www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - warning bug - bits and ints

reply =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
There is a bug with the brand new warning system in D.
Bit is treated like an integer type, which it is *not*.

Simple test case:


 bit t() { return 1; }
 bit f() { return 0; }
 
 void main()
 {
   bit b = t() && f();
 }


"implicit conversion of expression (t() && f()) of
type int to bit can cause loss of data"

This code works fine:


 void main()
 {
   bit b = true && false;
 }


(most likely because it is a compile-time expression?)


Converting any integer to a bit is *defined* as returning
0 for zero integers and 1 for non-zero integer values...

There is no "loss of data", as with going from int->short.

--anders
Mar 14 2005
next sibling parent reply Nick <Nick_member pathlink.com> writes:
In article <d13hhd$126q$1 digitaldaemon.com>,
=?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= says...

There is a bug with the brand new warning system in D.
Bit is treated like an integer type, which it is *not*.

Simple test case:


 bit t() { return 1; }
 bit f() { return 0; }
 
 void main()
 {
   bit b = t() && f();
 }




I think the problem here is that binary opperators like &&, || and ! return int,
not bit/bool. BTW, it this bug is fixed, I will start using warnings on a
permanent basis, as they have already unveiled two serious bugs in my code
(Thanks Walt!)

Nick
Mar 15 2005
parent Derek Parnell <derek psych.ward> writes:
On Tue, 15 Mar 2005 15:46:20 +0000 (UTC), Nick wrote:

[snip]

 ... I will start using warnings on a
 permanent basis, as they have already unveiled two serious bugs in my code
 (Thanks Walt!)


Agreed. I've found a few bugs because of the warnings too. Thank you Walter
for this helpful device.

-- 
Derek Parnell
Melbourne, Australia
16/03/2005 7:16:51 AM
Mar 15 2005
prev sibling next sibling parent reply "Andrew Fedoniouk" <news terrainformatica.com> writes:
One more on the subject:

int low, int high;

bool isEmpty()
{
   return low > hight;
}

Warning: "implicit conversion of expression (low > high) of type int to bit 
can cause loss of data".

Andrew.


"Anders F Björklund" <afb algonet.se> wrote in message 
news:d13hhd$126q$1 digitaldaemon.com...

 There is a bug with the brand new warning system in D.
 Bit is treated like an integer type, which it is *not*.

 Simple test case:


 bit t() { return 1; }
 bit f() { return 0; }

 void main()
 {
   bit b = t() && f();
 }


 "implicit conversion of expression (t() && f()) of
 type int to bit can cause loss of data"

 This code works fine:


 void main()
 {
   bit b = true && false;
 }


 (most likely because it is a compile-time expression?)


 Converting any integer to a bit is *defined* as returning
 0 for zero integers and 1 for non-zero integer values...

 There is no "loss of data", as with going from int->short.

 --anders
Mar 15 2005
parent reply Derek Parnell <derek psych.ward> writes:
On Tue, 15 Mar 2005 14:12:30 -0800, Andrew Fedoniouk wrote:


 One more on the subject:
 
 int low, int high;
 
 bool isEmpty()
 {
    return low > hight;
 }
 
 Warning: "implicit conversion of expression (low > high) of type int to bit 
 can cause loss of data".
 


This is not a bug.  The expression (low > high) results in an integer value
and it is possible to lose data when converting an integer to a bit (bool).

Now maybe such expressions should result in a true boolean value, but D
does not support true boolean values, only pseudo-boolean values called
bits ( a one-bit integer ).

Consequently, you cannot take the syntax shortcuts one would code in C/C++,
and in D such expressions as above need to be spelled out in full ...

  return low > high : true ? false;


-- 
Derek
Melbourne, Australia
16/03/2005 9:27:15 AM
Mar 15 2005
next sibling parent reply "Nick Sabalausky" <z a.a> writes:
I suppose this has probably been discussed already, but does it really make 
sense for the result of a comparison to be an integer?

"Derek Parnell" <derek psych.ward> wrote in message 
news:1abo3r0w4p3ct.1eds1mg8mobpz$.dlg 40tude.net...

 On Tue, 15 Mar 2005 14:12:30 -0800, Andrew Fedoniouk wrote:


 One more on the subject:

 int low, int high;

 bool isEmpty()
 {
    return low > hight;
 }

 Warning: "implicit conversion of expression (low > high) of type int to 
 bit
 can cause loss of data".


