MySQL Bugs: #92039: GCC 9 warnings

Bug #92039	GCC 9 warnings
Submitted:	16 Aug 2018 8:38	Modified:	22 Aug 2018 12:51
Reporter:	Steinar Gunderson	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Compiling	Severity:	S3 (Non-critical)
Version:	8.0.15	OS:	Any
Assigned to:		CPU Architecture:	Any

Description:
GCC 9 has some new warnings. In particular, it now warns about use of implicit copy constructors in classes with nontrivial destructors (and in a few other cases, such as operator=), which is deprecated in C++11:

class Foo {
  ~Foo();  // nontrivial destructor (even if it is actually empty)
};
void func(Foo a) {
  // ...
}
Foo b;
func(b);  // warning, invoking implicit Foo(Foo &), which is deprecated for Foo

The fix is either to remove ~Foo() where possible, or to make the copy constructor explicit:

class Foo {
  ~Foo();
  Foo(Foo &) = default;  // now explicit, so no longer deprecated
};

However, do note that making an explicit copy constructor disables the implicit assignment operator. If you want it, you'll need to specify it:

class Foo {
  ~Foo();
  Foo(Foo &) = default;
  Foo &operator=(const Foo &) = default;  // must be explicit due to previous line
};

And this, in turn, disables the implicit move assignment operator. Since you want it back, you need to write:

class Foo {
  ~Foo();
  Foo(Foo &) = default;
  Foo(Foo &&) = default;  // move constructor must be explicit because copy is
  Foo &operator=(const Foo &) = default;
  Foo &operator=(Foo &&) = default;  // move assignment must be explicit because move is
};

This is all rather cumbersome for virtual base classes, but that's what the standard says, and GCC has chosen to warn on the standard. See GCC bug #58407.

How to repeat:
N/A

Suggested fix:
Remove the destructor where possible. Otherwise, add the four lines as suggested.

Posted by developer:
 
Oh, I forgot to write: Making an explicit copy constructor will also disable the default constructor. So if you want it, you'll also need to do

  Foo() {}

Posted by developer:
 
Fixed in 8.0.14.

Configuration/build cleanup. No changelog entry needed.

I still cannot compile current mysql trunk on current gcc trunk due to this:
../../mysys/my_init.cc:475:63: error: could not convert ‘{0, "Waiting for disk space", 0, ""}’ from ‘<brace-enclosed initializer list>’ to ‘PSI_stage_info’ {aka ‘PSI_stage_info_v1’}
  475 |                                                PSI_DOCUMENT_ME};
      |                                                               ^
      |                                                               |
      |                                                               <brace-enclosed initializer list>
make[2]: *** [mysys/CMakeFiles/mysys.dir/my_init.cc.o] Error 1

any idea?

Posted by developer:
 
I'm unable to reproduce this. Save for a new GCC 9 internal compiler error that I'm reporting now (which forced me to use -DWITH_SYSTEM_LIBS=1), all of MySQL compiles warning-free with the latest GCC 9 snapshot I have here (20181019). Are you using any strange flags? What does the compile line for my_init.cc.o look like?

Posted by developer:
 
Fixed in 8.0.15.

Code cleanup. No changelog entry needed.