libreoffice/compilerplugins/clang/unnecessaryparen.cxx

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

669 lines
24 KiB
C++
Raw Normal View History

/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
/*
* This file is part of the LibreOffice project.
*
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*/
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
#ifndef LO_CLANG_SHARED_PLUGINS
#include <cassert>
#include <string>
#include <iostream>
#include <fstream>
#include <set>
#include <unordered_set>
#include <clang/AST/CXXInheritance.h>
#include "config_clang.h"
#include "compat.hxx"
#include "plugin.hxx"
/**
look for unnecessary parentheses
*/
namespace {
// Like clang::Stmt::IgnoreImplicit (lib/AST/Stmt.cpp), but also ignoring CXXConstructExpr and
// looking through implicit UserDefinedConversion's member function call:
Expr const * ignoreAllImplicit(Expr const * expr) {
while (true)
{
auto oldExpr = expr;
if (auto const e = dyn_cast<ExprWithCleanups>(expr)) {
expr = e->getSubExpr();
}
else if (auto const e = dyn_cast<CXXConstructExpr>(expr)) {
if (e->getNumArgs() == 1) {
expr = e->getArg(0);
}
}
else if (auto const e = dyn_cast<MaterializeTemporaryExpr>(expr)) {
expr = compat::getSubExpr(e);
}
else if (auto const e = dyn_cast<CXXBindTemporaryExpr>(expr)) {
expr = e->getSubExpr();
}
else if (auto const e = dyn_cast<ImplicitCastExpr>(expr)) {
expr = e->getSubExpr();
if (e->getCastKind() == CK_UserDefinedConversion) {
auto const ce = cast<CXXMemberCallExpr>(expr);
assert(ce->getNumArgs() == 0);
expr = ce->getImplicitObjectArgument();
}
}
#if CLANG_VERSION >= 80000
else if (auto const e = dyn_cast<ConstantExpr>(expr)) {
expr = e->getSubExpr();
}
#endif
if (expr == oldExpr)
return expr;
}
return expr;
}
class UnnecessaryParen:
public loplugin::FilteringRewritePlugin<UnnecessaryParen>
{
public:
explicit UnnecessaryParen(loplugin::InstantiationData const & data):
FilteringRewritePlugin(data) {}
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
virtual bool preRun() override
{
StringRef fn(handler.getMainFileName());
// fixing this, makes the source in the .y files look horrible
if (loplugin::isSamePathname(fn, WORKDIR "/YaccTarget/unoidl/source/sourceprovider-parser.cxx"))
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
return false;
if (loplugin::isSamePathname(fn, WORKDIR "/YaccTarget/idlc/source/parser.cxx"))
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
return false;
return true;
}
virtual void run() override
{
if( preRun())
TraverseDecl(compiler.getASTContext().getTranslationUnitDecl());
}
bool VisitParenExpr(const ParenExpr *);
bool VisitIfStmt(const IfStmt *);
bool VisitDoStmt(const DoStmt *);
bool VisitWhileStmt(const WhileStmt *);
bool VisitForStmt(ForStmt const * stmt);
bool VisitSwitchStmt(const SwitchStmt *);
bool VisitCaseStmt(const CaseStmt *);
bool VisitReturnStmt(const ReturnStmt* );
bool VisitCallExpr(const CallExpr *);
bool VisitVarDecl(const VarDecl *);
bool VisitCXXOperatorCallExpr(const CXXOperatorCallExpr *);
bool VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr const *);
bool VisitConditionalOperator(ConditionalOperator const * expr);
bool VisitBinaryConditionalOperator(BinaryConditionalOperator const * expr);
bool VisitMemberExpr(const MemberExpr *f);
bool VisitCXXDeleteExpr(const CXXDeleteExpr *);
bool VisitImplicitCastExpr(ImplicitCastExpr const * expr) {
if (ignoreLocation(expr)) {
return true;
}
if (expr->getCastKind() != CK_UserDefinedConversion) {
return true;
}
// Filter out a MemberExpr (resp. a ParenExpr sub-expr, if any, as would be found by
// VisitMemberExpr) that is part of a CXXMemberCallExpr which in turn is part of an
