType::Char_S);
|
|
return false;
|
|
}
|
|
|
|
/// If possible, return the text name of the signed type that corresponds to the
|
|
/// passed integer type. If the passed type is already signed then its name is
|
|
/// just returned. Only supports BuiltinTypes.
|
|
static std::optional<std::string>
|
|
getCorrespondingSignedTypeName(const clang::QualType &QT) {
|
|
using namespace clang;
|
|
const auto UQT = QT.getUnqualifiedType();
|
|
if (const auto *BT = llvm::dyn_cast<BuiltinType>(UQT)) {
|
|
switch (BT->getKind()) {
|
|
case BuiltinType::UChar:
|
|
case BuiltinType::Char_U:
|
|
case BuiltinType::SChar:
|
|
case BuiltinType::Char_S:
|
|
return "signed char";
|
|
case BuiltinType::UShort:
|
|
case BuiltinType::Short:
|
|
return "short";
|
|
case BuiltinType::UInt:
|
|
case BuiltinType::Int:
|
|
return "int";
|
|
case BuiltinType::ULong:
|
|
case BuiltinType::Long:
|
|
return "long";
|
|
case BuiltinType::ULongLong:
|
|
case BuiltinType::LongLong:
|
|
return "long long";
|
|
default:
|
|
llvm::dbgs() << "Unknown corresponding signed type for BuiltinType '"
|
|
<< QT.getAsString() << "'\n";
|
|
return std::nullopt;
|
|
}
|
|
}
|
|
|
|
// Deal with fixed-width integer types from <cstdint>. Use std:: prefix only
|
|
// if the argument type does.
|
|
const std::string TypeName = UQT.getAsString();
|
|
StringRef SimplifiedTypeName{TypeName};
|
|
const bool InStd = SimplifiedTypeName.consume_front("std::");
|
|
const StringRef Prefix = InStd ? "std::" : "";
|
|
|
|
if (SimplifiedTypeName.starts_with("uint") &&
|
|
SimplifiedTypeName.ends_with("_t"))
|
|
return (Twine(Prefix) + SimplifiedTypeName.drop_front()).str();
|
|
|
|
if (SimplifiedTypeName == "size_t")
|
|
return (Twine(Prefix) + "ssize_t").str();
|
|
|
|
llvm::dbgs() << "Unknown corresponding signed type for non-BuiltinType '"
|
|
<< UQT.getAsString() << "'\n";
|
|
return std::nullopt;
|
|
}
|
|
|
|
/// If possible, return the text name of the unsigned type that corresponds to
|
|
/// the passed integer type. If the passed type is already unsigned then its
|
|
/// name is just returned. Only supports BuiltinTypes.
|
|
static std::optional<std::string>
|
|
getCorrespondingUnsignedTypeName(const clang::QualType &QT) {
|
|
using namespace clang;
|
|
const auto UQT = QT.getUnqualifiedType();
|
|
if (const auto *BT = llvm::dyn_cast<BuiltinType>(UQT)) {
|
|
switch (BT->getKind()) {
|
|
case BuiltinType::SChar:
|
|
case BuiltinType::Char_S:
|
|
case BuiltinType::UChar:
|
|
case BuiltinType::Char_U:
|
|
return "unsigned char";
|
|
case BuiltinType::Short:
|
|
case BuiltinType::UShort:
|
|
return "unsigned short";
|
|
case BuiltinType::Int:
|
|
case BuiltinType::UInt:
|
|
return "unsigned int";
|
|
case BuiltinType::Long:
|
|
case BuiltinType::ULong:
|
|
return "unsigned long";
|
|
case BuiltinType::LongLong:
|
|
case BuiltinType::ULongLong:
|
|
return "unsigned long long";
|
|
default:
|
|
llvm::dbgs() << "Unknown corresponding unsigned type for BuiltinType '"
|
|
<< UQT.getAsString() << "'\n";
|
|
return std::nullopt;
|
|
}
|
|
}
|
|
|
|
// Deal with fixed-width integer types from <cstdint>. Use std:: prefix only
|
|
// if the argument type does.
