aboutsummaryrefslogtreecommitdiffstats
path: root/src
Commit message (Collapse)AuthorAgeFilesLines
* Workaround crash with LLVM 7upstream/1.3Sergio Martins2018-09-221-0/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | If clazy is compiled with clang instead of gcc it might crash with: ==10637== Process terminating with default action of signal 11 (SIGSEGV): dumping core ==10637== Access not within mapped region at address 0x8 ==10637== at 0x19CDD8C: clang::ast_matchers::MatchFinder::MatchFinder(clang::ast_matchers::MatchFinder::MatchFinderOptions) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0x9D75670: ClazyASTConsumer (Clazy.cpp:62) ==10637== by 0x9D75670: ClazyASTAction::CreateASTConsumer(clang::CompilerInstance&, llvm::StringRef) (Clazy.cpp:183) ==10637== by 0x9E29ED: clang::FrontendAction::CreateWrappedASTConsumer(clang::CompilerInstance&, llvm::StringRef) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0x9E8FCA: clang::FrontendAction::BeginSourceFile(clang::CompilerInstance&, clang::FrontendInputFile const&) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0x9AE3D5: clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0xA8C9FA: clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0x5822C7: cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (in /usr/lib/llvm-7/bin/clang) ==10637== by 0x571ACC: main (in /usr/lib/llvm-7/bin/clang) After debugging clazy and clang's code I couldn't find anything wrong with it. Valgrind's output doesn't make much sense, and simply compiling the Clazy.cpp translation unit with gcc instead of clang makes the crash go away and valgrind's output is clean. I'm assuming debian's LLVM was built with gcc and building clazy with clang will have some sort of incompatibility, or maybe it's simply a clang bug. The downside of this workaround is that qcolor-literal check will be disabled. Next step will be producing a minimal test case and reporting to LLVM. BUG: 392223 CCMAIL: Woebbeking@kde.org
* prospective fix for crash with clang-7.0Sergio Martins2018-09-192-3/+6
| | | | | | | | | | This shuffles the code around in hope of avoiding what seems to be a clang bug. The backtrace and valgrind report don't make sense to me, and the crash goes away if building clazy with gcc. CCBUG: 392223
* Fix build with clang 7.0Sergio Martins2018-09-191-4/+14
| | | | (cherry-picked from 06edd15fdda9155b54366d3887a4fbf25cf1d911)
* Don't recommend QStringLiteral in Q_GLOBAL_STATIC_WITH_ARGSSergio Martins2018-03-131-0/+7
| | | | | | Doesn't compile since it got noexcept BUG: 391807
* inefficient-qlist: Fix crash on incomplete typesWilliam Bader2018-01-241-1/+1
| | | | BUG: 389360
* detaching-temporary: Fix failing unit-testSergio Martins2018-01-092-3/+2
| | | | The operator path should only be executed on first iteration
* writing-to-temporary: Fix false-positiveSergio Martins2018-01-092-1/+6
| | | | | This was matching: (should() ? container1 : container2).append("")
* rule-of-three: Improve last commitSergio Martins2018-01-081-6/+11
| | | | | | | | | | Now it also fixes the second bug Our previous fix wasn't being run because it was in the if (numImplemeted == 1), so failed for the case where we have a user-dtor, user-copy-ctor and deleted-copy-assign. Moving the check out of the if fixes it. BUG: 388682
* rule-of-three: Don't warn if either copy-assign or copy-ctor is deletedSergio Martins2018-01-081-1/+1
| | | | | | | | It's safe, as it won't happen that one of the user-methods will do something different from the compiler-generated-method, as one of them can't be called. BUG: 388677
* qdeleteall: Reword the warning message a bitSergio Martins2017-12-191-9/+10
| | | | | | | | | | When you're using qDeleteAll(c.keys()) you should use qDeleteAll(c.keyBegin(), c.keyEnd()) The new messages are: warning: qDeleteAll() is being used on an unnecessary temporary container created by QMap::values(), use qDeleteAll(mycontainer) instead [-Wclazy-qdeleteall] warning: qDeleteAll() is being used on an unnecessary temporary container created by QMap::keys(), use qDeleteAll(mycontainer.keyBegin(), mycontainer.keyEnd()) instead [-Wclazy-qdeleteall] CCMAIL: aacid@kde.org
* Fix Utils::containsNonConstMemberCall()Sergio Martins2017-12-101-21/+41
| | | | | The parent() calls were not working since those nodes weren't visited yet. Also fixes the unit-tests.