// ImplicitCastExpr, so that VisitMemberExpr doesn't erroneously pick it up (and note that
// CXXMemberCallExpr's getImplicitObjectArgument() skips past the underlying MemberExpr):
if (auto const e1 = dyn_cast<CXXMemberCallExpr>(expr->getSubExpr())) {
if (auto const e2 = dyn_cast<ParenExpr>(
e1->getImplicitObjectArgument()->IgnoreImpCasts()))
{
handled_.insert(e2);
}
}
return true;
}
private:
void VisitSomeStmt(Stmt const * stmt, const Expr* cond, StringRef stmtName);
void handleUnreachableCodeConditionParens(Expr const * expr);
// Hack for libxml2's BAD_CAST object-like macro (expanding to "(xmlChar *)"), which is
// typically used as if it were a function-like macro, e.g., as "BAD_CAST(pName)" in
// SwNode::dumpAsXml (sw/source/core/docnode/node.cxx):
bool isPrecededBy_BAD_CAST(Expr const * expr);
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
bool badCombination(SourceLocation loc, int prevOffset, int nextOffset);
bool removeParens(ParenExpr const * expr);
std::unordered_set<ParenExpr const *> handled_;
};
bool UnnecessaryParen::VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr const * expr)
{
if (expr->getKind() == UETT_SizeOf && !expr->isArgumentType()) {
if (auto const e = dyn_cast<ParenExpr>(ignoreAllImplicit(expr->getArgumentExpr()))) {
handled_.insert(e);
}
}
return true;
}
bool UnnecessaryParen::VisitConditionalOperator(ConditionalOperator const * expr) {
handleUnreachableCodeConditionParens(expr->getCond());
return true;
}
bool UnnecessaryParen::VisitBinaryConditionalOperator(BinaryConditionalOperator const * expr) {
handleUnreachableCodeConditionParens(expr->getCond());
return true;
}
bool UnnecessaryParen::VisitParenExpr(const ParenExpr* parenExpr)
{
if (ignoreLocation(parenExpr))
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
if (handled_.find(parenExpr) != handled_.end())
return true;
auto subExpr = ignoreAllImplicit(parenExpr->getSubExpr());
if (auto subParenExpr = dyn_cast<ParenExpr>(subExpr))
{
if (compat::getBeginLoc(subParenExpr).isMacroID())
return true;
report(
DiagnosticsEngine::Warning, "parentheses around parentheses",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(subParenExpr);
}
// Somewhat redundantly add parenExpr to handled_, so that issues within InitListExpr don't get
// reported twice (without having to change TraverseInitListExpr to only either traverse the
// syntactic or semantic form, as other plugins do):
if (isa<DeclRefExpr>(subExpr)) {
if (!isPrecededBy_BAD_CAST(parenExpr)) {
report(
DiagnosticsEngine::Warning, "unnecessary parentheses around identifier",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
} else if (isa<IntegerLiteral>(subExpr) || isa<CharacterLiteral>(subExpr)
|| isa<FloatingLiteral>(subExpr) || isa<ImaginaryLiteral>(subExpr)
|| isa<CXXBoolLiteralExpr>(subExpr) || isa<CXXNullPtrLiteralExpr>(subExpr)
|| isa<ObjCBoolLiteralExpr>(subExpr))
{
auto const loc = compat::getBeginLoc(subExpr);
if (loc.isMacroID() && compiler.getSourceManager().isAtStartOfImmediateMacroExpansion(loc))
{
// just in case the macro could also expand to something that /would/ require
// parentheses here
return true;
}
report(
DiagnosticsEngine::Warning, "unnecessary parentheses around literal",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
} else if (auto const e = dyn_cast<clang::StringLiteral>(subExpr)) {
if (e->getNumConcatenated() == 1 && !isPrecededBy_BAD_CAST(parenExpr)) {
report(
DiagnosticsEngine::Warning,
"unnecessary parentheses around single-token string literal",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
} else if (auto const e = dyn_cast<UnaryOperator>(subExpr)) {
auto const op = e->getOpcode();