|
|
const std::string TypeName = UQT.getAsString();
|
|
StringRef SimplifiedTypeName{TypeName};
|
|
const bool InStd = SimplifiedTypeName.consume_front("std::");
|
|
const StringRef Prefix = InStd ? "std::" : "";
|
|
|
|
if (SimplifiedTypeName.starts_with("int") &&
|
|
SimplifiedTypeName.ends_with("_t"))
|
|
return (Twine(Prefix) + "u" + SimplifiedTypeName).str();
|
|
|
|
if (SimplifiedTypeName == "ssize_t")
|
|
return (Twine(Prefix) + "size_t").str();
|
|
if (SimplifiedTypeName == "ptrdiff_t")
|
|
return (Twine(Prefix) + "size_t").str();
|
|
|
|
llvm::dbgs() << "Unknown corresponding unsigned type for non-BuiltinType '"
|
|
<< UQT.getAsString() << "'\n";
|
|
return std::nullopt;
|
|
}
|
|
|
|
static std::optional<std::string>
|
|
castTypeForArgument(ConversionSpecifier::Kind ArgKind,
|
|
const clang::QualType &QT) {
|
|
if (ArgKind == ConversionSpecifier::Kind::uArg)
|
|
return getCorrespondingUnsignedTypeName(QT);
|
|
return getCorrespondingSignedTypeName(QT);
|
|
}
|
|
|
|
static bool isMatchingSignedness(ConversionSpecifier::Kind ArgKind,
|
|
const clang::QualType &ArgType) {
|
|
if (const auto *BT = llvm::dyn_cast<BuiltinType>(ArgType)) {
|
|
// Unadorned char never matches any expected signedness since it
|
|
// could be signed or unsigned.
|
|
const auto ArgTypeKind = BT->getKind();
|
|
if (ArgTypeKind == BuiltinType::Char_U ||
|
|
ArgTypeKind == BuiltinType::Char_S)
|
|
return false;
|
|
}
|
|
|
|
if (ArgKind == ConversionSpecifier::Kind::uArg)
|
|
return ArgType->isUnsignedIntegerType();
|
|
return ArgType->isSignedIntegerType();
|
|
}
|
|
|
|
namespace {
|
|
AST_MATCHER(clang::QualType, isRealChar) {
|
|
return clang::tidy::utils::isRealCharType(Node);
|
|
}
|
|
} // namespace
|
|
|
|
static bool castMismatchedIntegerTypes(const CallExpr *Call, bool StrictMode) {
|
|
/// For printf-style functions, the signedness of the type printed is
|
|
/// indicated by the corresponding type in the format string.
|
|
/// std::print will determine the signedness from the type of the
|
|
/// argument. This means that it is necessary to generate a cast in
|
|
/// StrictMode to ensure that the exact behaviour is maintained.
|
|
/// However, for templated functions like absl::PrintF and
|
|
/// fmt::printf, the signedness of the type printed is also taken from
|
|
/// the actual argument like std::print, so such casts are never
|
|
/// necessary. printf-style functions are variadic, whereas templated
|
|
/// ones aren't, so we can use that to distinguish between the two
|
|
/// cases.
|
|
if (StrictMode) {
|
|
const FunctionDecl *FuncDecl = Call->getDirectCallee();
|
|
assert(FuncDecl);
|
|
return FuncDecl->isVariadic();
|
|
}
|
|
return false;
|
|
}
|
|
|
|
FormatStringConverter::FormatStringConverter(
|
|
ASTContext *ContextIn, const CallExpr *Call, unsigned FormatArgOffset,
|
|
const Configuration ConfigIn, const LangOptions &LO, SourceManager &SM,
|
|
Preprocessor &PP)
|
|
: Context(ContextIn), Config(ConfigIn),
|
|
CastMismatchedIntegerTypes(
|
|
castMismatchedIntegerTypes(Call, ConfigIn.StrictMode)),
|
|
Args(Call->getArgs()), NumArgs(Call->getNumArgs()),
|
|
ArgsOffset(FormatArgOffset + 1), LangOpts(LO) {
|
|
assert(ArgsOffset <= NumArgs);
|
|
FormatExpr = llvm::dyn_cast<StringLiteral>(
|
|
Args[FormatArgOffset]->IgnoreUnlessSpelledInSource());
|
|
|
|
assert(FormatExpr && FormatExpr->isOrdinary());
|
|
|
|
if (const std::optional<StringRef> MaybeMacroName =
|
|
formatStringContainsUnreplaceableMacro(Call, FormatExpr, SM, PP);
|
|
MaybeMacroName) {
|
|
conversionNotPossible(
|
|
("format string contains unreplaceable macro '" + *MaybeMacroName + "'")
|
|
.str());
|
|
return;
|
|
}
|
|
|
|
PrintfFormatString = FormatExpr->getString();
|
|
|
|
// Assume that the output will be approximately the same size as the input,
|
|
// but perhaps with a few escapes expanded.
