|
|
|
Home » U++ Library support » U++ Core » The problem with 'Null'
| The problem with 'Null' [message #20446] |
Thu, 19 March 2009 08:04  |
gridem
Messages: 45
Registered: August 2008
|
Member |
|
|
I found that Upp uses the following practice: instead of creating already prepared object it creates the object with default constructor and than fill the necessary members in later calls. But using such approach the programmer should distinguish between init and non init state. One of the possible solution: apply 'Null' to the fields and than check by using IsNull.
The simple types (int, long etc) and Value already have such posibility. String also can use this but using another functionality:
return 'super' empty string that may be treated as 'Null'. The problem is that and IsNull(String::GetVoid())
the solution exist: return String::GetVoid() on error. But what I can do when I must return Vector:
Vector<Templates> GetTemplateList()
Empty list denotes the there are no templates. But how can I return error without using terrible One<Vector<T> >?
|
|
|
|
|
|
| Re: The problem with 'Null' [message #20450 is a reply to message #20446] |
Thu, 19 March 2009 08:45  |
cbpporter
Messages: 1401
Registered: September 2007
|
Ultimate Contributor |
|
|
| gridem wrote on Thu, 19 March 2009 09:04 |
But using such approach the programmer should distinguish between init and non init state. One of the possible solution: apply 'Null' to the fields and than check by using IsNull.
|
I think that with U++ design there is no such thing as init and non-init state. Pretty much all classes are in "init" state after the call of a constructor, even the default one. Such objects are initialized and ready to use, but generally hold no extra information. A String() will have zero length, a Vector() will have zero elements, a Button() will have no picture, text and other properties that diverge from default, but otherwise is ready to be inserted into a parent.
So basically the default constructor creates object in an already prepared state. I think the reason that we use default constructor in this way rather than constructor with parameters to set all the meaningful data is largely related to the intrinsic construction rules of C++. Since we don't put everything on the heap, the objects construction can't be deferred to the allocation moment and is constructed at the runtime point equivalent to the declaration. At this point you often don't have all the information required to call a constructor with parameters.
There are of course exceptions. Things like Size and Point will not be initialized by default construction and can result in bugs if the user is not aware of this fact. Here denoting the absence of an initialization would be useful, but I think these classes are supposed to be lightweight and this is why they are not initialized. Adding unfertilized state detection would make them more heavy weight and less efficient. Having a vector of 10 points would initialize all of them to default, and in most cases, you would reinitialize them in code again. By leaving them uninitialized you only get the meaningful initialization, but you risk bugs if you forget it.
As for the One<Vector <T> > solution, I think is not a very good one. Not only do you have to test if the Vector is not Null and loose the guarantee that Vector is initialized, but you also increase the level of indirection. Such tricks might look like something good, but my experience tells me that if you are going to write programs with hundreds of thousands of line of code and use less straight forward solutions, you'll end up causing more problems than you solve.
Wouldn't a call to IsEmpty or something equivalent on you return value solve the problem? Maybe I'm not understanding exactly what you are trying to achieve. Maybe if you can give a real world example where U++ initialization scheme is not working out and you need to determine initialization status, then I could offer a better solution.
|
|
|
|
|
|
|
|
|
|
| Re: The problem with 'Null' [message #20469 is a reply to message #20467] |
Fri, 20 March 2009 09:52 |
 |
mirek
Messages: 13975
Registered: November 2005
|
Ultimate Member |
|
|
| gridem wrote on Fri, 20 March 2009 03:16 |
Thank you very much for detailed answers! It's very useful for me to understanding intrinsics of Upp.
But let's me to understand my opinion. I think that Null approach is not narrow-minded but generic. It uses for Value, it uses for many simple types.
Because I use a lot the templates, overloaded functions and code generation I have to utilize the generic methods for every types that I want to use. And I use Null as some kind of parameter state that I can treat as:
1. Init/Non init
2. Error on function return.
3. Default values to call the function to distinguish it from nondefault values:
void some_fun(int a, int b = Null, bool c = Null, ...)
{
if (b == Null)
b = some_complex_calculated_value(a);
...
}
etc
|
Two notes:
You cannot really define Null for bool, as it has only 2 values..
It is more correct to use "IsNull" instead of "== Null".
| Quote: |
So to see this approach for String's and Vector's I have to do workaround. May be I should not use the generics but another mechanism?
|
Well, I can see where you are heading, but I do not really like that path 
My only apology at this moment is that U++ is "practice driven", and in the whole history (which now spans abuot 10 years), we never missing IsNull for containers...
BTW, as you have noticed, there is the small issue with String Null - empty string is considered Null.
I agree this is sort of controversial decision. Indeed, a couple of years ago, we identified it as mistake and tried
IsNull(String()) == false
IsNull(String(Null)) == true
String(Null) == String()
variant. Well, what happened is that in practice, this was found to be rather unfortunate. I guess the primary problem is that it is very convenient and natural, when working with databases and GUI, that all empty String gui fields are inserted as Nulls. With above, you would need to have additional GUI buttons to say whether the field is empty or whether it is null.
Similar issues can be found across the code. That is why we went back to
IsNull(String()) == true
As a sidenote, this equivalence was originally taken from Oracle.
Mirek
|
|
|
|
| Re: The problem with 'Null' [message #20524 is a reply to message #20469] |
Sun, 22 March 2009 09:15 |
gridem
Messages: 45
Registered: August 2008
|
Member |
|
|
| luzr wrote on Fri, 20 March 2009 11:52 |
Two notes:
You cannot really define Null for bool, as it has only 2 values..
It is more correct to use "IsNull" instead of "== Null".
|
Oh, my mistake: IsNull of course instead of == Null.
| luzr wrote on Fri, 20 March 2009 11:52 |
Well, I can see where you are heading, but I do not really like that path
My only apology at this moment is that U++ is "practice driven", and in the whole history (which now spans abuot 10 years), we never missing IsNull for containers...
BTW, as you have noticed, there is the small issue with String Null - empty string is considered Null.
I agree this is sort of controversial decision. Indeed, a couple of years ago, we identified it as mistake and tried
IsNull(String()) == false
IsNull(String(Null)) == true
String(Null) == String()
variant. Well, what happened is that in practice, this was found to be rather unfortunate. I guess the primary problem is that it is very convenient and natural, when working with databases and GUI, that all empty String gui fields are inserted as Nulls. With above, you would need to have additional GUI buttons to say whether the field is empty or whether it is null.
Similar issues can be found across the code. That is why we went back to
IsNull(String()) == true
As a sidenote, this equivalence was originally taken from Oracle.
Mirek
|
Thank you for your explanations. I caught the main idea. So for SQL programming it's possible to introduce 'local IsNull':
template<typename T>
bool IsNullSql(const T& t)
{
return IsNull(t);
}
bool IsNullSql(const String& s)
{
return s.IsEmpty();
}
|
|
|
|
|
|
|
|
|
|
Goto Forum:
Current Time: Mon Apr 29 15:37:05 CEST 2024
Total time taken to generate the page: 0.03283 seconds
|
|
|
Members
Pages
Help
Register
Login
Home