if (op == UO_Plus || op == UO_Minus) {
auto const e2 = e->getSubExpr();
if (isa<IntegerLiteral>(e2) || isa<FloatingLiteral>(e2) || isa<ImaginaryLiteral>(e2)) {
report(
DiagnosticsEngine::Warning,
"unnecessary parentheses around signed numeric literal",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
}
} else if (isa<CXXNamedCastExpr>(subExpr)) {
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
if (!removeParens(parenExpr)) {
report(
DiagnosticsEngine::Warning, "unnecessary parentheses around cast",
compat::getBeginLoc(parenExpr))
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
<< parenExpr->getSourceRange();
}
handled_.insert(parenExpr);
} else if (auto memberExpr = dyn_cast<MemberExpr>(subExpr)) {
if (isa<CXXThisExpr>(ignoreAllImplicit(memberExpr->getBase()))) {
report(
DiagnosticsEngine::Warning, "unnecessary parentheses around member expr",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
}
return true;
}
bool UnnecessaryParen::VisitIfStmt(const IfStmt* ifStmt)
{
handleUnreachableCodeConditionParens(ifStmt->getCond());
VisitSomeStmt(ifStmt, ifStmt->getCond(), "if");
return true;
}
bool UnnecessaryParen::VisitDoStmt(const DoStmt* doStmt)
{
VisitSomeStmt(doStmt, doStmt->getCond(), "do");
return true;
}
bool UnnecessaryParen::VisitWhileStmt(const WhileStmt* whileStmt)
{
handleUnreachableCodeConditionParens(whileStmt->getCond());
VisitSomeStmt(whileStmt, whileStmt->getCond(), "while");
return true;
}
bool UnnecessaryParen::VisitForStmt(ForStmt const * stmt) {
if (auto const cond = stmt->getCond()) {
handleUnreachableCodeConditionParens(cond);
}
return true;
}
bool UnnecessaryParen::VisitSwitchStmt(const SwitchStmt* switchStmt)
{
VisitSomeStmt(switchStmt, switchStmt->getCond(), "switch");
return true;
}
bool UnnecessaryParen::VisitCaseStmt(const CaseStmt* caseStmt)
{
VisitSomeStmt(caseStmt, caseStmt->getLHS(), "case");
return true;
}
bool UnnecessaryParen::VisitReturnStmt(const ReturnStmt* returnStmt)
{
if (ignoreLocation(returnStmt))
return true;
if (!returnStmt->getRetValue())
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(returnStmt->getRetValue()));
if (!parenExpr)
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
// assignments need extra parentheses or they generate a compiler warning
auto binaryOp = dyn_cast<BinaryOperator>(parenExpr->getSubExpr());
if (binaryOp && binaryOp->getOpcode() == BO_Assign)
return true;
// only non-operator-calls for now
auto subExpr = ignoreAllImplicit(parenExpr->getSubExpr());
if (isa<CallExpr>(subExpr) && !isa<CXXOperatorCallExpr>(subExpr))
{
report(
DiagnosticsEngine::Warning, "parentheses immediately inside return statement",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
return true;
}
void UnnecessaryParen::VisitSomeStmt(const Stmt * stmt, const Expr* cond, StringRef stmtName)
{
if (ignoreLocation(stmt))
return;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(cond));
if (parenExpr) {
if (handled_.find(parenExpr) != handled_.end()) {
Enable -Wunreachable-code ...motivated by <https://gerrit.libreoffice.org/#/c/41565/2> adding dead code at the end of a switch statement, after the last case's "break". -Wunreachable-code appears to work well on Clang, while it appears to have no effect on GCC. Most of the affected places are apparently temporary/TODO/FIXME cases of disabling code via "if (false)", which can be written with an extra set of parentheses as "if ((false))" to silence -Wunreachable-code on Clang (which thus needed loplugin:unnecessaryparen to be adapted accordingly). In some cases, the controlling expression was more complex than just "false" and needed to be rewritten by taking it out of the if