[Bug c++/95849] New: Universal forwarding constructor inheritance resolution issue

[njormrod at fb dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95849

Bug ID: 95849
   Summary: Universal forwarding constructor inheritance
resolution issue
   Product: gcc
   Version: 10.1.0
Status: UNCONFIRMED
  Severity: normal
  Priority: P3
 Component: c++
  Assignee: unassigned at gcc dot gnu.org
  Reporter: njormrod at fb dot com
  Target Milestone: ---

Universal forwarding constructors have a known oddity: they are a better match
than the copy constructor for non-const objects.

If a derived class inherits its parent's constructors, where the parent has a
universal forwarding constructor, then the derived class's default copy
constructor (having signature `const Derived&`) should likewise be usurped by
the non-const universal-forwarding constructor.

This is not the case in gcc.

Code demonstrating the bug: foo() and bar() should both return 8, but bar()
actually returns 7.

```
struct A {
A(const A&) : x(7) {}

template 
A(U&&) : x(8) {}

int x;
};

int foo() {
A o1(true);
A o2(o1);
return o2.x;
}

struct B : A {
using A::A;
};

int bar() {
B o1(true);
B o2(o1);
return o2.x;
}
```


Godbolt link for gcc: https://gcc.godbolt.org/z/Y_sXD6. Bug is present.
Godbolt link for clang: https://gcc.godbolt.org/z/BFjYph. Bug is present.
Godbolt link for msvc: https://gcc.godbolt.org/z/GQ2z3x. msvc has the correct
behavior.

[Bug libstdc++/93205] New: std::discrete_distribution's operator>> causes OOM

[njormrod at fb dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93205

Bug ID: 93205
   Summary: std::discrete_distribution's operator>> causes OOM
   Product: gcc
   Version: 7.5.0
Status: UNCONFIRMED
  Severity: normal
  Priority: P3
 Component: libstdc++
  Assignee: unassigned at gcc dot gnu.org
  Reporter: njormrod at fb dot com
  Target Milestone: ---

std::discrete_distribution's operator>> should work with an invalid stream
(29.6.1.6, Table 105, row 15). Instead, it may OOM.

Relevant code from libstdc++/bits/random.tcc:
   2770:  size_t __n;
   2771:  __is >> __n;
   2772:
   2773:  std::vector __prob_vec;
   2774:  __prob_vec.reserve(__n);

If the istream fails to read __n on line 2771, then the vector is reserved with
garbage size on line 2774, potentially OOMing.

This could be fixed by initializing __n to 0 on line 2770.

[Bug libstdc++/93205] std::discrete_distribution's operator>> causes OOM

[njormrod at fb dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93205

--- Comment #1 from Nicholas Ormrod  ---
(This bug was discovered when some empty-istream test code OOMed)

[Bug c++/98938] New: throw calls move constructor

[njormrod at fb dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98938

Bug ID: 98938
   Summary: throw calls move constructor
   Product: gcc
   Version: 10.1.0
Status: UNCONFIRMED
  Severity: normal
  Priority: P3
 Component: c++
  Assignee: unassigned at gcc dot gnu.org
  Reporter: njormrod at fb dot com
  Target Milestone: ---

Throwing invokes the move constructor instead of the copy constructor.

Demonstration of move-constructor in godbolt: https://gcc.godbolt.org/z/zhjv5z
Demonstration of copy-constructor using clang: https://gcc.godbolt.org/z/E3MEef

>From 14.2 Throwing an exception [except.throw]: "Throwing an exception
copy-initializes (9.4, 11.4.5.3) a temporary object, called the exception
object."

Code (as used in godbolt):

```
#include 

struct A : std::exception {
A(int* q) : p(q) {}
A(const A& a) : p(a.p) { *p = 12345; }
A(A&& a) : p(a.p) { *p = 56789; }

int* p;
};

int foo() {
int i = 10;
A a(&i);
try {
throw a;
} catch (A&) {}
return i;
}
```