* Utils::containsNonConstMemberCall() now honours assign operatorSergio Martins2017-12-088-41/+44
| | | | | | | Was returning false previously. Fixes a false positive in range-loop. CCMAIL: Ch.Ehrlicher@gmx.de
* Fix build due to missing includeSergio Martins2017-12-041-0/+1
|
* Don't try to fix strings with escaped bytesSergio Martins2017-12-033-1/+36
| | | | | | | | | While QString::fromUtf8("ö") is equal to QString::fromUtf8("\xc3\xb6"), QStringLiteral("ö") is not equal to QStringLiteral("\xc3\xb6") because the escaped bytes won't be converted to utf-16 but instead used directly as is. CCMAIL: faure@kde.org CCMAIL: montel@kde.org
* connect-3arg-lambda: refactor this checkSergio Martins2017-12-011-17/+22
| | | | | No longer advises to pass 'this' as 3rd argument, because it might make sense to pass another object there, should be the user to decide.
* Update ClazyAnchorHeader.h for the 1.3 releaseSergio Martins2017-11-241-2/+22
|
* Introduce lambda-unique-connectionSergio Martins2017-11-183-0/+133
| | | | | | | Qt::UniqueConnection only works with member functions, not functors or lambdas. CCMAIL: filipe.azevedo@kdab.com
* const-signal-or-slot: Don't print an error when connecting to lambdaSergio Martins2017-11-151-3/+1
| | | | | | returning nullptr is enough BUG: 386940
* old-style-connect: Only execute Qt specific code in Qt developer modeSergio Martins2017-11-081-2/+3
|
* old-style-connect: Fix fixit bug when using Q_PRIVATE_SLOTSergio Martins2017-11-081-4/+19
| | | | | | | | | | | If the QObject has a signal with the same name of a Q_PRIVATE_SLOT then the fixit would connect to the signal instead of the private slot. The fix is generic, if the user used the SLOT macro we simply never rewrite to use a signal. CCMAIL: dfaure@kdab.com
* Fix build on MinGW due to missing includeIvan Donchevskii2017-10-271-0/+1
|
* Add flag to change build of ClangLazy from shared to staticIvan Donchevskii2017-10-232-0/+4
| | | | Differential Revision: https://phabricator.kde.org/D8377
* Protect CheckManager with mutexKevin Funk2017-10-224-1/+28
| | | | | Needed in case Clazy is used as a plugin inside libclang. There, we might have multiple threads accessing CheckManager concurrently.
* Minor cleanup to previous commit regarding visit-implicit-codeSergio Martins2017-10-192-2/+2
| | | | CCBUG: 385851
* Add option to visit implicit code, such as compiler generated CTORsRichard Oehlinger2017-10-194-1/+17
| | | | | | | | | Although clazy's built-in checks don't need this, some custom user checks might Not enabled by default as it causes a slight slowdown BUG: 385851
* Merge branch '1.2'Sergio Martins2017-10-182-4/+4
|\
| * detaching-temporary: Don't warn for intersect, unite and subtractSergio Martins2017-10-151-3/+3
| | | | | | | | | | | | These are fine to be called on a temporary, since they return a reference to the object itself, which then must be stored or something done with it due to Q_REQUIRED_RESULT
| * Add missing \n in debug messageSergio Martins2017-10-151-1/+1
| |
* | Inform which is the second argument => number of argumentMontel Laurent2017-10-161-1/+1
| | | | | | | | | | | | | | | | + add space after this one. It avoids line as : error, invalid argIndex 3 3/compile/kde5/framework/kde/pim/kmail-account-wizard/src/cryptopage.cpp:226:9: warning: Pass 'this' as the 3rd connect parameter [-Wclazy-connect-3arg-lambda] => we have "3" without space so we can't copy path directly and we didn"t know when was the second "3"
* | Merge branch '1.2' into masterSergio Martins2017-10-125-5/+15
|\|
| * rule-of-three: Don't warn for std:: typesSergio Martins2017-10-121-1/+1
| |
| * Improve the warning message for some checksSergio Martins2017-10-123-4/+4
| |
| * qenums: Don't warn when importing enums from other classesSergio Martins2017-10-121-0/+10
| | | | | | | | Q_ENUM() doesn't support that
* | qproperty-without-notify: Don't warn for Q_GADGETSSergio Martins2017-10-122-1/+16
| | | | | | | | CCMAIL: jpnurmi@qt.io
* | Reduce some duplicated codeSergio Martins2017-10-122-2/+8
| |
* | Merge branch '1.2' into masterSergio Martins2017-10-111-0/+6
|\|
| * old-style-connect: Add a fixit for one more caseOlivier Goffart (Woboq GmbH)2017-10-111-0/+6
| | | | | | | | | | Fixes fixit when the argument types have different const reference, for example "const QString &" vs "QString", which are compatible.
