aboutsummaryrefslogtreecommitdiffstats
path: root/sources/shiboken2/libshiboken/basewrapper.cpp
Commit message (Collapse)AuthorAgeFilesLines
* Ensure that signature strings never overflow againChristian Tismer2019-04-161-2/+2
| | | | | | | | | | | | | | | | | | | | The signature module used to use large strings with the signatures of all functions in a class. This can lead to an overflow in MSVC, because the maximum string length funnily still is 32K unicode characters. This patch solves that by using a single string per function. Instead of a huge string, a list of strings is passed to each class. To prevent any runtime increase, the string list creation is deferred until the actual usage. At initialization time only a ssize_t holding the structure address is passed. As a result, the signature module should be even slightly faster. Task-number: PYSIDE-955 Change-Id: I99faf942a3cca03456928b8aec5e8a4b9924b8b2 Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
* Cleanup And Simplify the Signature Related CodeChristian Tismer2019-04-161-18/+8
| | | | | | Task-number: PYSIDE-510 Change-Id: I73c441b56a19a0ac836e3598ff6fc8c9ba4d1cd2 Reviewed-by: Christian Tismer <tismer@stackless.com>
* Fix Shiboken.ObjectType.__new__ for Python 2.7Christian Tismer2019-02-071-0/+13
| | | | | | | | | | | | | | | | | | | | | | | | | | | type(QtWidgets.QWidget).__new__(type(QtWidgets.QWidget), "", (), {}) gave a problem in Python 2.7 after transition to PEP 384. The reason for the problem is a check in Python 2.7 that tries to find out if the function used to create a new object is a secure (builtin) one. Therefore, all new types that are generated by a Python function are filtered out. Unfortunately, Python 2.7 did that assuming that only Python classes are heap types. This is at least no longer true since Python 3 migrated to the new type API where all new types are heap types. The internal check was therefore changed to do the test for a builtin "new" function differently. But not in Python 2.7 . The workaround was to create the Shiboken.ObjectType as a heap type and then remove that flag from the type. This seems to have no bad effects, probably because the types did barely change when doing the transition. Anyway, I will stay tuned and watch out if this later creates a problem. Task-number: PYSIDE-816 Change-Id: Ia596716b0e5dff3f1a7155ab6765186ef0b9d179 Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io>
* Add pyObj check to setSequenceOwnershipCristian Maureira-Fredes2019-02-061-0/+4
| | | | | | | | | | | | | Ownership is always set recursively, so for each element we check the children of the object. When we don't have any, and also no parent the mechanism we have fails, so for constructors that allow having a null parent, we need to just skip this function. Change-Id: Iee6a81409cad7dd08c6ecbafd1b11c9a7bb0db85 Fixes: PYSIDE-922 Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
* Produce TypeError Messages Using the Signature ModuleChristian Tismer2019-01-101-36/+4
| | | | | | | | | | | | | | | The TypeError messages can now be produced, based upon the signature module. As a feature under test, we produce ValueErrors instead in certain cases. This will probably improve, later. We are currently investigating how much can be determined, automatically. Task-number: PYSIDE-795 Change-Id: Ie8a648beaf8a3bed388e3c01ba501bb36859722e Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io> Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io>
* Fix crash related to multiple inheritanceFriedemann Kleint2019-01-031-8/+20
| | | | | | | | | | | | | | | | | | | | | | | | | | In the <class>_PTR_CppToPython_<class> converter function (written by CppGenerator::writeConverterFunctions()), the generated code used typeid(*ptr).name() to retrieve the name to use for the SbkObjectTypes. This construct returns the name of the outermost class (for example, "QWidget" for a QWidget-type paint device returned by QPainter::device()), as opposed to "QPaintDevice *" returned by typeid(ptr).name(). This caused a crash with multiple inheritance since QWidget inherits QObject and QPaintDevice and the "QWidget" type was associated with the QPaintDevice pointer. To fix this: - Add API to libshiboken to obtain the SbkObjectType* by name and check for the presence of a special cast function (multiple inheritance). - Generate the code of <class>_PTR_CppToPython_<class> as follows: Check whether the outermost type obtained by typeid(*ptr).name() has a special cast function. If that is the case, use the type name obtained by typeid(ptr).name() (base class) to create the wrapper. Change-Id: I8ee6b4c084e9dafa434623433661809b83aedee5 Fixes: PYSIDE-868 Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io>
