What is std::string::c_str() lifetime?

问题:

In one of my programs, I have to interface with some legacy code that works with const char*.

Lets say I have a structure which looks like:

struct Foo
{
  const char* server;
  const char* name;
};

My higher-level application only deals with std::string, so I thought of using std::string::c_str() to get back const char* pointers.

But what is the lifetime of c_str() ?

Can I do something like this without facing undefined behavior ?

{
  std::string server = \"my_server\";
  std::string name = \"my_name\";

  Foo foo;
  foo.server = server.c_str();
  foo.name = name.c_str();

  // We use foo
  use_foo(foo);

  // Foo is about to be destroyed, before name and server
}

Or am I supposed to immediately copy the result of c_str() to another place ?

Thank you.

回答1:

The c_str() result becomes invalid if the std::string is destroyed or if a non-const member function of the string is called. So, usually you will want to make a copy of it if you need to keep it around.

In the case of your example, it appears that the results of c_str() are used safely, because the strings are not modified while in that scope. (However, we don\'t know what use_foo() or ~Foo() might be doing with those values; if they copy the strings elsewhere, then they should do a true copy, and not just copy the char pointers.)

回答2:

Technically your code is fine.

BUT you have written in such a way that makes it easy to break for somebody that does not know the code. For c_str() the only safe usage is when you pass it as a parameter to a function. Otherwise you open yourself up-to maintenance problems.

Example 1:

{
  std::string server = \"my_server\";
  std::string name   = \"my_name\";

  Foo foo;
  foo.server = server.c_str();
  foo.name = name.c_str();

  //
  // Imagine this is a long function
  // Now a maintainer can easily come along and see name and server
  // and would never expect that these values need to be maintained as
  // const values so why not re-use them

  name += \"Martin\";
  // Oops now its broken.

  // We use foo
  use_foo(foo);

  // Foo is about to be destroyed, before name and server
}

So for maintenance make it obvious:

Better solution:

{
  // Now they can\'t be changed.
  std::string const server = \"my_server\";
  std::string const name   = \"my_name\";

  Foo foo;
  foo.server = server.c_str();
  foo.name = name.c_str();

  use_foo(foo);    
}

But if you have const strings you don\'t actually need them:

{
  char const* server = \"my_server\";
  char const* name   = \"my_name\";

  Foo foo;
  foo.server = server;
  foo.name   = name;

  use_foo(foo);
}

OK. For some reason you want them as strings:
Why not use them only in the call:

{
  std::string server = \"my_server\";
  std::string name = \"my_name\";

  // guaranteed not to be modified now!!!     
  use_foo(Foo(server.c_str(), name.c_str());
}

回答3:

It is valid until one of the following happens to the corresponding string object:

• the object is destroyed
• the object is modified

You\'re fine with your code unless you modify those string objects after c_str()s are copied into foo but before use_foo() is called.

回答4:

Return value of c_str() is valid only until the next call of a nonconstant member function for the same string

回答5:

The const char* returned from c_str() is only valid until the next non-const call to the std::string object. In this case you\'re fine because your std::string is still in scope for the lifetime of Foo and you aren\'t doing any other operations that would change the string while using foo.

回答6:

As long as the string isn\'t destroyed or modified, using c_str() is OK. If the string is modified using a previously returned c_str() is implementation defined.

回答7:

For completeness, here\'s a reference and quotation from cppreference.com:

The pointer obtained from c_str() may be invalidated by:

• Passing a non-const reference to the string to any standard library function, or
• Calling non-const member functions on the string, excluding operator[], at(), front(), back(), begin(), rbegin(), end() and rend().