In the ArduinoUnit unit testing library I have provided a mechanism for giving a TestSuite a name. A user of the library can write the following:
TestSuite suite("my test suite");
// ...
suite.run(); // Suite name is used here
This is the expected usage - the name of the TestSuite is a string literal. However to prevent hard-to-find bugs I feel obliged to cater for different usages, for example:
char* name = (char*) malloc(14);
strcpy(name, "my test suite");
TestSuite suite(name);
free(name);
// ...
suite.run(); // Suite name is used here
As such I have implemented TestSuite like this:
class TestSuite {
public:
TestSuite(const char* name) {
name_ = (char*) malloc(strlen(name) + 1);
strcpy(name_, name);
}
~TestSuite() {
free(name_);
}
private:
char* name_;
};
Putting aside the issue of failing to deal with memory allocation failures in the constructor I'd prefer to simply allocate the pointer to a member variable like this:
class TestSuite {
public:
TestSuite(const char* name) : name_(name) {
}
private:
const char* name_;
};
Is there any way I can change the interface to force it to be used 'correctly' so that I can do away with the dynamic memory allocation?
What if you provide two overloaded constructors?
TestSuite(const char* name) ...
TestSuite(char* name) ...
If called with a const char*
, then the constructor could make a copy of the pointer, assuming that the string will not go away. If called with a char*
, the constructor could make a copy of the whole string.
Note that it is still possible to subvert this mechanism by passing a const char*
to the constructor when the name
is in fact dynamically allocated. However, this may be sufficient for your purposes.
I should note that I have never actually seen this technique used in an API, it was just a thought that occurred to me as I was reading your question.