Fix a crash in String::changeBuffer()

[raheelh]

Calling String::reserve(0) causes a crash if String object is in invalidated state. Per the comment on the method's declaration in WString.h, reserve(0) would validate the string object but in fact it just causes a crash instead of recovering from invalidated strings.

This change fixes the edge case bug in String::changeBuffer() which is the root cause of the crash exposed from String::reserve().

Following test code was used to reproduce the problem and also to validate the fix:

String result;
while(true){ // suppose this is a loop which reads data from a web page.
  char c = 'A'; // suppose this is a byte read from client.
  result += c; // the loop will eventually cause malloc() to fail here
  if (result.c_str()==0) // even if we detect this
  {
    Serial.println("String INVALIDATED!!!!!");
    // attempt to validate string object so we can use it.
    result.reserve(0);   // before fix, this would crash.
    // if reserve(0) worked, the object is validated, i.e. available for use again.
    Serial.println("Trying to empty...."); // try to make use of object to see.
    result=""; 
    Serial.println("Emptied!!!!"); // works!
    break;
  } 
}

[igrr]

umm_realloc will attempt to extend the buffer if it is possible, i.e. the adjacent block is not allocated.
You can check umm_malloc source code (it is included in this repository).

[raheelh]

Makes sense. I just created a pull request with a change to leverage realloc() in this method.