Re: [boost] Re: std::complex design (Was: N-Dimensional array library submission...)

From: Gabriel Dos Reis (Gabriel.Dos-Reis_at_[hidden])
Date: 2002-02-22 10:30:37

nbecker_at_[hidden] writes:

| >>>>> "Gabriel" == Gabriel Dos Reis <Gabriel.Dos-Reis_at_[hidden]> writes:
| 
|     Gabriel> nbecker_at_[hidden] writes:
|     Gabriel> | I totally agree that it would be much easier to port legacy code if
|     Gabriel> | your could say:
|     Gabriel> | 
|     Gabriel> | real(z) = x;
|     Gabriel> | 
|     Gabriel> | However, I should point out that it is easy to make a proxy class that
|     Gabriel> | does just this.
| 
|     Gabriel> Whether it is easy to make a proxy class for that doesn't suppress the
|     Gabriel> problem. In fact, there is no way one could make a modifiable lvalue out
|     Gabriel> of real(z), that was my point.
| 
| Let me make 2 points:
| 
| 1) I completely agree that the design of complex, which disallows
|    direct access to real and imag, is a mistake.  real() and imag()
|    should return a ref.  I argued this before, but I got responses to
|    the effect that someone may want to represent complex numbers as
|    mag, phase, etc.  Personally, I find this argument unconvincing.

You're right in finding that argument unconvincing because the
standard unambigously says that complex<> stores the real and
imaginary parts.

But as noted by Dave, having a modifiable lvalue is half of the
story.  The other half is layout.  C99 makes a little step further: a
_Complex has a layout compatible with that of an array of two scalars,
with the slot-0 <-> real-part and slot-1 <-> imaginary-part.

| 2) I don't understand why you say you can't make a modifiable lvalue.

You cannot say &real(z);

There is no way of saying that, using only proxy classes.

-- Gaby