* Merge remote-tracking branch 'origin/5.11' into 5.12Friedemann Kleint2018-10-291-2/+4
|\ | | | | | | Change-Id: I3bb491686968e81382c135ab737da259d9796f52
| * Fix Memory Leak Caused By Wrong Limited API DefaultChristian Tismer2018-10-291-2/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When a type has nullptr as tp_dealloc, there apply different defaults. Static types had object_dealloc as default, while new heaptypes created with type_new have subtype_dealloc as default. A problem was now that PyType_FromSpec also has subtype_dealloc as default. But that is wrong, because a type that was written with the static type approach is already written with object_dealloc in mind and takes somehow care about further issues with that type. When we now convert this type and suddenly use subtype_dealloc instead of object_dealloc, things get pretty wrong. Finding that out was pretty hard and took quite long to understand. The fix was then very easy and is the best proof: Replacing our former (wrong) solution of supplying an SbkDummyDealloc with a function object_dealloc works perfectly, and the leakage completely vanished. The documentation now is also corrected. Task-number: PYSIDE-832 Change-Id: Ifc20c28172eb5663cd5e60dac52e0a43acfb626c Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
* | Fix crash when garbage collecting in a non-GUI threadFriedemann Kleint2018-10-151-4/+38
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | If a GUI class happens to be detected unreferenced when garbage collecting in a non-GUI thread and is subsequently deleted, crashes can occur for QWidgets and similar classes. The hitherto unimplemented delete-in-main-thread" attribute should be used. Add the missing implementation. Add the field to shiboken's type entry and SbkObjectTypePrivate class and pass it via newly introduced flags to introduceWrapperType(). Defer the deletion when invoked from the background thread and store the list of destructors in a list in binding manager run by Py_AddPendingCall(). Task-number: PYSIDE-743 Task-number: PYSIDE-810 Change-Id: Id4668a6a1e32392be9dcf1229e1e10c492b2a5f5 Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org> Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io> Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* | libshikoken: Refactor the Visitor class hierarchyFriedemann Kleint2018-09-191-47/+50
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Implementing the deleteInMainThread feature requires being able to collect a list of destructors and potentially storing it. This requires splitting out the actual deallocation/destructor calls from the DtorCallerVisitor and DeallocVisitor classes. Since this is the only use of the virtual HierarchyVisitor::done() method (and it does not really belong to the visitor pattern), remove it. Change the void visit() method into a bool from which true can be returned to terminate. The finish()/wasFinished() methods can then also be removed from HierarchyVisitor and the code simplified accordingly. Replace the DtorCallerVisitor and DeallocVisitor classes by DtorAccumulatorVisitor that collects a list of DestructorEntry structs containing destructor function and C++ instance. Polish the code a bit, use member initialization, add override, move implementations to source and some spacing for clarity. Task-number: PYSIDE-810 Change-Id: I5e3ef6df753679ec111a5f0d1b75305bd5cf1c0c Reviewed-by: Christian Tismer <tismer@stackless.com>
* | Merge remote-tracking branch 'origin/5.11' into devFriedemann Kleint2018-09-041-2/+2
|\| | | | | | | Change-Id: Ie8025300580981c5349b31d4846a7f659481991d
| * libshiboken: Use empty() to check for empty containers as clang-tidy advisesFriedemann Kleint2018-08-301-2/+2
| | | | | | | | | | | | | | | | | | Using the size() method for that purpose can be expensive for standard containers. Task-number: PYSIDE-727 Change-Id: I0da34e271503384a741d856fff5e84fee67bc97f Reviewed-by: Christian Tismer <tismer@stackless.com>
* | libshiboken: Fix container typesFriedemann Kleint2018-07-311-70/+64
| | | | | | | | | | | | | | | | | | | | | | | | | | | | Change the ref count map to a unordered_multimap and the remaining lists to vectors. A linked std::list is not suitable for the lists used in libshiboken. Task-number: PYSIDE-727 Change-Id: Ibd65486a58cf43ac66b981bea65597df5a732b63 Reviewed-by: Christian Tismer <tismer@stackless.com> Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io> Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
* | libshiboken: Refactor loopsFriedemann Kleint2018-07-301-77/+52