statement to silence Clang. One noteworthy case where the nature of the disabled code wasn't immediately apparent: Sep 12 16:59:58 <sberg> quikee, is that "if (false)" in ScExponentialSmoothingDialog::ApplyOutput (sc/source/ui/StatisticsDialogs/ExponentialSmoothingDialog.cxx) some work-in- progress or dead code? Sep 12 17:02:03 <quikee> sberg: WIP, but you can remove it Sep 12 17:04:47 <sberg> quikee, I'll wrap the false in an extra set of parentheses for now, to silence -Wunreachable-code (I wouldn't want to remove it, as I have no idea whether I should then also remove the "Initial value" comment preceding it) Sep 12 17:07:29 <quikee> sberg: both are different ways to calculate the "intital value"... so no Another case where the nature of the dead code, following while (true) loops without breaks, is unclear is sd/source/ui/remotecontrol/BluetoothServer.cxx, where I added TODO markers to the workarounds that silence the warnings for now. basic/source/sbx/sbxvalue.cxx had a variable of type double, of automatic storage duration, and without an initalizer at the top of a switch statement. Clang warning about it is arguably a false positive. Apart from that, this didn't find any cases of genuinely dead code in the existing code base. Change-Id: Ib00b822c8efec94278c048783d5997b8ba86a94c Reviewed-on: https://gerrit.libreoffice.org/42217 Tested-by: Stephan Bergmann <sbergman@redhat.com> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2017-09-12 18:24:46 +02:00
return;
}
if (compat::getBeginLoc(parenExpr).isMacroID())
return;
// assignments need extra parentheses or they generate a compiler warning
auto binaryOp = dyn_cast<BinaryOperator>(parenExpr->getSubExpr());
if (binaryOp && binaryOp->getOpcode() == BO_Assign)
return;
if (auto const opCall = dyn_cast<CXXOperatorCallExpr>(parenExpr->getSubExpr())) {
if (opCall->getOperator() == OO_Equal) {
return;
}
}
report(
DiagnosticsEngine::Warning, "parentheses immediately inside %0 statement",
compat::getBeginLoc(parenExpr))
<< stmtName
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
}
}
bool UnnecessaryParen::VisitCallExpr(const CallExpr* callExpr)
{
if (ignoreLocation(callExpr))
return true;
if (callExpr->getNumArgs() != 1 || isa<CXXOperatorCallExpr>(callExpr))
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(callExpr->getArg(0)));
if (!parenExpr)
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
// assignments need extra parentheses or they generate a compiler warning
auto binaryOp = dyn_cast<BinaryOperator>(parenExpr->getSubExpr());
if (binaryOp && binaryOp->getOpcode() == BO_Assign)
return true;
report(
DiagnosticsEngine::Warning, "parentheses immediately inside single-arg call",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
return true;
}
bool UnnecessaryParen::VisitCXXDeleteExpr(const CXXDeleteExpr* deleteExpr)
{
if (ignoreLocation(deleteExpr))
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(deleteExpr->getArgument()));
if (!parenExpr)
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
// assignments need extra parentheses or they generate a compiler warning
auto binaryOp = dyn_cast<BinaryOperator>(parenExpr->getSubExpr());
if (binaryOp && binaryOp->getOpcode() == BO_Assign)
return true;
report(
DiagnosticsEngine::Warning, "parentheses immediately inside delete expr",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
return true;
}
bool UnnecessaryParen::VisitCXXOperatorCallExpr(const CXXOperatorCallExpr* callExpr)
{
if (ignoreLocation(callExpr))
return true;
if (callExpr->getNumArgs() != 2)
return true;
// Same logic as CXXOperatorCallExpr::isAssignmentOp(), which our supported clang
// doesn't have yet.