|
|
const size_t EstimatedGrowth = 8;
|
|
StandardFormatString.reserve(PrintfFormatString.size() + EstimatedGrowth);
|
|
StandardFormatString.push_back('\"');
|
|
|
|
const bool IsFreeBsdkPrintf = false;
|
|
|
|
using clang::analyze_format_string::ParsePrintfString;
|
|
ParsePrintfString(*this, PrintfFormatString.data(),
|
|
PrintfFormatString.data() + PrintfFormatString.size(),
|
|
LangOpts, Context->getTargetInfo(), IsFreeBsdkPrintf);
|
|
finalizeFormatText();
|
|
}
|
|
|
|
std::optional<StringRef>
|
|
FormatStringConverter::formatStringContainsUnreplaceableMacro(
|
|
const CallExpr *Call, const StringLiteral *FormatExpr, SourceManager &SM,
|
|
Preprocessor &PP) {
|
|
// If a macro invocation surrounds the entire call then we don't want that to
|
|
// inhibit conversion. The whole format string will appear to come from that
|
|
// macro, as will the function call.
|
|
std::optional<StringRef> MaybeSurroundingMacroName;
|
|
if (SourceLocation BeginCallLoc = Call->getBeginLoc();
|
|
BeginCallLoc.isMacroID())
|
|
MaybeSurroundingMacroName =
|
|
Lexer::getImmediateMacroName(BeginCallLoc, SM, PP.getLangOpts());
|
|
|
|
for (auto I = FormatExpr->tokloc_begin(), E = FormatExpr->tokloc_end();
|
|
I != E; ++I) {
|
|
const SourceLocation &TokenLoc = *I;
|
|
if (TokenLoc.isMacroID()) {
|
|
const StringRef MacroName =
|
|
Lexer::getImmediateMacroName(TokenLoc, SM, PP.getLangOpts());
|
|
|
|
if (MaybeSurroundingMacroName != MacroName) {
|
|
// glibc uses __PRI64_PREFIX and __PRIPTR_PREFIX to define the prefixes
|
|
// for types that change size so we must look for multiple prefixes.
|
|
if (!MacroName.starts_with("PRI") && !MacroName.starts_with("__PRI"))
|
|
return MacroName;
|
|
|
|
const SourceLocation TokenSpellingLoc = SM.getSpellingLoc(TokenLoc);
|
|
const OptionalFileEntryRef MaybeFileEntry =
|
|
SM.getFileEntryRefForID(SM.getFileID(TokenSpellingLoc));
|
|
if (!MaybeFileEntry)
|
|
return MacroName;
|
|
|
|
HeaderSearch &HS = PP.getHeaderSearchInfo();
|
|
// Check if the file is a system header
|
|
if (!isSystem(HS.getFileDirFlavor(*MaybeFileEntry)) ||
|
|
llvm::sys::path::filename(MaybeFileEntry->getName()) !=
|
|
"inttypes.h")
|
|
return MacroName;
|
|
}
|
|
}
|
|
}
|
|
return std::nullopt;
|
|
}
|
|
|
|
void FormatStringConverter::emitAlignment(const PrintfSpecifier &FS,
|
|
std::string &FormatSpec) {
|
|
ConversionSpecifier::Kind ArgKind = FS.getConversionSpecifier().getKind();
|
|
|
|
// We only care about alignment if a field width is specified
|
|
if (FS.getFieldWidth().getHowSpecified() != OptionalAmount::NotSpecified) {
|
|
if (ArgKind == ConversionSpecifier::sArg) {
|
|
// Strings are left-aligned by default with std::format, so we only
|
|
// need to emit an alignment if this one needs to be right aligned.
|
|
if (!FS.isLeftJustified())
|
|
FormatSpec.push_back('>');
|
|
} else {
|
|
// Numbers are right-aligned by default with std::format, so we only
|
|
// need to emit an alignment if this one needs to be left aligned.
|
|
if (FS.isLeftJustified())
|
|
FormatSpec.push_back('<');
|
|
}
|
|
}
|
|
}
|
|
|
|
void FormatStringConverter::emitSign(const PrintfSpecifier &FS,
|
|
std::string &FormatSpec) {
|
|
const ConversionSpecifier Spec = FS.getConversionSpecifier();
|
|
|
|
// Ignore on something that isn't numeric. For printf it's would be a
|
|
// compile-time warning but ignored at runtime, but for std::format it
|
|
// ought to be a compile-time error.