|/ | | | | | | | | | | Avoid repeated invocation of container.end() in the loop condition by either assigning to a variable or use range-based-for where possible. Use auto for iterators to allow for changing the container type. Task-number: PYSIDE-727 Change-Id: I5217207a3a7fb60823ccbcbf0915bf6cf21a3e4d Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* Pep 384 Final Cut: Remove PepTypeChristian Tismer2018-07-231-16/+16
| | | | | | | | | | | | | | | | | | | | | | | | | While trying to document the Limited API Project, it suddenly struck me: We can make the patch much much simpler and implement it without the necessity to have an extra PepType! Now I am happy to continue the documentation, because it is now no more improvable. This version will last as long as the layout of PyTypeObject does not change substantially. When that happens, then we need to rewrite stuff with the according PyType_GetSlot() access functions. These access functions will until then be complete enough so that we can live without the tricks like inventing a reduced PyTypeObject as was done in the current implementation. Task-number: PYSIDE-560 Change-Id: I49849cc377baa6794a5b53292691e21d6e2853ab Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org> Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io> Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* libshiboken: Use std::set::find() instead of searching the setFriedemann Kleint2018-07-191-1/+1
| | | | | | Task-number: PYSIDE-727 Change-Id: Ia79f34ed466eb7652946e49db6ef6db8dd9fa03d Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* Fix limited API incompatibility with Python 3.7Christian Tismer2018-07-101-0/+4
| | | | | | | | | | | | | | | | | When Python 3.7 appeared, the structure of pystate.h was changed, substantially. Unfortunately this structure contains the trashcan code, which is only available as a macro and not part of the limited API. This code is normally not used by application programs. It prevents crashes when chains of millions of objects are deallocated. I disabled this for now when the limited API is active. As soon as somebody complains about crashes, I will try to implement it again in a safe way, but I am not sure if it is worth it in the first place. Task-number: PYSIDE-737 Change-Id: Id0daf391448ddcb9df3d299f859ef024714fa736 Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
* Add ownership condition for finite PySequencesCristian Maureira-Fredes2018-07-031-1/+8
| | | | | | | | | | | | | | | | | | | | | After the fix for PYSIDE-441, another issue appeared related to QVariants and PySequences PYSIDE-641, which was related due to the nature of this python data type. The problem had the same root cause, using PySequences assuming they are always finite, but not including the case of a class implementing the __getitem__ method without a length. The fix for PYSIDE-441 did not include the option of having incomplete PySequences, so this change add an extra condition to transfer the ownership of a incomplete PySequence element. Task-number: PYSIDE-671 Change-Id: I72ed1f5ea51c0c5b5a40ec51ab850732eea3c3b9 Reviewed-by: Christian Tismer <tismer@stackless.com>
* Undo renaming of Pep384_InitChristian Tismer2018-06-061-1/+1
| | | | | | | | | | | | The functions which were named Pep384XXX were renamed in order to get "PepType" as a function-like macro that suggests the similarity between "PyTypeObject" and "PepTypeObject". But the renaming of the module initialization function was not intended. Change-Id: I555633ccbd8e1354c27f2c1957c81905be54d86b Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* PEP 384-squash: Implement PEP 384Christian Tismer2018-05-301-241/+259
| | | | | | | | | | This is the condensed checkin of 18 commits which created the implementation of PEP 384. Task-number: PYSIDE-560 Change-Id: I834c659af4c2b55b268f8e8dc4cfa53f02502409 Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org> Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* Change license from all the filesCristian Maureira-Fredes2018-05-031-1/+1
| | | | | | | | | | | | | | | | | Removing the word 'project' from all the headers, and changing the PySide reference from the examples to Qt for Python: The following line was used inside the source/ and build_scripts/ directory: for i in $(grep -r "the Qt for Python project" * |grep -v "pyside2-tools" | awk '{print $1}' | sed 's/:.*//g');do sed -i 's/the\ Qt\ for\ Python\ project/Qt\ for\ Python/g' $i;done and the following line was used inside the examples/ directory: for i in $(grep -r "of the PySide" * |grep -v "pyside2-tools" | awk '{print $1}' | sed 's/:.*//g');do sed -i 's/of\ the\ PySide/of\ the\ Qt\ for\ Python/g' $i;done Change-Id: Ic480714686ad62ac4d81c670f87f1c2033d4ffa1 Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io> Reviewed-by: Alex Blasche <alexander.blasche@qt.io>
