Hello,

Who knows more about default parameters? I do not know the wording of the
current standard, nor do I know of any considerations of the committee about
default-parameters. I mainly got my knowledge by experimenting with gcc
version 2.4.3. Please correct me if I got something wrong.

( I ) 'x=exp'
( II ) 'x=X(exp)'
 HOW TO CALL CONSTRUCTORS WITH MORE THAN 1 ARGUMENT /
 DEFAULT-CONSTRUCTORS (= CTOR WITH 0 ARGUMENTS) ?
( III ) 'x=exp' 'x=X(exp)' and 'x(exp)'
( IV ) 'x=(exp)' GCC-BUG ?

--------------------------------------------------------------------
( I ) 'a=exp'

It is possible to supply (from right to left) default-parameters for
function-calls. Any expression can be given as default-param, it
is evaluated every time the function is called.
 struct X { int     v;
   X(int i)    { v=i;   printf("called X(%d)\n",i); }
   X(X& x)     { v=x.v; printf("called X(x.%d)\n",x.v); }
 } xx (-1);
 char CC=3;

 void f( X a=printf("using default args for g\n")
       , X b=2+4*5, X c=CC, X d=xx);

 >called X(-1)
 >using default args for g
 >called X(25)
 >called X(22)
 >called X(3)
 >called X(x.-1)
--------------------------------------------------------------------
( II ) 'x=X(exp)'
 HOW TO CALL CONSTRUCTORS WITH MORE THAN 1 ARGUMENT /
 DEFAULT-CONSTRUCTORS (= CTOR WITH 0 ARGUMENTS) ?

I found out that 'X x=X()' can be used to supply the default-ctor as
default-param and 'XY xy=XY(1,2)' can be used to supply ctors with 2 args.

When I look at it my first thought is:
"a X/XY is contructed and than this is used as   argument to copy-ctor-call
 to contruct x/xy"
But I am wrong, x/xy are constructed directly. 'X x=X();' 'X x=X(1);'
'X x=X(1,2);' etc are idioms I can also use in var-declarations, being the
same as 'X x;' 'X x(1)' 'X x(1,2)'
So 'f(X x=1)' is the same as 'f(X x=X(1))'.

My question: Is this official syntax?

--------------------------------------------------------------------
( III ) 'x=exp' 'x=X(exp)' and 'x(exp)'

I thought 'X x(1)' is always the C++ form of C's form ('X x=1') of the
constructor-call with 1 arg.
If so,  'f(X x(1))' (*) should be the same as
        'f(X x=1)'
    and 'f(X x=X(1))'
and also 'f(X x(1,2))'  (**) should be the same as
         'f(X x=X(1,2)'
Well, there seems to be no solution how to express 'f(X x=X())' this way.
However gcc does not accept (*) and (**) (see IV).

Have (*) and (**) been discussed by the committee? Are they allowed?

--------------------------------------------------------------------
( IV ) 'x(exp)' GCC-BUG ?

When I tried to use 'X x(exp)' instead of 'X x=exp' with gcc:

 void f( X a(printf("using default args for g\n"))
       , X b(2+4*5), X c(CC), X d(xx));

then a, b, c and d are constructed before main() like static vars of f().
And if I try to call 'f( xx );' or 'f( X(1) );' gcc flags an error:
 >too many arguments to function `void  f ()'

And gcc believes that
 void f( X a=printf("using default args for g\n")
     ^
       , X b(2+4*5), X c(CC), X d=xx);
            ^     ^     ^  ^     ^
has the type:
 `void  f (struct X (= ( printf ((const char *)((char *)
     "using default args for g\012"))))
  , struct X (=  xx ))'

I suppose this is a bug. If have no idea how gcc could come to this
result in a correct way. Or have I missed some new C++ features?

Thanks for your help,

Ulf Schuenemann

--------------------------------------------------------------------
Ulf Sch   nemann
Institut f   r Informatik, Technische Universit   t M   nchen.
email: schuenem@informatik.tu-muenchen.de
WWW:   http://hphalle2/~schuenem  (currently not available from outside)