default constructor for int

Question

Possible Duplicate:
Why is it an error to use an empty set of brackets to call a constructor with no arguments?

In an answer to this question it's said that

ints are default-constructed as 0, as if you initialized them with int(). Other primitive types are initialized similarly (e.g., double(), long(), bool(), etc.).

Just while I was explaining this to a colleague of mine I made up the following code, compiled (gcc-4.3.4) and ran, and observed unexpected behavior.

#include <iostream>

int main() {
  int i(); 
  std::cout << i << std::endl; // output is 1
}

Why is the output 1 and not 0 ?

This is explained in: [Why is it an error to use an empty set of brackets to call a constructor with no arguments?](http://stackoverflow.com/questions/180172/why-is-it-an-error-to-use-an-empty-set-of-brackets-to-call-a-constructor-with-no) — R. Martinho Fernandes, Jun 20 '12 at 10:20
I'm surprised it even compiles, as it looks like you stumbled upon the [most vexing parse](http://en.wikipedia.org/wiki/Most_vexing_parse). — Some programmer dude, Jun 20 '12 at 10:21
Congratulations on discovering [the most vexing parse](http://en.wikipedia.org/wiki/Most_vexing_parse). — Sergey Kalinichenko, Jun 20 '12 at 10:21

score 36 · Accepted Answer · edited Jun 20 '12 at 11:00

36

Most vexing parse comes into play here. You're actually declaring a function i, not an int variable. It shouldn't even compile (unless you actually have a function i defined somewhere... do you?).

To value-initialize the int, you need:

int i = int();

edited Jun 20 '12 at 11:00

David Rodríguez - dribeas

198,982
21
284
478

answered Jun 20 '12 at 10:20

Luchian Grigore

245,575
61
446
609

1

Implicit cast to `bool`, I think. – aschepler Jun 20 '12 at 10:21
It must be bug in previous versions of GCC; in 4.7, it doesn't compile. – Griwes Jun 20 '12 at 10:21
@Griwes maybe he defines the function somewhere and is un-aware of it. – Luchian Grigore Jun 20 '12 at 10:22
@LuchianGrigore, yet, on Ideone's GCC 4.3.4, it compiles; same for its 4.5.1 in C++11 mode. – Griwes Jun 20 '12 at 10:23
g++ 4.5.1 gives a warning: `Warnung: Die Adresse von »int i()« wird immer zu »wahr« auswerten`, but compiles it fine. – Daniel Fischer Jun 20 '12 at 10:25
Also see source code and output here: http://ideone.com/8yB1R – moooeeeep Jun 20 '12 at 10:27
1

@moooeeeep yes, but the important part here is that `i` is not an `int` - see http://ideone.com/f7bEP – Luchian Grigore Jun 20 '12 at 10:29
It would be nice to see the assembler (can't compile it myself now) – Antonio Pérez Jun 20 '12 at 10:41
@AntonioPérez It would indeed, but VS2005 doesn't accept the code :) – Luchian Grigore Jun 20 '12 at 10:45
10

@AntonioPérez I would not expect the assembly to be of any interest, but rather dumb. The equivalent of `std::cout< – David Rodríguez - dribeas Jun 20 '12 at 11:04
3

@DavidRodríguez-dribeas And continuing: the function is "used" if its address is taken, so the absence of a definition somewhere is undefined behavior. Typically, however... if the compiler optimizes the "use" out (since it knows that the address is non-null), the code will compile and run; if it doesn't, you'll get an undefined symbol when linking. – James Kanze Jun 20 '12 at 11:18
Most ODR violations (including this one) are "undefined behaviour, no diagnostic required". As James suggests, it would make the optimizer's job considerably harder if it were required to generate an error here. – M.M Jun 02 '15 at 23:22

default constructor for int

1 Answers1

Linked

Related