* Rename PySide references to Qt for PythonCristian Maureira-Fredes2018-04-271-1/+1
| | | | | | | | | When referring to the project one should use "Qt for Python" and for the module "PySide2" Change-Id: I36497df245c9f6dd60d6e160e2fc805e48cefcae Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org> Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
* libshiboken: Code cleanupFriedemann Kleint2018-01-271-25/+0
| | | | | | | | Remove apparently unused/deprecated code. Change-Id: I566014bafdcbf0b3b46e9dc836c451c64d409511 Reviewed-by: Cristian Maureira-Fredes <cristian.maureira-fredes@qt.io> Reviewed-by: Christian Tismer <tismer@stackless.com>
* Prevent infinite loop due to bad __getitem__ implCristian Maureira-Fredes2018-01-151-7/+9
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Looking at PYSIDE-441 there was an issue regarding overloading the __getitem__ method on a class that inherits from QObject. The bug report showed that setting an object of the same class to be the parent of another object of the same class ended up causing an infinite loop when trying to get the parent element. Overloading __getitem__ implicitly converts the class into an iterator, and the developer *must* include a proper implementation of the method which raises a StopIteration exception when needed. Commonly, people that overload this method included access to class data structures where in most of the cases an IndexError is raised which forces the iteration to stop. Since the bug report did not include this code and also there was no access to any internal variable, no exception was raised and ended up causing an infinite loop. This can be replicated in python as folows: class A(object): def __getitem__(self, arg=None): print("getitem called:", arg) #raise StopIteration a = A() print(list(a)) This small fix avoids the infinite loop when the method __len__ is not implemented (length = -1) or when the length of the pyObj is zero. Without a proper implementation of __getitem__ (Raising IndexError or StopIteration) the infinite loop will happen. If __len__ is not implemented, then the application will complain, but does not matter since this is never checked during the iteration. Change-Id: I74e7bf1755c265dbc309bb6c5a760f11643fd7ed Task-number: PYSIDE-441 Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io> Reviewed-by: Christian Tismer <tismer@stackless.com>
* Merge remote-tracking branch 'origin/5.6' into 5.9Friedemann Kleint2017-11-061-2/+5
|\ | | | | | | Change-Id: I72583df407fc5b3caa8bf35fd997889a4ac86512
| * Implement proper void pointer (void*) supportAlexandru Croitor2017-11-061-0/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This change introduces a new type into the shiboken2 module which is imported by calling "import PySide2.support.VoidPtr". The type takes care of conversions from / to void* values in function signatures. Creating an instance can be done by passing either a shiboken wrapped object, or an integer representing an address, or a python object that implements the buffer interface. For example, this is useful for passing numpy arrays to C OpenGL functions that take void* parameters. First you convert the array into a bytestring (using numpy.array.tobytes(), then you instantiate a VoidPtr from that bytestring, and finally you pass it along to a GL function. One corner case that is currently not supported is void** parameters. Change-Id: I01e291d6856cb6bd8b5175adc3ead6b728036535 Reviewed-by: Christian Tismer <tismer@stackless.com>
| * Remove dead / unused code regarding conversionsAlexandru Croitor2017-11-021-2/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | There's a lot of code that was previously used for doing conversions between C++ / Python types (apparently relying on extensive RTTI / typeid manipulations), which got superseded by a cleaner templated approach. The old code was left behind, and there were a few instances where it was still used even though it wasn't needed, like in QtScript typesystem XML and shiboken's enum handling. Remove the old code, apply the small changes needed to make it work with new the code. This is cleanup to reduce the confusion regarding conversion behavior, and also preparation for a proper implementation of handling "void*" types. Change-Id: I8f16bb31436f9a677bb2d64c7197c4375005b656 Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
* | Merge remote-tracking branch 'origin/5.6' into 5.9Friedemann Kleint2017-10-271-5/+41
|\| | | | | | | Change-Id: I94cb5a7dab97cff3591bac534228bfd3e3ad5938
| * Support the qApp macro correctly, final version incl. debugChristian Tismer2017-10-271-5/+41