* | Introduce thread-with-slotsSergio Martins2017-10-086-24/+217
| | | | | | | | | | | | | | | | | | slots in a QThread derived classe is usually a code smell, because they'll run in the thread where the QThread QObject lives, and not in the thread itself. Disabled by default, you'll have to explicitly enable it, and check case by case if you have races.
* | Introduce const-signal-or-slot, to find getters unintentionally marked as slotsSergio Martins2017-10-053-0/+151
| | | | | | | | | | | | | | Warns for: - non-void const method marked as slot - const method marked as signal - connecting to a method which isn't marked as slot, is const and returns non-void
* | Merge branch '1.2' into masterSergio Martins2017-10-051-0/+1
|\|
| * unused-non-trivial-variable: Fix false positive when decl begins with macroSergio Martins2017-10-051-0/+1
| | | | | | | | | | | | We check if the variable is used after the location where it was declared, but it the expression begins with a macro, it was using the location of where the macro was written, and where it was used.
* | Only run QtUtils::isBootstrapping() if we're building QtSergio Martins2017-10-032-2/+3
| | | | | | | | for performance reasons
* | Merge branch '1.2' into masterSergio Martins2017-10-031-0/+7
|\|
| * qstring-allocations: Don't warn during bootstrapSergio Martins2017-10-031-0/+7
| | | | | | | | | | Too many false positives due to QString::fromLatin1() being used instead tr().
* | Fix false warning in qhash-namespace check for enum (classes)Christian Ehrlicher2017-09-281-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Summary: When a qHash() function is defined for a enum class in a namespace, clazy wrongly prints the warning 'Move qHash(Foo::Bar) out of namespace Foo' because ContextUtils::namespaceForType() does not consider enumerations. Therefore I changed QualType::getAsCXXRecorDecl() to QualType::getAsTagDec() which returns the base class of RecordType and EnumType (and is the one which is passed to namespaceForDecl()) Type::getAsTagDecl() should be slightly faster since it avoids a dyna cast, don't know since when this function is available though, at last it's in 4.0.0 version: clang version 4.0.1 (tags/RELEASE_401/final 305264) Attn: there is also a change in main.cpp.expected since clang 4.0.0 prints qHash(IntFoo) instead the expected qHash(NS::IntFoo) in the warning output ... Test Plan: - added enum class check to qhash-namespace unit test - run test without modification which fails with the additional warning +qhash-namespace/main.cpp:22:5: warning: Move qHash(NS::EnumClass) out of namespace NS [-Wclazy-qhash-namespace] - apply my modification, re-run unit test, no additional warnings Reviewers: smartins Reviewed By: smartins Subscribers: chehrlic Differential Revision: https://phabricator.kde.org/D8010
* | qhash-namespace: Very qhash() is inside QT_*_NAMESPACE macrosSergio Martins2017-09-244-5/+77
| | | | | | | | | | This will only be done if -qt-developer is passed, as it's not useful for user code
* | Added -qt-developer optionSergio Martins2017-09-243-1/+15
| | | | | | | | | | | | When building Qt it will hounor Qt guidelines. Used for performance reasons, so it doesn't run stuff that doesn't apply to user code
* | Introducing qhash-namespaceSergio Martins2017-09-244-0/+157
| | | | | | | | Checks that qHash() is declared inside the namespace of it's argument.
* | Introducing overridden-signalSergio Martins2017-09-226-7/+171
| | | | | | | | | | | | | | | | | | | | Warns when overriding a signal, which might make existing connects not work, if done unintentionally. Warns for: - Overriding signal with non-signal - Overriding non-signal with signal - Overriding signal with signal
* | Merge branch '1.2' into masterSergio Martins2017-09-213-20/+55
|\|
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-09 22:18:23 +0000