office-gobmx/compilerplugins/clang/typeidcomparison.cxx
Stephan Bergmann d2ba5d98ec Fix some tautological std::type_info (in-)equality comparisons
...which could never succeed.

I became aware of this when Clang 15 trunk -std=c++2b, implementing
<http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2021/p1328r1.html> "Making
std::type_info::operator== constexpr", pointed at two of the broken comparisons
with

> sw/source/uibase/app/docsh2.cxx:478:25: error: code will never be executed [-Werror,-Wunreachable-code]
>                         pTmpFrame->GetFrame().Appear();
>                         ^~~~~~~~~
> sw/source/uibase/app/docsh2.cxx:475:33: error: code will never be executed [-Werror,-Wunreachable-code]
>                         bOnly = false;
>                                 ^~~~~

(It didn't emit warnings pointing at any of the other broken comparisons,
though.)

All of these broken comparisons were regressions introduced with
89d39bc100 "tdf#94559: 4th step to remove
rtti.hxx", replacing uses of the IS_TYPE macro (from include/tools/rtti.hxx,
meanwhile removed with d64e535fe9 "Remove unused
rtti.hxx").

I now added loplugin:typeidcomparison to also find the other broken comparisons
introduced by that commit.  (The remaining cases where that commit replaced uses
of TYPE_INFO with typeid comparisons were correct and/or have meanwhile been
replaced with code not using typeid, see
553ee72041 "simplify some use of typeid" and
d656da9bc4 "~SwModify: do not silently tolerate
clients registered past death".)

The original IS_TYPE macro made sure not to dereference null pointers,

> #define IS_TYPE(T,pObj) \
>         ( pObj && (pObj)->Type() == TYPE(T) )

I don't know if any of the pointers now dereferenced in those typeid expressions
can legitimately be null.  But to be on the safe side, I replicated that check
in the newly introduced isType (sw/inc/istype.hxx).

(It is interesting to note that none of the static analysis that we routinely
employ seems to have noticed these broken comparisons.)

Change-Id: I65baffdd27bac1abf744283ff98c2dc864fa63b4
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/129865
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
2022-02-12 19:37:43 +01:00

90 lines
2.7 KiB
C++

/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; fill-column: 100 -*- */
/*
* 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/.
*/
// Find (in-)equality comparisons between typeid expressions that can never succeed. For now, just
// detects cases where the two involved types are structurally different, one a pointer type and the
// other a non-pointer type.
#ifndef LO_CLANG_SHARED_PLUGINS
#include "plugin.hxx"
namespace
{
class TypeidComparison final : public loplugin::FilteringPlugin<TypeidComparison>
{
public:
explicit TypeidComparison(loplugin::InstantiationData const& data)
: FilteringPlugin(data)
{
}
// For CXXRewrittenBinaryOperator `typeid(...) != typeid(...)`:
bool shouldVisitImplicitCode() const { return true; }
bool preRun() override { return compiler.getLangOpts().CPlusPlus; }
void run() override
{
if (preRun())
{
TraverseDecl(compiler.getASTContext().getTranslationUnitDecl());
}
}
bool VisitCXXOperatorCallExpr(CXXOperatorCallExpr const* expr)
{
if (ignoreLocation(expr))
{
return true;
}
auto const op = expr->getOperator();
if (op != OO_EqualEqual && op != OO_ExclaimEqual)
{
return true;
}
assert(expr->getNumArgs() == 2);
auto const e1 = dyn_cast<CXXTypeidExpr>(expr->getArg(0)->IgnoreParenImpCasts());
if (e1 == nullptr)
{
return true;
}
auto const e2 = dyn_cast<CXXTypeidExpr>(expr->getArg(1)->IgnoreParenImpCasts());
if (e2 == nullptr)
{
return true;
}
auto const t1 = getOperandType(e1);
auto const t2 = getOperandType(e2);
if (t1->isPointerType() == t2->isPointerType())
{
return true;
}
report(DiagnosticsEngine::Warning,
"comparison of type info of mixed pointer and non-pointer types %0 and %1 can never "
"succeed",
expr->getExprLoc())
<< t1 << t2 << expr->getSourceRange();
return true;
}
private:
QualType getOperandType(CXXTypeidExpr const* expr)
{
return expr->isTypeOperand() ? expr->getTypeOperand(compiler.getASTContext())
: expr->getExprOperand()->getType();
}
};
static loplugin::Plugin::Registration<TypeidComparison> typeidcomparison("typeidcomparison");
}
#endif
/* vim:set shiftwidth=4 softtabstop=4 expandtab cinoptions=b1,g0,N-s cinkeys+=0=break: */