|
|
if (Spec.isAnyIntArg() || Spec.isDoubleArg()) {
|
|
// + is preferred to ' '
|
|
if (FS.hasPlusPrefix())
|
|
FormatSpec.push_back('+');
|
|
else if (FS.hasSpacePrefix())
|
|
FormatSpec.push_back(' ');
|
|
}
|
|
}
|
|
|
|
void FormatStringConverter::emitAlternativeForm(const PrintfSpecifier &FS,
|
|
std::string &FormatSpec) {
|
|
if (FS.hasAlternativeForm()) {
|
|
switch (FS.getConversionSpecifier().getKind()) {
|
|
case ConversionSpecifier::Kind::aArg:
|
|
case ConversionSpecifier::Kind::AArg:
|
|
case ConversionSpecifier::Kind::eArg:
|
|
case ConversionSpecifier::Kind::EArg:
|
|
case ConversionSpecifier::Kind::fArg:
|
|
case ConversionSpecifier::Kind::FArg:
|
|
case ConversionSpecifier::Kind::gArg:
|
|
case ConversionSpecifier::Kind::GArg:
|
|
case ConversionSpecifier::Kind::xArg:
|
|
case ConversionSpecifier::Kind::XArg:
|
|
case ConversionSpecifier::Kind::oArg:
|
|
FormatSpec.push_back('#');
|
|
break;
|
|
default:
|
|
// Alternative forms don't exist for other argument kinds
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
void FormatStringConverter::emitFieldWidth(const PrintfSpecifier &FS,
|
|
std::string &FormatSpec) {
|
|
{
|
|
const OptionalAmount FieldWidth = FS.getFieldWidth();
|
|
switch (FieldWidth.getHowSpecified()) {
|
|
case OptionalAmount::NotSpecified:
|
|
break;
|
|
case OptionalAmount::Constant:
|
|
FormatSpec.append(llvm::utostr(FieldWidth.getConstantAmount()));
|
|
break;
|
|
case OptionalAmount::Arg:
|
|
FormatSpec.push_back('{');
|
|
if (FieldWidth.usesPositionalArg()) {
|
|
// std::format argument identifiers are zero-based, whereas printf
|
|
// ones are one based.
|
|
assert(FieldWidth.getPositionalArgIndex() > 0U);
|
|
FormatSpec.append(llvm::utostr(FieldWidth.getPositionalArgIndex() - 1));
|
|
}
|
|
FormatSpec.push_back('}');
|
|
break;
|
|
case OptionalAmount::Invalid:
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
void FormatStringConverter::emitPrecision(const PrintfSpecifier &FS,
|
|
std::string &FormatSpec) {
|
|
const OptionalAmount FieldPrecision = FS.getPrecision();
|
|
switch (FieldPrecision.getHowSpecified()) {
|
|
case OptionalAmount::NotSpecified:
|
|
break;
|
|
case OptionalAmount::Constant:
|
|
FormatSpec.push_back('.');
|
|
FormatSpec.append(llvm::utostr(FieldPrecision.getConstantAmount()));
|
|
break;
|
|
case OptionalAmount::Arg:
|
|
FormatSpec.push_back('.');
|
|
FormatSpec.push_back('{');
|
|
if (FieldPrecision.usesPositionalArg()) {
|
|
// std::format argument identifiers are zero-based, whereas printf
|
|
// ones are one based.
|
|
assert(FieldPrecision.getPositionalArgIndex() > 0U);
|
|
FormatSpec.append(
|
|
llvm::utostr(FieldPrecision.getPositionalArgIndex() - 1));
|
|
}
|
|
FormatSpec.push_back('}');
|
|
break;
|
|
case OptionalAmount::Invalid:
|
|
break;
|
|
}
|
|
}
|
|
|
|
void FormatStringConverter::maybeRotateArguments(const PrintfSpecifier &FS) {
|
|
unsigned ArgCount = 0;
|
|
const OptionalAmount FieldWidth = FS.getFieldWidth();
|
|
const OptionalAmount FieldPrecision = FS.getPrecision();
|
|
|
|
if (FieldWidth.getHowSpecified() == OptionalAmount::Arg &&
|
|
!FieldWidth.usesPositionalArg())
|
|
++ArgCount;
|
|
if (FieldPrecision.getHowSpecified() == OptionalAmount::Arg &&
|
|
!FieldPrecision.usesPositionalArg())
|
|
++ArgCount;
|
|
|
|
if (ArgCount)
|
|
ArgRotates.emplace_back(FS.getArgIndex() + ArgsOffset, ArgCount);
|
|
}
|
|
|
|
void FormatStringConverter::emitStringArgument(unsigned ArgIndex,
|
|
const Expr *Arg) {
|
|
// If the argument is the result of a call to std::string::c_str() or
|
|
// data() with a return type of char then we can remove that call and
|
|
// pass the std::string directly. We don't want to do so if the return
|
|
// type is not a char pointer (though it's unlikely that such code would
|
|
// compile without warnings anyway.) See RedundantStringCStrCheck.