auto Opc = callExpr->getOperator();
if (Opc != OO_Equal && Opc != OO_StarEqual &&
Opc != OO_SlashEqual && Opc != OO_PercentEqual &&
Opc != OO_PlusEqual && Opc != OO_MinusEqual &&
Opc != OO_LessLessEqual && Opc != OO_GreaterGreaterEqual &&
Opc != OO_AmpEqual && Opc != OO_CaretEqual &&
Opc != OO_PipeEqual)
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(callExpr->getArg(1)));
if (!parenExpr)
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
// Sometimes parentheses make the RHS of an assignment easier to read by
// visually disambiguating the = from a call to ==
auto sub = parenExpr->getSubExpr();
if (auto subBinOp = dyn_cast<BinaryOperator>(sub))
{
if (!(subBinOp->isMultiplicativeOp() || subBinOp->isAdditiveOp() || subBinOp->isPtrMemOp()))
return true;
}
if (auto subOperatorCall = dyn_cast<CXXOperatorCallExpr>(sub))
{
auto op = subOperatorCall->getOperator();
if (!((op >= OO_Plus && op <= OO_Exclaim) || (op >= OO_ArrowStar && op <= OO_Subscript)))
return true;
}
if (isa<ConditionalOperator>(sub))
return true;
report(
DiagnosticsEngine::Warning, "parentheses immediately inside assignment",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
return true;
}
bool UnnecessaryParen::VisitVarDecl(const VarDecl* varDecl)
{
if (ignoreLocation(varDecl))
return true;
if (!varDecl->getInit())
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(varDecl->getInit()));
if (!parenExpr)
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
// Sometimes parentheses make the RHS of an assignment easier to read by
// visually disambiguating the = from a call to ==
auto sub = parenExpr->getSubExpr();
#if CLANG_VERSION >= 100000
if (auto const e = dyn_cast<CXXRewrittenBinaryOperator>(sub)) {
sub = e->getDecomposedForm().InnerBinOp;
}
#endif
if (auto subBinOp = dyn_cast<BinaryOperator>(sub))
{
if (!(subBinOp->isMultiplicativeOp() || subBinOp->isAdditiveOp() || subBinOp->isPtrMemOp()))
return true;
}
if (auto subOperatorCall = dyn_cast<CXXOperatorCallExpr>(sub))
{
auto op = subOperatorCall->getOperator();
if (!((op >= OO_Plus && op <= OO_Exclaim) || (op >= OO_ArrowStar && op <= OO_Subscript)))
return true;
}
if (isa<ConditionalOperator>(sub))
return true;
// these two are for "parentheses were disambiguated as a function declaration [-Werror,-Wvexing-parse]"
auto const sub2 = sub->IgnoreImplicit();
if (isa<CXXTemporaryObjectExpr>(sub2)
|| isa<CXXFunctionalCastExpr>(sub2))
return true;
report(
DiagnosticsEngine::Warning, "parentheses immediately inside vardecl statement",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
return true;
}
bool UnnecessaryParen::VisitMemberExpr(const MemberExpr* memberExpr)
{
if (ignoreLocation(memberExpr))
return true;
auto parenExpr = dyn_cast<ParenExpr>(ignoreAllImplicit(memberExpr->getBase()));
if (!parenExpr)
return true;
if (handled_.find(parenExpr) != handled_.end())
return true;
if (compat::getBeginLoc(parenExpr).isMacroID())
return true;
auto sub = parenExpr->getSubExpr();
if (isa<CallExpr>(sub)) {
if (isa<CXXOperatorCallExpr>(sub))
return true;
} else if (isa<CXXConstructExpr>(sub)) {
// warn
} else if (isa<MemberExpr>(sub)) {
// warn
} else if (isa<DeclRefExpr>(sub)) {
// warn
} else
return true;
report(
DiagnosticsEngine::Warning, "unnecessary parentheses around member expr",
compat::getBeginLoc(parenExpr))
<< parenExpr->getSourceRange();
handled_.insert(parenExpr);
return true;
}
// Conservatively assume any parenthesised integer or Boolean (incl. Objective-C ones) literal in
// certain condition expressions (i.e., those for which handleUnreachableCodeConditionParens is
// called) to be parenthesised to silence Clang -Wunreachable-code, if that is either the whole
// condition expression or appears as a certain sub-expression (looking at what isConfigurationValue
// in Clang's lib/Analysis/ReachableCode.cpp looks for, descending into certain unary and binary
// operators):
void UnnecessaryParen::handleUnreachableCodeConditionParens(Expr const * expr) {
