NetNews Offline 2/NetNews Offline Volume 2.iso/news/comp/lang/c++-part2/17741

Internet Message Format | 1996-08-05 | 3.8 KB

Path: navet.enator.se!usenet
From: Hardy Henneberg <[e-mail]>
Newsgroups: comp.lang.c++,fj.lang.c++
Subject: Re: [Q]When temporary object be created?
Date: 17 Apr 1996 09:07:16 GMT
Organization: Enator AB
Message-ID: <4l2cc4$[e-mail]>
NNTP-Posting-Host: 193.162.31.53

[e-mail] (Hiroshi-MATSUSHITA) writes:
> When a temporary object will be created?
>
> Suppose, I create a class named PERSON that takes a name
> in its constructor like following:
>
>   class PERSON {
>     public:
>       PERSON(const STRING& nm);
>       ...
>     private:
>       STRING name;
>   };
>
> Then, when I create an instance of PERSON using raw string,
> I may write like this:
>
>   PERSON nomo("Hideo Nomo");
>
> STRING class, in this case, is not a smart pointer, but it
> creates a character buffer newly using 'new' operator in its
> constructor and delete it in its destructor.
>
> In this case, will a temporary string instance be created?

Yes, Your PERSON constructor takes a STRING as parameter.
You initialize  a PERSON object with a const char*, so the compiler
to find a PERSON constructor with a const char* as parameter. Because
that doesn't exits, it will use the rules for function matching to find
a constructor which matches the call. One of these rules says, there
is a match if the const char* can be typeconverted into a STRING with a
typeconverting function. I guess the STRING class must have a typecast
constructor (a constructor with 1 parameter), with a const char* parameter.
So, this string constructor must be called resulting in a temporary STRING
object.

> Does it depend on how to create the constructor?
I dont' understand the question.

> Does it depend on a compiler? (undefined in C++ spec.?)
No.

>
> If I initialize the private member 'name' using initializing
> list like following, can I avoid to create a temporary object?
>
>   PERSON::PERSON(const STRING& nm) : name(nm) {...}

No, the implementation of the PERSON constructor and the initialisation
of the PERSON object will normally be in two seperate compilated modules
with no possibility for the compiler to optimize.

>
> Why I want to know this is because I don't want to take extra time
> and memory space even if it's temporary.
> I'm pazzling my brain which I'd better to take between reference
> approach described above and pointer approach like following:
>
>   class PERSON {
>     public:
>       PERSON(const char* nm);
>       ...
>     private:
>       STRING name;
>   };
>
> In this case (pointer approach), *no* temporary object will be
> created. Also, this constructor can accept STRING object, because
> STRING class has a conversion operator to "const char*", and it
> works implicitly. (Does this feature bring another argue?)
>
This is a possible solution, and you can even have both constructors.
Then the function matching rules will tell the compiler to use
the constructor with a const char* in your case, because it's
an exact match.

But remember, that when you define a new constructor with 1 parameter you
also make a new hole in the type control, which may let errors pass
the compiler.

If you are worried about memory space a solution could be
to use an intelligent STRING class, which only uses one buffer for
all strings with the same content.

If intialising PERSON with a const char* really is a performance issue,
I'm sure there is also another solution for that.

I think using char* all over in a program should be avoided - it's much
better to use the STRING class.

I'm also against misusing type cast constructors.

My experience from working 8 years in C++ is that char* and type
cast constructors cause a lot of problems.

cheers,
Hardy Henneberg
ENATOR, Denmark

>
> (If this is a FAQ, please forgive me. Then I appreciate
> if you let me know.)
>
> Thank you,
>
> --Hiroshi Matsushita @ Seiko Epson Corp. ([e-mail])
>