|
|
|
|
if (!StringCStrCallExprMatcher) {
|
|
// Lazily create the matcher
|
|
const auto StringDecl = type(hasUnqualifiedDesugaredType(recordType(
|
|
hasDeclaration(cxxRecordDecl(hasName("::std::basic_string"))))));
|
|
const auto StringExpr = expr(
|
|
anyOf(hasType(StringDecl), hasType(qualType(pointsTo(StringDecl)))));
|
|
|
|
StringCStrCallExprMatcher =
|
|
cxxMemberCallExpr(
|
|
on(StringExpr.bind("arg")), callee(memberExpr().bind("member")),
|
|
callee(cxxMethodDecl(hasAnyName("c_str", "data"),
|
|
returns(pointerType(pointee(isRealChar()))))))
|
|
.bind("call");
|
|
}
|
|
|
|
auto CStrMatches = match(*StringCStrCallExprMatcher, *Arg, *Context);
|
|
if (CStrMatches.size() == 1)
|
|
ArgCStrRemovals.push_back(CStrMatches.front());
|
|
else if (Arg->getType()->isPointerType()) {
|
|
const QualType Pointee = Arg->getType()->getPointeeType();
|
|
// printf is happy to print signed char and unsigned char strings, but
|
|
// std::format only likes char strings.
|
|
if (Pointee->isCharType() && !isRealCharType(Pointee))
|
|
ArgFixes.emplace_back(ArgIndex, "reinterpret_cast<const char *>(");
|
|
}
|
|
}
|
|
|
|
bool FormatStringConverter::emitIntegerArgument(
|
|
ConversionSpecifier::Kind ArgKind, const Expr *Arg, unsigned ArgIndex,
|
|
std::string &FormatSpec) {
|
|
const clang::QualType &ArgType = Arg->getType();
|
|
if (ArgType->isBooleanType()) {
|
|
// std::format will print bool as either "true" or "false" by default,
|
|
// but printf prints them as "0" or "1". Be compatible with printf by
|
|
// requesting decimal output.
|
|
FormatSpec.push_back('d');
|
|
} else if (ArgType->isEnumeralType()) {
|
|
// std::format will try to find a specialization to print the enum
|
|
// (and probably fail), whereas printf would have just expected it to
|
|
// be passed as its underlying type. However, printf will have forced
|
|
// the signedness based on the format string, so we need to do the
|
|
// same.
|
|
if (const auto *ET = ArgType->getAs<EnumType>()) {
|
|
if (const std::optional<std::string> MaybeCastType = castTypeForArgument(
|
|
ArgKind,
|
|
ET->getOriginalDecl()->getDefinitionOrSelf()->getIntegerType()))
|
|
ArgFixes.emplace_back(
|
|
ArgIndex, (Twine("static_cast<") + *MaybeCastType + ">(").str());
|
|
else
|
|
return conversionNotPossible(
|
|
(Twine("argument ") + Twine(ArgIndex) + " has unexpected enum type")
|
|
.str());
|
|
}
|
|
} else if (CastMismatchedIntegerTypes &&
|
|
!isMatchingSignedness(ArgKind, ArgType)) {
|
|
// printf will happily print an unsigned type as signed if told to.
|
|
// Even -Wformat doesn't warn for this. std::format will format as
|
|
// unsigned unless we cast it.
|
|
if (const std::optional<std::string> MaybeCastType =
|
|
castTypeForArgument(ArgKind, ArgType))
|
|
ArgFixes.emplace_back(
|
|
ArgIndex, (Twine("static_cast<") + *MaybeCastType + ">(").str());
|
|
else
|
|
return conversionNotPossible(
|
|
(Twine("argument ") + Twine(ArgIndex) + " cannot be cast to " +
|
|
Twine(ArgKind == ConversionSpecifier::Kind::uArg ? "unsigned"
|
|
: "signed") +
|
|
" integer type to match format"
|
|
" specifier and StrictMode is enabled")
|
|
.str());
|
|
} else if (isRealCharType(ArgType) || !ArgType->isIntegerType()) {
|
|
// Only specify integer if the argument is of a different type
|
|
FormatSpec.push_back('d');
|
|
}
|
|
return true;
|
|
}
|
|
|
|
/// Append the corresponding standard format string type fragment to FormatSpec,
|
|
/// and store any argument fixes for later application.
|
|
/// @returns true on success, false on failure
|
|
bool FormatStringConverter::emitType(const PrintfSpecifier &FS, const Expr *Arg,
|
|
std::string &FormatSpec) {
|
|
ConversionSpecifier::Kind ArgKind = FS.getConversionSpecifier().getKind();
|
|
switch (ArgKind) {
|
|
case ConversionSpecifier::Kind::sArg:
|
|
emitStringArgument(FS.getArgIndex() + ArgsOffset, Arg);
|
|
break;
|
|
case ConversionSpecifier::Kind::cArg:
|
|
// The type must be "c" to get a character unless the type is exactly
|
|
// char (whether that be signed or unsigned for the target.)
