[intro.object] Definition of "most derived object" does not match its use CWG 1517 #2950
Labels
cwg
Issue must be reviewed by CWG.
Comments
jensmaurer
added
the
decision-required
jensmaurer
changed the title
|
Editorial meeting: During the period of construction, the dynamic type might be different from the most derived object. "typeid" currently refers to most derived object, though. See CWG1517 for rewrite of class.ctor. The "type of an object" should not change throughout the lifetime. |
jensmaurer
added
cwg
|
"typeid" issue submitted to CWG: http://lists.isocpp.org/core/2019/09/7236.php |
|
"most derived object" definition issue submitted to CWG: http://lists.isocpp.org/core/2019/09/7237.php |
|
@jensmaurer Any way to get access to the list? Publicly available one is quite outdated. |
jensmaurer
changed the title
|
I'm wondering what is the most derived object in this example as per the current definition struct A{
virtual ~A(){}
};
struct B:A{
B(){
A* ptr = this;
std::cout<< typeid(*ptr).name()<<std::endl; // [expr.typeid] p2 applies here.
}
};
struct C:B{
};
int main() {
C c;
} |
|
That particular example is covered by the resolution of CWG1517. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
According to [intro.object] p6, a most derived object is an object of a most derived class type or of a non-class type, however, almost all mentions of it in the standard use it as a property that relates two object to each other (Ex:
Ais the most derived object ofB), while the definition currently just describes an object (Ex:Ais a most derived object).Examples of use:
[defns.dynamic.type] type of the most derived object to which the glvalue refers
In this definition, most derived object is used as if it relates two objects, which it doesn't. What this means is that "dynamic type" is undefined for an expression that refers to an object that is not a complete object. For instance, if you had a a reference to
B, and you initialized it with an object of typeD, whereBis a base class ofD, the reference would be bound to the base class subobject, rather than the object itself, so the dynamic type of an expression naming that reference would be undefined, according to the current definition.[class.ctor] p5 Here, the wording alludes to "the most derived object", referring to some specific object, that has not been defined.
There are many such cases like these, and can be fixed by defining "most derived object" as a term that relates two objects like so:
Some rewording would be required for clauses that currently refer to most derived objects (they are incorrect even now), which I will submit in a pull request if this change is deemed good (I'm on vacation right now, so it will take about a week)
The text was updated successfully, but these errors were encountered: