Bug 102378

Summary: missing -Waddress in template code at definition time rather than instantiation
Product: gcc Reporter: Martin Sebor <msebor>
Component: c++Assignee: Not yet assigned to anyone <unassigned>
Status: ASSIGNED ---    
Severity: enhancement CC: webrown.cpp
Priority: P3 Keywords: diagnostic
Version: 12.0   
Target Milestone: ---   
Host: Target:
Build: Known to work:
Known to fail: Last reconfirmed: 2021-09-16 00:00:00
Bug Depends on:    
Bug Blocks: 103360    

Description Martin Sebor 2021-09-16 21:19:54 UTC 
While implementing the suggestion in the review below I noticed that G++ fails to issue -Waddress in template code even when it could:
https://gcc.gnu.org/pipermail/gcc-patches/2021-September/579082.html

$ cat t.C && gcc -O2 -S -Wall -Waddress t.C
int f ()
{
  int a[2];
  return &a == 0;      // -Waddress (good)
}

template <class T>
int g ()
{
  {
    int a[2];
    return &a == 0;    // missing -Waddress
  }

  {
    T t;
    return &t == 0;    // no -Waddress for type-dependent expr (good)
  }

  {
    T a[2];
    return &a == 0;    // missing -Waddress
  }
}

template <class T>
int h ()
{
  {
    int a[2];
    return &a == 0;    // missing -Waddress
  }

  {
    T t;
    return &t == 0;    // missing -Waddress for T = int
  }

  {
    T a[2];
    return &a == 0;    // missing -Waddress
  }
}

template <int> int h ();
t.C: In function ‘int f()’:
t.C:4:13: warning: the address of ‘a’ will never be NULL [-Waddress]
    4 |   return &a == 0;      // -Waddress (good)
      |          ~~~^~~~


In contrast, Clang issues three warnings:

t.C:4:11: warning: comparison of address of 'a' equal to a null pointer is
      always false [-Wtautological-pointer-compare]
  return &a == 0;      // -Waddress (good)
          ^    ~
t.C:12:13: warning: comparison of address of 'a' equal to a null pointer is
      always false [-Wtautological-pointer-compare]
    return &a == 0;    // missing -Waddress
            ^    ~
t.C:31:13: warning: comparison of address of 'a' equal to a null pointer is
      always false [-Wtautological-pointer-compare]
    return &a == 0;    // missing -Waddress
            ^    ~
3 warnings generated.
Comment 1 Drea Pinski 2021-09-16 21:24:48 UTC 
GCC warns only at instantiation time.
Comment 2 Drea Pinski 2021-09-16 21:28:23 UTC 
(In reply to Andrew Pinski from comment #1)
> GCC warns only at instantiation time.

And it has been that way since at least 4.4.0.
Comment 3 Martin Sebor 2021-09-16 22:11:27 UTC 
With the correct explicit instantiation directive things look much better:

$ cat pr102378.C && gcc -S -Wall pr102378.C
int f ()
{
  int a[2];
  return &a == 0;      // -Waddress (good)
}

template <class T>
int g ()
{
  {
    int a[2];
    return &a == 0;    // missing -Waddress
  }

  {
    T t;
    return &t == 0;    // no -Waddress for type-dependent expr (good)
  }

  {
    T a[2];
    return &a == 0;    // missing -Waddress
  }
}

template <class T>
int h ()
{
  {
    int a[2];
    return &a == 0;    // -Waddress (good)
  }

  {
    T t;
    return &t == 0;    // -Waddress (good)
  }

  {
    T a[2];
    return &a == 0;    // -Waddress (good)
  }
}

template int h<int> ();
pr102378.C: In function ‘int f()’:
pr102378.C:4:13: warning: the address of ‘a’ will never be NULL [-Waddress]
    4 |   return &a == 0;      // -Waddress (good)
      |          ~~~^~~~
pr102378.C: In instantiation of ‘int h() [with T = int]’:
pr102378.C:45:22:   required from here
pr102378.C:31:15: warning: the address of ‘a’ will never be NULL [-Waddress]
   31 |     return &a == 0;    // -Waddress (good)
      |            ~~~^~~~
pr102378.C:36:15: warning: the address of ‘t’ will never be NULL [-Waddress]
   36 |     return &t == 0;    // -Waddress (good)
      |            ~~~^~~~
pr102378.C:41:15: warning: the address of ‘a’ will never be NULL [-Waddress]
   41 |     return &a == 0;    // -Waddress (good)
      |            ~~~^~~~
Comment 4 Martin Sebor 2021-09-16 22:25:51 UTC 
(In reply to Andrew Pinski from comment #2)
> (In reply to Andrew Pinski from comment #1)
> > GCC warns only at instantiation time.
> 
> And it has been that way since at least 4.4.0.

Not quite.  It depends on where a warning is implemented.  As the test case below shows, G++ does diagnose some problems in uninstantiated templates but not others.  Jason's suggestion in the review I pointed to is to try to do better and the purpose of this bug is to capture that limitation and improvement for the work I'm doing.

$ cat t.C && gcc -S -Wall -Wredundant-tags t.C
struct A { A (); ~A (); };

template <class T>
void f ()
{
  volatile struct A a;            // -Wredundant-tags (good)
  register int n = sizeof a;      // -Wregister (good)
  __builtin_memset (&a, 0, n);    // missing -Wclass-memaccess and volatile A* to A* conversion warning
}

t.C: In function ‘void f()’:
t.C:6:12: warning: redundant class-key ‘struct’ in reference to ‘struct A’ [-Wredundant-tags]
    6 |   volatile struct A a;            // -Wredundant-tags (good)
      |            ^~~~~~
      |            ------
t.C:7:16: warning: ISO C++17 does not allow ‘register’ storage class specifier [-Wregister]
    7 |   register int n = sizeof a;      // -Wregister (good)
      |                ^
Comment 5 Jakub Jelinek 2022-05-06 08:31:05 UTC 
GCC 12.1 is being released, retargeting bugs to GCC 12.2.
Comment 6 Richard Biener 2022-08-19 08:24:17 UTC 
GCC 12.2 is being released, retargeting bugs to GCC 12.3.
Comment 7 Richard Biener 2023-05-08 12:22:29 UTC 
GCC 12.3 is being released, retargeting bugs to GCC 12.4.
Comment 8 Richard Biener 2024-06-20 08:58:01 UTC 
GCC 12.4 is being released, retargeting bugs to GCC 12.5.