|
|
if (!isRealCharType(Arg->getType()))
|
|
FormatSpec.push_back('c');
|
|
break;
|
|
case ConversionSpecifier::Kind::dArg:
|
|
case ConversionSpecifier::Kind::iArg:
|
|
case ConversionSpecifier::Kind::uArg:
|
|
if (!emitIntegerArgument(ArgKind, Arg, FS.getArgIndex() + ArgsOffset,
|
|
FormatSpec))
|
|
return false;
|
|
break;
|
|
case ConversionSpecifier::Kind::pArg: {
|
|
const clang::QualType &ArgType = Arg->getType();
|
|
// std::format knows how to format void pointers and nullptrs
|
|
if (!ArgType->isNullPtrType() && !ArgType->isVoidPointerType())
|
|
ArgFixes.emplace_back(FS.getArgIndex() + ArgsOffset,
|
|
"static_cast<const void *>(");
|
|
break;
|
|
}
|
|
case ConversionSpecifier::Kind::xArg:
|
|
FormatSpec.push_back('x');
|
|
break;
|
|
case ConversionSpecifier::Kind::XArg:
|
|
FormatSpec.push_back('X');
|
|
break;
|
|
case ConversionSpecifier::Kind::oArg:
|
|
FormatSpec.push_back('o');
|
|
break;
|
|
case ConversionSpecifier::Kind::aArg:
|
|
FormatSpec.push_back('a');
|
|
break;
|
|
case ConversionSpecifier::Kind::AArg:
|
|
FormatSpec.push_back('A');
|
|
break;
|
|
case ConversionSpecifier::Kind::eArg:
|
|
FormatSpec.push_back('e');
|
|
break;
|
|
case ConversionSpecifier::Kind::EArg:
|
|
FormatSpec.push_back('E');
|
|
break;
|
|
case ConversionSpecifier::Kind::fArg:
|
|
FormatSpec.push_back('f');
|
|
break;
|
|
case ConversionSpecifier::Kind::FArg:
|
|
FormatSpec.push_back('F');
|
|
break;
|
|
case ConversionSpecifier::Kind::gArg:
|
|
FormatSpec.push_back('g');
|
|
break;
|
|
case ConversionSpecifier::Kind::GArg:
|
|
FormatSpec.push_back('G');
|
|
break;
|
|
default:
|
|
// Something we don't understand
|
|
return conversionNotPossible((Twine("argument ") +
|
|
Twine(FS.getArgIndex() + ArgsOffset) +
|
|
" has an unsupported format specifier")
|
|
.str());
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
/// Append the standard format string equivalent of the passed PrintfSpecifier
|
|
/// to StandardFormatString and store any argument fixes for later application.
|
|
/// @returns true on success, false on failure
|
|
bool FormatStringConverter::convertArgument(const PrintfSpecifier &FS,
|
|
const Expr *Arg,
|
|
std::string &StandardFormatString) {
|
|
// The specifier must have an associated argument
|
|
assert(FS.consumesDataArgument());
|
|
|
|
StandardFormatString.push_back('{');
|
|
|
|
if (FS.usesPositionalArg()) {
|
|
// std::format argument identifiers are zero-based, whereas printf ones
|
|
// are one based.
|
|
assert(FS.getPositionalArgIndex() > 0U);
|
|
StandardFormatString.append(llvm::utostr(FS.getPositionalArgIndex() - 1));
|
|
}
|
|
|
|
// std::format format argument parts to potentially emit:
|
|
// [[fill]align][sign]["#"]["0"][width]["."precision][type]
|
|
std::string FormatSpec;
|
|
|
|
// printf doesn't support specifying the fill character - it's always a
|
|
// space, so we never need to generate one.
|
|
|
|
emitAlignment(FS, FormatSpec);
|
|
emitSign(FS, FormatSpec);
|
|
emitAlternativeForm(FS, FormatSpec);
|
|
|
|
if (FS.hasLeadingZeros())
|
|
FormatSpec.push_back('0');
|
|
|
|
emitFieldWidth(FS, FormatSpec);
|
|
emitPrecision(FS, FormatSpec);
|
|
maybeRotateArguments(FS);
|
|
|
|
if (!emitType(FS, Arg, FormatSpec))
|
|
return false;
|
|
|
|
if (!FormatSpec.empty()) {
|
|
StandardFormatString.push_back(':');
|
|
StandardFormatString.append(FormatSpec);
|
|
}
|
|
|
|
StandardFormatString.push_back('}');
|
|
return true;
|
|
}
|
|
|
|
/// Called for each format specifier by ParsePrintfString.