// Cf. :
auto const e = ignoreAllImplicit(expr);
if (auto const e1 = dyn_cast<ParenExpr>(e)) {
auto const sub = e1->getSubExpr();
if (isa<IntegerLiteral>(sub) || isa<CXXBoolLiteralExpr>(sub)
|| isa<ObjCBoolLiteralExpr>(sub))
{
handled_.insert(e1);
}
} else if (auto const e1 = dyn_cast<UnaryOperator>(e)) {
if (e1->getOpcode() == UO_LNot) {
handleUnreachableCodeConditionParens(e1->getSubExpr());
}
} else if (auto const e1 = dyn_cast<BinaryOperator>(e)) {
if (e1->isLogicalOp() || e1->isComparisonOp()) {
handleUnreachableCodeConditionParens(e1->getLHS());
handleUnreachableCodeConditionParens(e1->getRHS());
}
}
}
bool UnnecessaryParen::isPrecededBy_BAD_CAST(Expr const * expr) {
if (compat::getBeginLoc(expr).isMacroID()) {
return false;
}
SourceManager& SM = compiler.getSourceManager();
const char *p1 = SM.getCharacterData( compat::getBeginLoc(expr).getLocWithOffset(-10) );
const char *p2 = SM.getCharacterData( compat::getBeginLoc(expr) );
return std::string(p1, p2 - p1).find("BAD_CAST") != std::string::npos;
}
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
namespace {
bool badCombinationChar(char c) {
return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') || c == '_'
|| c == '+' || c == '-' || c == '\'' || c == '"';
}
}
bool UnnecessaryParen::badCombination(SourceLocation loc, int prevOffset, int nextOffset) {
//TODO: check for start/end of file; take backslash-newline line concatenation into account
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
auto const c1
= compiler.getSourceManager().getCharacterData(loc.getLocWithOffset(prevOffset))[0];
auto const c2
= compiler.getSourceManager().getCharacterData(loc.getLocWithOffset(nextOffset))[0];
// An approximation of avoiding whatever combinations that would cause two adjacent tokens to be
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
// lexed differently, using, for now, letters (TODO: non-ASCII ones) and digits and '_'; '+' and
// '-' (to avoid ++, etc.); '\'' and '"' (to avoid u'x' or "foo"bar, etc.):
return badCombinationChar(c1) && badCombinationChar(c2);
}
bool UnnecessaryParen::removeParens(ParenExpr const * expr) {
if (rewriter == nullptr) {
return false;
}
auto const firstBegin = compat::getBeginLoc(expr);
auto secondBegin = compat::getEndLoc(expr);
Enable loplugin:cstylecast for some more cases ...mostly of C-style casts among arithmetic types, and automatically rewrite those into either static_cast or a functional cast (which should have identical semantics, but where the latter probably looks better for simple cases like casting a literal to a specific type, as in "sal_Int32(0)" vs. "static_cast<sal_Int32>(0)"). The main benefit of reducing the amount of C-style casts across the code base further is so that other plugins (that have not been taught about the complex semantics of C-style cast) can pick those up (cf. the various recent "loplugin:redundantcast" commits, which address those findings after this improved loplugin:cstylecast has been run). Also, I found some places where a C-style cast has probably been applied only to the first part of a larger expression in error (because it's easy to forget parentheses in cases like "(sal_uInt16)VOPT_CLIPMARKS+1"); I'll follow up on those individually. The improved loplugin:cstylecast is careful to output either "(performs: static_cast)" or "(performs: functional cast)", so that compilerplugins/clang/test/cstylecast.cxx can check that the plugin would automatically rewrite to one or the other form. To allow fully-automatic rewriting, this also required loplugin:unnecessaryparen to become a rewriting plugin, at least for the parens-around-cast case (where "((foo)bar)" first gets rewritten to "(static_cast<foo>(bar))", then to "static_cast<foo>(bar)". Rewriting could probably be added to other cases of loplugin:unnecessaryparen in the future, too. (The final version of this patch would even have been able to cope with 361dd2576a09fbda83f3ce9a26ecb590c38f74e3 "Replace some C-style casts in ugly macros with static_cast", so that manual change would not have been necessary after all.) Change-Id: Icd7e319cc38eb58262fcbf7643d177ac9ea0220a Reviewed-on: https://gerrit.libreoffice.org/47798 Tested-by: Jenkins <ci@libreoffice.org> Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2018-01-10 15:55:22 +01:00