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | For short the new features: - there is a qApp in QtCore, QtGui and QtWidgets for compatibility, and also in __builtins__ for a true macro-like experience. - if you delete any qApp variable, the Q*Application is reset and you can start over. Long description: There is a qApp macro in Qt5 which is equivalent to Q*Application.instance() . Python does not have macros. Both PyQt5 and PySide2 have an according structure in QtWidgets. In the case of PySide2, the qApp variable is first initialized to None and later to QApplication(). This does not reflect the original sense of the qApp macro, because - it only handles QApplication, - it does not handle destruction. This "macro" should live in QtCore, but both PyQt5 and PySide2 decided to put this in QtWidgets. As a compromize, I propose to put qApp into all three modules, and into __builtins__ as well, so wherever you create an application, you find this "macro" in place. While changing the code, I stumbled over the template set_qapp_parent_for_orphan. I tried to make sense out of it and finally removed it. There were no side effects but bug PYSIDE-85 is gone, now. With some extra effort, I created a singleton qApp that changes itself. This way, a true macro was simulated. Note that this was not possible with a garbage collected variable, and I had to make shiboken aware of this. As the final optimization, I turned qApp also into a fuse variable: Delete any qApp variable and Q*Application will finish when there is no extra reference. Task-number: PYSIDE-85 Task-number: PYSIDE-571 Change-Id: I7a56b19858f63349c98b95778759a6a6de856938 Reviewed-by: Christian Tismer <tismer@stackless.com>
* | Merge remote-tracking branch 'origin/5.6' into 5.9Friedemann Kleint2017-09-281-5/+10
|\| | | | | | | Change-Id: I47521e21977b1f17fcc65590f565270b2440a48b
| * Implement introspection with __signature__ packageChristian Tismer2017-09-201-5/+10
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The signature module was turned into a package under 'PySide2/support/signature'. The package is completely isolated so that nothing is leaking into the normal import machinery. The package is also not initialized unless a __signature__ attribute is accessed. The only change to Python during a PySide run is the existence of the __signature__ attribute. As a side effect, all tests run at the same speed as before this extension. The module does not actively import PySide modules. Instead, it inspects sys.modules and reloads its mapping.py if needed. Example usage: >>> PySide2.QtWidgets.QGraphicsAnchorLayout.addAnchors.__signature__ >>> PySide2.QtWidgets.QGraphicsAnchorLayout.__signature__ The module has been thoroughly tested on macOS. I consider this ready. Task-number: PYSIDE-510 Change-Id: Ibb231a7fbb4ccc1a7249df55e3881a4e21a19c0d Reviewed-by: Christian Tismer <tismer@stackless.com>
* | Merge remote-tracking branch 'origin/5.6' into 5.9Friedemann Kleint2017-06-091-4/+4
|\| | | | | | | Change-Id: I73f39966a2af7aa935e1890e29c9b71573aae97c
| * libshiboken: Fix/Silence some MSVC warningsFriedemann Kleint2017-06-091-4/+4
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | autodecref.h(110): warning C4522: 'Shiboken::AutoDecRef': multiple assignment operators specified autodecref.h(78): warning C4800: 'PyObject *const ': forcing value to bool 'true' or 'false' (performance warning) conversions.h(282): warning C4800: 'long': forcing value to bool 'true' or 'false' (performance warning) basewrapper.cpp(625): warning C4800: 'int': forcing value to bool 'true' or 'false' (performance warning) basewrapper.cpp(654): warning C4800: 'SpecialCastFunction': forcing value to bool 'true' or 'false' (performance warning) basewrapper.cpp(1014): warning C4800: 'Shiboken::ParentInfo *': forcing value to bool 'true' or 'false' (performance warning) basewrapper.cpp(1044): warning C4800: 'void *': forcing value to bool 'true' or 'false' (performance warning) helper.cpp(56): warning C4244: 'initializing': conversion from 'Py_ssize_t' to 'int', possible loss of data shibokenbuffer.cpp(46): warning C4800: 'int': forcing value to bool 'true' or 'false' (performance warning) Change-Id: If1517fde8e7670f258a56f6d845a66ebb3d82141 Reviewed-by: Christian Tismer <tismer@stackless.com>
| * move everying into sources/shiboken2Oswald Buddenhagen2017-05-221-0/+1518
| | | | | | in preparation for a subtree merge. this should not be necessary to do in a separate commit, but git is a tad stupid about following history correctly without it.
* move everying into sources/shiboken2 (5.9 edition)Oswald Buddenhagen2017-05-221-0/+1516
in preparation for a subtree merge. this should not be necessary to do in a separate commit, but git is a tad stupid about following history correctly without it.
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-10 03:28:24 +0000