Not an ideal solution; ideally we should not throw the exceptions,
and stop the parser as soon as something bad like this happens; but
hopefully exception throwing is reasonable exceptional.
Change-Id: If619592533b2929c671e2b03eb8a83480bd92c54
This required some changes to the framework:
* Init-/DeInitVCL is no longer done per individual test in BootstrapFixture, but
once per CppunitTest invocation in a new vclbootstrapprotector (similarly to
the exisiting unobootstrapprotector). CppunitTests that need VCL now need to
declare gb_CppunitTest_use_vcl.
* For things to work properly, the UNO component context needs to be disposed
from within DeInitVCL (cf. Desktop's Application::DeInit called from
DeInitVCL). The easiest solution was to introduce an
Application::setDeInitHook (where the hook is called from DeInitVCL)
specifically for vclbootstrapprotector to call.
* PythonTests don't (yet) call DeInitVCL; they still hook into
BootstrapFixture's original test_init functionality (to call InitVCL), and do
not make use of the vclbootstrapprotector.
Change-Id: I4f3a3c75db30b58c1cd49d81c51db14902ed68b2
It appears that the C++ standard allows overriding destructors to be marked
"override," but at least some MSVC versions complain about it, so at least make
sure such destructors are explicitly marked "virtual."
Change-Id: I0e1cafa7584fd16ebdce61f569eae2373a71b0a1
...mostly done with a rewriting Clang plugin, with just some manual tweaking
necessary to fix poor macro usage.
Change-Id: I71fa20213e86be10de332ece0aa273239df7b61a
It's not very efficient, because we generally end up copying it twice -
once into the parameter and again into the destination OUString.
So I create a clang plugin that finds such places and generates a
warning so that we can convert them to pass-by-reference.
Change-Id: I5341a6ea9e3190f4b4c05c42c85595e3dcd83361
This change is inspired by startElementNS & singleElementNS which
have namespace as first parameter to define the namespace of an
element. Some attributes of a element can be in different namespace
but until now FastAttributeList "add" method did not have a namespace
parameter. This commit adds "addNS" which accepts namespace as a
first parameter.
Change-Id: Iebf8b5e890c17f00a56923efc3506580eed070a9
a pity the other one wasn't fixed when the first one was fixed
back in 2002 with 0645e6a1a1917fef833358f328b5d56396380698
Change-Id: Ia7ac0f57a6805dc8ff09668198d0530b1cfc31db
This is much better approach compared to the callback function, as it allows
passing arguments to the c++ constructor directly, while still allowing some
additional initialization after having acquired the instance.
Change-Id: I5a0f981915dd58f1522ee6054e53a3550b29d624
Many of the initalizations (in eg. framework) have to be done on an
acquire()'d object, so instead of doing the initialization directly, return
the initialization member function back to the createInstance() /
createInstanceWithContext() / ... and perform the initialization there.
As a sideeffect, I belive the calling initialize() from servicemanager is not
that much a hack any more - whoever converts the implementation to be
constructor-base has the choice to provide the callback, or still initialize
through XInitialization, where the callback is preferred by servicemanager
when it exists.
Change-Id: I8a87b75c54c1441ca0f184967d31ff4902fc4081
Most of the constructors are supposed to be only a call of
new TheInstance(arguments)
or an equivalent; so let's just change the constructor caller accordingly, to
accept unacquired new instance.
If there are exceptions that need to do more heavy lifting, they do not have
to use the constructor feature, or there can be a wrapper for the real
implementation, doing the additional work in their (C++) constructor.
Change-Id: I035c378778aeda60d15af4e56ca3761c586d5ded