if (firstBegin.isMacroID() || secondBegin.isMacroID()) {
return false;
}
unsigned firstLen = Lexer::MeasureTokenLength(
firstBegin, compiler.getSourceManager(), compiler.getLangOpts());
for (auto l = firstBegin.getLocWithOffset(std::max<unsigned>(firstLen, 1));;
l = l.getLocWithOffset(1))
{
unsigned n = Lexer::MeasureTokenLength(
l, compiler.getSourceManager(), compiler.getLangOpts());
if (n != 0) {
break;
}
++firstLen;
}
unsigned secondLen = Lexer::MeasureTokenLength(
secondBegin, compiler.getSourceManager(), compiler.getLangOpts());
for (;;) {
auto l = secondBegin.getLocWithOffset(-1);
auto const c = compiler.getSourceManager().getCharacterData(l)[0];
if (c == '\n') {
if (compiler.getSourceManager().getCharacterData(l.getLocWithOffset(-1))[0] == '\\') {
break;
}
} else if (!(c == ' ' || c == '\t' || c == '\v' || c == '\f')) {
break;
}
secondBegin = l;
++secondLen;
}
if (!replaceText(firstBegin, firstLen, badCombination(firstBegin, -1, firstLen) ? " " : "")) {
if (isDebugMode()) {
report(
DiagnosticsEngine::Fatal,
"TODO: cannot rewrite opening parenthesis, needs investigation",
firstBegin);
report(
DiagnosticsEngine::Note, "when removing these parentheses", expr->getExprLoc())
<< expr->getSourceRange();
}
return false;
}
if (!replaceText(secondBegin, secondLen, badCombination(secondBegin, -1, secondLen) ? " " : ""))
{
//TODO: roll back first change
if (isDebugMode()) {
report(
DiagnosticsEngine::Fatal,
"TODO: cannot rewrite closing parenthesis, needs investigation",
secondBegin);
report(
DiagnosticsEngine::Note, "when removing these parentheses", expr->getExprLoc())
<< expr->getSourceRange();
}
return false;
}
return true;
}
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
loplugin::Plugin::Registration< UnnecessaryParen > unnecessaryparen("unnecessaryparen", true);
}
make (some) clang plugins share the same RecursiveASTVisitor Each plugin currently uses its own recursive AST run, which adds up. This patch adds another shared plugin which internally contains all (suitable) plugins and dispatches to them from the same one recursive run. This patch converts ~25 plugins and for starmath's accessibility.cxx reduces clang build time from 5.43s to 5.14s (and it's 4.39s without any plugins). As there are almost 50 more plugins to go, this can theoretically result in 4.56s final time, although probably not all plugins can be that easily converted, if at all. This mostly requires very little change in many plugins (see e.g. BadStatics), some even work without any functionality change (e.g. CharRightShift). Traverse* calls require some changes but are often not that difficult. WalkUp* probably can't be supported, although some plugins can(?) possibly be adjusted to not rely on them. And of course some plugins can be left as they are, using their own recursive run. See description at the top of generator.cxx for description of how to convert a plugin. The sharedvisitor.cxx source is generated based on scanning relevant plugin sources using a clang-based scanner/generator. The generated source is intentionally included instead of getting always generated, as the generating currently takes some time, so it should get updated in git whenever a change in a plugin triggers a source change in it. Change-Id: Ia0d2e3a5a464659503dbb4ed6c20b6cc89b4de01 Reviewed-on: https://gerrit.libreoffice.org/68026 Tested-by: Jenkins Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk> Reviewed-by: Luboš Luňák <l.lunak@collabora.com>
2019-03-07 14:31:17 +01:00
#endif // LO_CLANG_SHARED_PLUGINS
/* vim:set shiftwidth=4 softtabstop=4 expandtab: */