|
|
bool FormatStringConverter::HandlePrintfSpecifier(const PrintfSpecifier &FS,
|
|
const char *StartSpecifier,
|
|
unsigned SpecifierLen,
|
|
const TargetInfo &Target) {
|
|
|
|
const size_t StartSpecifierPos = StartSpecifier - PrintfFormatString.data();
|
|
assert(StartSpecifierPos + SpecifierLen <= PrintfFormatString.size());
|
|
|
|
// Everything before the specifier needs copying verbatim
|
|
assert(StartSpecifierPos >= PrintfFormatStringPos);
|
|
|
|
appendFormatText(StringRef(PrintfFormatString.begin() + PrintfFormatStringPos,
|
|
StartSpecifierPos - PrintfFormatStringPos));
|
|
|
|
const ConversionSpecifier::Kind ArgKind =
|
|
FS.getConversionSpecifier().getKind();
|
|
|
|
// Skip over specifier
|
|
PrintfFormatStringPos = StartSpecifierPos + SpecifierLen;
|
|
assert(PrintfFormatStringPos <= PrintfFormatString.size());
|
|
|
|
FormatStringNeededRewriting = true;
|
|
|
|
if (ArgKind == ConversionSpecifier::Kind::nArg) {
|
|
// std::print doesn't do the equivalent of %n
|
|
return conversionNotPossible("'%n' is not supported in format string");
|
|
}
|
|
|
|
if (ArgKind == ConversionSpecifier::Kind::PrintErrno) {
|
|
// std::print doesn't support %m. In theory we could insert a
|
|
// strerror(errno) parameter (assuming that libc has a thread-safe
|
|
// implementation, which glibc does), but that would require keeping track
|
|
// of the input and output parameter indices for position arguments too.
|
|
return conversionNotPossible("'%m' is not supported in format string");
|
|
}
|
|
|
|
if (ArgKind == ConversionSpecifier::PercentArg) {
|
|
StandardFormatString.push_back('%');
|
|
return true;
|
|
}
|
|
|
|
const unsigned ArgIndex = FS.getArgIndex() + ArgsOffset;
|
|
if (ArgIndex >= NumArgs) {
|
|
// Argument index out of range. Give up.
|
|
return conversionNotPossible(
|
|
(Twine("argument index ") + Twine(ArgIndex) + " is out of range")
|
|
.str());
|
|
}
|
|
|
|
return convertArgument(FS, Args[ArgIndex]->IgnoreImplicitAsWritten(),
|
|
StandardFormatString);
|
|
}
|
|
|
|
/// Called at the very end just before applying fixes to capture the last part
|
|
/// of the format string.
|
|
void FormatStringConverter::finalizeFormatText() {
|
|
appendFormatText(
|
|
StringRef(PrintfFormatString.begin() + PrintfFormatStringPos,
|
|
PrintfFormatString.size() - PrintfFormatStringPos));
|
|
PrintfFormatStringPos = PrintfFormatString.size();
|
|
|
|
// It's clearer to convert printf("Hello\r\n"); to std::print("Hello\r\n")
|
|
// than to std::println("Hello\r");
|
|
// Use StringRef until C++20 std::string::ends_with() is available.
|
|
const auto StandardFormatStringRef = StringRef(StandardFormatString);
|
|
if (Config.AllowTrailingNewlineRemoval &&
|
|
StandardFormatStringRef.ends_with("\\n") &&
|
|
!StandardFormatStringRef.ends_with("\\\\n") &&
|
|
!StandardFormatStringRef.ends_with("\\r\\n")) {
|
|
UsePrintNewlineFunction = true;
|
|
FormatStringNeededRewriting = true;
|
|
StandardFormatString.erase(StandardFormatString.end() - 2,
|
|
StandardFormatString.end());
|
|
}
|
|
|
|
StandardFormatString.push_back('\"');
|
|
}
|
|
|
|
/// Append literal parts of the format text, reinstating escapes as required.
|
|
void FormatStringConverter::appendFormatText(const StringRef Text) {
|
|
for (const char Ch : Text) {
|
|
if (Ch == '\a')
|
|
StandardFormatString += "\\a";
|
|
else if (Ch == '\b')
|
|
StandardFormatString += "\\b";
|
|
else if (Ch == '\f')
|
|
StandardFormatString += "\\f";
|
|
else if (Ch == '\n')
|
|
StandardFormatString += "\\n";
|
|
else if (Ch == '\r')
|
|
StandardFormatString += "\\r";
|
|
else if (Ch == '\t')
|
|
StandardFormatString += "\\t";
|
|
else if (Ch == '\v')
|
|
StandardFormatString += "\\v";
|
|
else if (Ch == '\"')
|
|
StandardFormatString += "\\\"";
|
|
else if (Ch == '\\')
|
|
StandardFormatString += "\\\\";
|
|
else if (Ch == '{') {
|
|
StandardFormatString += "{{";
|
|
FormatStringNeededRewriting = true;
|
|
} else if (Ch == '}') {
|
|
StandardFormatString += "}}";
|
|
FormatStringNeededRewriting = true;
|
|
} else if (Ch < 32) {
|
|
StandardFormatString += "\\x";
|
|
StandardFormatString += llvm::hexdigit(Ch >> 4, true);
|
|
StandardFormatString += llvm::hexdigit(Ch & 0xf, true);
|
|
} else
|
|
StandardFormatString += Ch;
|
|
}
|
|
}
|
|
|
|
static std::string withoutCStrReplacement(const BoundNodes &CStrRemovalMatch,
|
|
ASTContext &Context) {
|
|
const auto *Arg = CStrRemovalMatch.getNodeAs<Expr>("arg");
|
|
const auto *Member = CStrRemovalMatch.getNodeAs<MemberExpr>("member");
|
|
const bool Arrow = Member->isArrow();
|
|
return Arrow ? utils::fixit::formatDereference(*Arg, Context)
|
|
: tooling::fixit::getText(*Arg, Context).str();
|
|
}
|
|
|
|
/// Called by the check when it is ready to apply the fixes.
