[PATCH] make cmakeduchaintest run without asserts

[Peter Oberndorfer]

On Sonntag 24 Mai 2009, David Nolden wrote:
> Am Samstag 23 Mai 2009 23:18:05 schrieb Peter Oberndorfer:
> > vm.insert CMAKE_CURRENT_SOURCE_DIR is needed since 953842
> > where a Q_ASSERT was added to make sure CMAKE_CURRENT_SOURCE_DIR is defined
> >
> > addDocumentChain in testUses is needed to avoid assert indexed.data() ==
> > context in removeDocumentChain that fires when a addDocumentChain was never
> > called
> >
> > addDocumentChain + removeDocumentChain testDUChainWalk
> > are needed to avoid assert m_recursiveImports.isEmpty()
> > in TopDUContextLocalPrivate::clearImportedContextsRecursively
> > ---
> >
> > I am no duchain/cmake expert, so please review carefully.
> > Also i don't know if addDocumentChain + removeDocumentChain
> > is the right cure for the m_recursiveImports.isEmpty() assert.
> Your duchain fixes are correct. A top-context is only partially functional 
> when it was not added using addDocumentChain.
> 
> Greetings, David
> 
Good,
while testing my changes for cmakeduchaintest i found another problem,
which happend to me several times during shutdown of kdevelop after normal work.
But it can be reproduced easier with this test.

It manifests itself in a QMutex::lock: mutex lock failure

KDevelop::shouldDoDUChainReferenceCountingInternal can access the mutex in the already destructed RefCountDecider.
I also attached the part of the valgrind log.

The attached test patch 1 makes this obvious by adding a assert.
Attached test patch 2 "solves" this by never destructing RefCountDecider.
You have another idea how to solve this properly?
(My hack is prbably not thread safe)

Greetings Peter

==24232==
==24232== Invalid read of size 1
==24232==    at 0x4088B7E: QMutex::lock() (qmutex.cpp:152)
==24232==    by 0x42E2318: KDevelop::shouldDoDUChainReferenceCountingInternal(void*) (referencecounting.cpp:57)
==24232==    by 0x4369C62: KDevelop::AppendedList<true, KDevelop::IndexedTypeIdentifier>::free(KDevelop::IndexedTypeIdentifier*) (appendedlist_static.h:94)
==24232==    by 0x437286E: KDevelop::IdentifierPrivate<true>::~IdentifierPrivate() (identifier.cpp:45)
==24232==    by 0x4363CDA: KDevelop::Identifier::~Identifier() (identifier.cpp:301)
==24232==    by 0x58DAE0B: __cxa_finalize (in /lib/libc-2.8.so)
==24232==    by 0x42BA2A8: ??? (in /home/kde-devel/kde/lib/libkdevplatformlanguage.so.1.0.0)
==24232==    by 0x44422B3: ??? (in /home/kde-devel/kde/lib/libkdevplatformlanguage.so.1.0.0)
==24232==    by 0x400E62D: _dl_fini (in /lib/ld-2.8.so)
==24232==    by 0x58DAB0B: exit (in /lib/libc-2.8.so)
==24232==    by 0x58C4613: (below main) (in /lib/libc-2.8.so)
==24232==  Address 0x6680920 is 0 bytes inside a block of size 96 free'd
==24232==    at 0x40235AC: operator delete(void*) (vg_replace_malloc.c:313)
==24232==    by 0x4088E24: QMutex::~QMutex() (qmutex.cpp:134)
==24232==    by 0x42E312E: RefCountDecider::~RefCountDecider() (referencecounting.cpp:37)
==24232==    by 0x58DAB0B: exit (in /lib/libc-2.8.so)
==24232==    by 0x58C4613: (below main) (in /lib/libc-2.8.so)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-test-patch-to-assert-when-a-already-dead-RefCountDec.patch
Type: text/x-diff
Size: 1698 bytes
Desc: not available
URL: <http://mail.kde.org/pipermail/kdevelop-devel/attachments/20090524/dd85c13c/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-test-patch-to-not-destroy-RefCountDecider-during-lib.patch
Type: text/x-diff
Size: 4988 bytes
Desc: not available
URL: <http://mail.kde.org/pipermail/kdevelop-devel/attachments/20090524/dd85c13c/attachment-0001.bin>