0

My emplyer is considering upgrading to RAD11 so I have to test if it suite our needs so I am in process of porting one of mine projects (Huge win32 CAD/CAM app) from BDS2006 into RAD 11 (trial version 28.0.44500.8973) and have workaround most of the problems (related to changes in VCL and WinAPI interface related to Unicode) but with this bug I hit a wall...

For some classes and structs (not all of them) sometimes (only on certain lines not all of them) the compiler is complaining about missing constructor which is always present in code like this (see the comments marked with // ***):

//---------------------------------------------------------------------------
//--- Crypto ver: 1.24 ------------------------------------------------------
//---------------------------------------------------------------------------
#ifndef _crypto_h
#define _crypto_h
//---------------------------------------------------------------------------
#define _mod_type DWORD
//---------------------------------------------------------------------------
struct _mod_type2
    {
    _mod_type h,l;  // (high,low) or (integer,fractional)
    _mod_type2()                { h=0; l=0; }
    _mod_type2(_mod_type2& a)   { *this=a; }    // *** this is the "missing" constructor
    ~_mod_type2()               {};
    _mod_type2* operator = (const _mod_type2 *a) { *this=a; return this; }
    //_mod_type2* operator = (const _mod_type2 &a) { ...copy...; return this; }

    _mod_type2(int        a)    { h=a; l=0; }
    _mod_type2(_mod_type& a)    { h=a; l=0; }
    _mod_type2 operator = (int        a)    { h=a; l=0; return *this; }
    _mod_type2 operator = (_mod_type& a)    { h=a; l=0; return *this; }

    int operator == (_mod_type2 a) { return (h==a.h)&&(l==a.l); }
    int operator != (_mod_type2 a) { return (h!=a.h)||(l!=a.l); }
    int operator >= (_mod_type2 a) { if (h!=a.h) return (h>a.h); return (l>=a.l); }
    int operator >  (_mod_type2 a) { if (h!=a.h) return (h>a.h); return (l> a.l); }
    int operator <= (_mod_type2 a) { if (h!=a.h) return (h<a.h); return (l<=a.l); }
    int operator <  (_mod_type2 a) { if (h!=a.h) return (h<a.h); return (l< a.l); }
    };
//---------------------------------------------------------------------------
bool error(_mod_type2 a,_mod_type2 b)
    {
    bool x;
    x = (_mod_type2(a)>=_mod_type2(b));     // *** this line produce error
/*
    [bcc32 Error] crypto.h(34): E2285 Could not find a match for '_mod_type2::_mod_type2(_mod_type2&)'
  Full parser context
    win_main.cpp(5): #include crypto.h
    crypto.h(32): parsing: bool error(_mod_type2,_mod_type2)
*/
    return x;
    }
//---------------------------------------------------------------------------
#endif
//---------------------------------------------------------------------------

Its always a constructor like this:

class_name(class_name& a)   { *this=a; }

Which I have in all of my structs and classes due to compiler bug in BDS2006

If I comment out the constructor (so C++ will use default one) the bug usually disappears but not for all classes/struct ...

The simple code above produce the error (just #include it to empty C++ win32 Form App project using Classic compiler).

So the Question is:

Q1 Is anything wrong with my code?

Q2 How to workaround this?

Using new compiler seems to compile this without problems however that is not an option as the whole project is not written for it and porting it would be a huge effort taking too long time (hitting much more walls like this and without the working RAD help in trial version its not a good idea in general at this point)...

PS All the QAs in here dealing with this error I found was due to really missing constructor which is not my case ...

Spektre
  • 49,595
  • 11
  • 110
  • 380
  • You don't have an assignment operator taking `_mod_type2` – Mat May 12 '22 at 09:39
  • The one that takes a pointer is highly suspicious, I'd get rid of it. Your operator= should return a reference not a copy, and take const references not modifiable ones unless you have a very good reason to do otherwise. – Mat May 12 '22 at 09:46
  • operator= should return a reference to the object, not a pointer. You shouldn't have one that takes a pointer either, that's super strange - why would you want `foo a; foo b; a = &b;` to compile? – Mat May 12 '22 at 09:49
  • And no you don't get default generated constructors & other special member functions once you start writing your own. – Mat May 12 '22 at 09:50
  • @Mat looks like my solution worked out ... thx for your assistance ... – Spektre May 12 '22 at 12:13
  • You're claiming a **compiler** bug in BDS2006, but as you write yourself there: it's a double deletion bug in your **own** code. – MSalters May 12 '22 at 12:14
  • @MSalters the default constructor and destructors from BDS are the cause of the problem ... and not just memory it also solves some compiler problems like this one [Too many initializers error for a simple array in bcc32](https://stackoverflow.com/q/33829983/2521214) – Spektre May 12 '22 at 12:15
  • @Spektre: Why do you expect the compiler to generate a default constructor when you provided your own `S::S(S&a)`, which is a non-default constructor? It seems to me that you're blaming the compiler with little or no proof. – MSalters May 12 '22 at 12:20
  • @MSalters I dont ... but when there are none then I expect BDS do them on its own and then the problems start to appear ... – Spektre May 12 '22 at 12:22

1 Answers1

0

Looks like I found out solution. I needed just change all the constructors of type:

T(T& a)

into

T(const T& a)

Where T is struct or class name ...

After changing all the constructors in my project I got 0 compiler errors I am left with just single linker error which is a fight for another day.

Spektre
  • 49,595
  • 11
  • 110
  • 380
  • 1
    This makes sense. The latter is a normal copy constructor, which can copy any T value including temporaries. – MSalters May 12 '22 at 12:22