|
|
void FormatStringConverter::applyFixes(DiagnosticBuilder &Diag,
|
|
SourceManager &SM) {
|
|
if (FormatStringNeededRewriting) {
|
|
Diag << FixItHint::CreateReplacement(
|
|
CharSourceRange::getTokenRange(FormatExpr->getBeginLoc(),
|
|
FormatExpr->getEndLoc()),
|
|
StandardFormatString);
|
|
}
|
|
|
|
// ArgCount is one less than the number of arguments to be rotated.
|
|
for (auto [ValueArgIndex, ArgCount] : ArgRotates) {
|
|
assert(ValueArgIndex < NumArgs);
|
|
assert(ValueArgIndex > ArgCount);
|
|
|
|
// First move the value argument to the right place. But if there's a
|
|
// pending c_str() removal then we must do that at the same time.
|
|
if (const auto CStrRemovalMatch =
|
|
std::find_if(ArgCStrRemovals.cbegin(), ArgCStrRemovals.cend(),
|
|
[ArgStartPos = Args[ValueArgIndex]->getBeginLoc()](
|
|
const BoundNodes &Match) {
|
|
// This c_str() removal corresponds to the argument
|
|
// being moved if they start at the same location.
|
|
const Expr *CStrArg = Match.getNodeAs<Expr>("arg");
|
|
return ArgStartPos == CStrArg->getBeginLoc();
|
|
});
|
|
CStrRemovalMatch != ArgCStrRemovals.end()) {
|
|
const std::string ArgText =
|
|
withoutCStrReplacement(*CStrRemovalMatch, *Context);
|
|
assert(!ArgText.empty());
|
|
|
|
Diag << FixItHint::CreateReplacement(
|
|
Args[ValueArgIndex - ArgCount]->getSourceRange(), ArgText);
|
|
|
|
// That c_str() removal is now dealt with, so we don't need to do it again
|
|
ArgCStrRemovals.erase(CStrRemovalMatch);
|
|
} else
|
|
Diag << tooling::fixit::createReplacement(*Args[ValueArgIndex - ArgCount],
|
|
*Args[ValueArgIndex], *Context);
|
|
|
|
// Now shift down the field width and precision (if either are present) to
|
|
// accommodate it.
|
|
for (size_t Offset = 0; Offset < ArgCount; ++Offset)
|
|
Diag << tooling::fixit::createReplacement(
|
|
*Args[ValueArgIndex - Offset], *Args[ValueArgIndex - Offset - 1],
|
|
*Context);
|
|
|
|
// Now we need to modify the ArgFix index too so that we fix the right
|
|
// argument. We don't need to care about the width and precision indices
|
|
// since they never need fixing.
|
|
for (auto &ArgFix : ArgFixes) {
|
|
if (ArgFix.ArgIndex == ValueArgIndex)
|
|
ArgFix.ArgIndex = ValueArgIndex - ArgCount;
|
|
}
|
|
}
|
|
|
|
for (const auto &[ArgIndex, Replacement] : ArgFixes) {
|
|
SourceLocation AfterOtherSide =
|
|
Lexer::findNextToken(Args[ArgIndex]->getEndLoc(), SM, LangOpts)
|
|
->getLocation();
|
|
|
|
Diag << FixItHint::CreateInsertion(Args[ArgIndex]->getBeginLoc(),
|
|
Replacement, true)
|
|
<< FixItHint::CreateInsertion(AfterOtherSide, ")", true);
|
|
}
|
|
|
|
for (const auto &Match : ArgCStrRemovals) {
|
|
const auto *Call = Match.getNodeAs<CallExpr>("call");
|
|
const std::string ArgText = withoutCStrReplacement(Match, *Context);
|
|
if (!ArgText.empty())
|
|
Diag << FixItHint::CreateReplacement(Call->getSourceRange(), ArgText);
|
|
}
|
|
}
|
|
} // namespace clang::tidy::utils
|