+1
I'm also not sure that a smart pointer wrapper type brings a lot of
value; at the very least it is highly disruptive to existing code bases.
The Core C++ guidelines are clear that smart pointers imply ownership of
some sort, while naked pointers are non-owning; the various pointers to
QObjects stored as member variables of other QObject sub-classes are
indeed non-owning so there is no semantics or convention problem there IMO.
But I bit more than that - the promise a smart pointer of this kind can
make is something along the lines of "as long as I'm in scope, the
contained QObject pointer will at all times have some other object
responsible for freeing it" can't really be fulfilled by another class -
it can check that at creation, but can't prevent a object losing its'
parent via a `setParent(nullptr)`. Only the QObject class itself can
enforce that I think.
To that extent, it can indeed be useful to have a `make_child` member
template, or a free-standing `qMakeParented` function template, or
something else like that - which can hide the "new" call to avoid the
problem of having new-s without matching delete-s in the code while
setting the created object in a mode that ensures it can't become
parent-less.
Best,
Igor
On 16/10/2025 12:42, André Somers via Development wrote:
Hi,
On 15-10-2025 16:30, Daniel Schürmann wrote:
Hello
My name is Daniel Schürmann and I am one of the core members in the
Mixxx team. We develop the QT based FOSS DJ software Mixxx.
Since a couple of years we use a custom smart pointer to wrap
pointers to QObjects, where the lifetime is managed by the QT object
tree.
This is very useful for two reasons:
* In reviews it is clear that there is no explicit delete required,
it eliminates the nowadays exceptional case of new() without delete().
* The pointer asserts that the object has actually a parent and is
not leaked.
It has proven to be useful and disappointing that it is not available
in all my QT projects. That's why I like to discuss to put it or
anything similar into the QT source itself.
I have provided some more details in a QT bug here:
https://bugreports.qt.io/browse/QTBUG-141134
What do you think?
I have seen similar approaches in customer code before, but I don't
quite see what parented_ptr itself brings to the table here?
You can work around the "no naked new()" rule by creating a custom
template function that creates the type explicitly and enforces a
non-null parent. I don't understand why you need a new pointer type
for that. And for monitoring the validity of pointers to QObjects, we
already have QPointer.
As the "naked pointer" way of working is ingrained quite deeply into
Qt, I don't quite see how you expect this to find a place in the
current API either?
Perhaps it might be easier or better to add a make_child template
method to QObject that returns a new instance of the template type,
parented to the object you call it on? So you could do something like:
auto button = myDialog->make_child<QPushButton>(tr("Some button"));
instead of
auto button = new QPushButton(tr("Some button"), myDialog);
I think it should be possible to specialize these where needed (like
in this case, where the expected parent is actually a QWidget, not a
QObject), but I didn't try.
Cheers,
André
--
Development mailing list
[email protected]
https://lists.qt-project.org/listinfo/development
