c++/8317: Confusing diagnostic for code that misuses conversion operators

[austern at apple dot com]

>Number:         8317
>Category:       c++
>Synopsis:       Confusing diagnostic for code that misuses conversion operators
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Oct 22 21:36:01 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     austern@apple.com
>Release:        3.1, 3.2, pre-3.3 TOT
>Organization:
>Environment:
Linux
>Description:
Consider the following code fragment:
struct X {
  char &operator[](char i);
  operator char *();
  operator const char *() const;
};

void f(X &x)
{
  x[0] += 1;
}

Obviously this is bad practice, since x[0] should be interpreted too ways.  The compiler ought to tell the programmer that he/she is doing something wrong.

The compiler does do that.  However, the diagnostic is so confusing that a user who would make this mistake is unlikely to be able to figure out what the compiler is saying.

With 3.1 and 3.2, the output is:
bash-2.05$ /usr/local/gcc-3.1/bin/g++ -c foo.cc
foo.cc: In function `void f(X&)':
foo.cc:11: choosing `char& X::operator[](char)' over `operator[]'
foo.cc:11:   because worst conversion for the former is better than worst 
   conversion for the latter
foo.cc:11: choosing `char& X::operator[](char)' over `operator[]'
foo.cc:11:   because worst conversion for the former is better than worst 
   conversion for the latter

Note that (1) the warning is repeated; and (2) it doesn't get to the heart of the matter and tell the user that it's a bad idea to provide both operator* and operator[].  A user is unlikely to be able to understand the problem from reading this message.

With 3.3 the behavior is exactly the same, except that 'warning' is replaced by 'error'.  This is not an improvement.
>How-To-Repeat:

>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: