Balog Pal wrote in a message to Wim Veldhuis:
 BM> #define WHEEL_MAX 13
 BM> #define STARTING_CASH 300
 TM> const unsigned int WHEEL_MAX = 13;
 TM> const unsigned int STARTING_CASH = 300;
 WV> This is not entirely correct. Compilers usually allocate memory for
 WV> something declared as const ...
 BP> What compilers do that?  I tried a plenty (MS, Borland, Sym,
 BP> Watcom, SUN, HP) and didn't notice that.  They allocate
 BP> memory if you comile in C (!) mode but not in C++.  Unless
 BP> you happen to reference the address of the const that is not
 BP> very likely for an int constant as above.
If you do it with ints, there is not a problem, if you do it with double or 
another more complex type the problem does arise.
If you place a const double inch  2.54; in a header file and include the 
header file in a module which does not use the const, you might get a warning 
like I said. I had this problem when using Borland C++ 4.5x quite some time 
ago. The "reasoning" of the compiler probably is to allocate it just once in 
memory as a read-only variable, since it is difficult to load it into 
processor registers directly and the code for transfering a double into 
registers is already present when it is located somewhere in memory.
The compiler however generates a warning about an identifier not being used 
if you include the headerfile in a module which does not use the identifier.
 WV> This can result in warnings if you do
 WV> this inside a header file and you compile another module which uses
 WV> the headerfile but doesn't use the constant.
 BP> I have a plenty of such const int's in headers and have no
 BP> warnings. Hm? 
I omitted to say that this is usually a problem with other data types than 
int. Int is mostly used for consts, so this is indeed usually not a problem.
 
 WV> In segmented programs there is also a gotscha's if you put initialised
 WV> data in a non default segment for some reason. At least with the
 WV> borland compiler this results in errors as it ALWAYS assumes const
 WV> variables to be in the default data segment !
 BP> Which version and path level? Probably a serious bug.
If I recall correctly it was with Borland C++ v4.52. Target model is large, 
and windows.
I agree it is a bug, but Borland will probably call it a feature :-(
I haven't tested this with v5.x since I only used it for 32-bit applications 
so far, which do not have the 64K barrier.
PS. My excuse for my late response, I am very busy at the moment and do not 
regularly check my fido mail. You can reach me also at the internet at 
veldhuis@xs4all.nl 
mvg/wr
  
--- timEd/2 1.01.g3+
---------------