 This is not a bug.  The expression (low > high) results in an integer 
 value
 and it is possible to lose data when converting an integer to a bit 
 (bool).

 Now maybe such expressions should result in a true boolean value, but D
 does not support true boolean values, only pseudo-boolean values called
 bits ( a one-bit integer ).

 Consequently, you cannot take the syntax shortcuts one would code in 
 C/C++,
 and in D such expressions as above need to be spelled out in full ...

  return low > high : true ? false;


 -- 
 Derek
 Melbourne, Australia
 16/03/2005 9:27:15 AM
Mar 15 2005
parent reply Derek Parnell <derek psych.ward> writes:
On Tue, 15 Mar 2005 21:49:49 -0500, Nick Sabalausky wrote:


 "Derek Parnell" <derek psych.ward> wrote in message 
 news:1abo3r0w4p3ct.1eds1mg8mobpz$.dlg 40tude.net...

 On Tue, 15 Mar 2005 14:12:30 -0800, Andrew Fedoniouk wrote:


 One more on the subject:

 int low, int high;

 bool isEmpty()
 {
    return low > hight;
 }

 Warning: "implicit conversion of expression (low > high) of type int to 
 bit
 can cause loss of data".


 This is not a bug.  The expression (low > high) results in an integer 
 value
 and it is possible to lose data when converting an integer to a bit 
 (bool).

 Now maybe such expressions should result in a true boolean value, but D
 does not support true boolean values, only pseudo-boolean values called
 bits ( a one-bit integer ).

 Consequently, you cannot take the syntax shortcuts one would code in 
 C/C++,
 and in D such expressions as above need to be spelled out in full ...

  return low > high : true ? false;


 I suppose this has probably been discussed already, but does it really make 
 sense for the result of a comparison to be an integer?
 


Not to me. I think its plain daft. But I believe DMD does it this way
because it tries to be compatible with (legacy) C/C++ code.

-- 
Derek
Melbourne, Australia
16/03/2005 1:58:09 PM
Mar 15 2005
parent "Unknown W. Brackets" <unknown simplemachines.org> writes:
More importantly, it does it this way for reasons of processor 
optimization, to my understanding...

-[Unknown]



 On Tue, 15 Mar 2005 21:49:49 -0500, Nick Sabalausky wrote:
 
 

"Derek Parnell" <derek psych.ward> wrote in message 
news:1abo3r0w4p3ct.1eds1mg8mobpz$.dlg 40tude.net...


On Tue, 15 Mar 2005 14:12:30 -0800, Andrew Fedoniouk wrote:



One more on the subject:

int low, int high;

bool isEmpty()
{
   return low > hight;
}

Warning: "implicit conversion of expression (low > high) of type int to 
bit
can cause loss of data".


This is not a bug.  The expression (low > high) results in an integer 
value
and it is possible to lose data when converting an integer to a bit 
(bool).

Now maybe such expressions should result in a true boolean value, but D
does not support true boolean values, only pseudo-boolean values called
bits ( a one-bit integer ).

Consequently, you cannot take the syntax shortcuts one would code in 
C/C++,
and in D such expressions as above need to be spelled out in full ...

 return low > high : true ? false;


I suppose this has probably been discussed already, but does it really make 
sense for the result of a comparison to be an integer?


 
 Not to me. I think its plain daft. But I believe DMD does it this way
 because it tries to be compatible with (legacy) C/C++ code.
Mar 15 2005
prev sibling parent =?ISO-8859-1?Q?Anders_F_Bj=F6rklund?= <afb algonet.se> writes:
Derek Parnell wrote:


 This is not a bug.  The expression (low > high) results in an integer value
 and it is possible to lose data when converting an integer to a bit (bool).


As long as D uses pseudo-boolean casts to convert integers and pointers 
into bits, there is no possibility for "loss of data" when doing it...

The result is the same as : (integer != 0) or (pointer != null)


 Now maybe such expressions should result in a true boolean value, but D
 does not support true boolean values, only pseudo-boolean values called
 bits ( a one-bit integer ).


Well, bits are *not* 1-bit integers, but they sure are pseudo-booleans.

If you convert for instance the integer 256 into a byte, you get 0.
But if you convert the integer 2 into a bit, you get 1 (i.e. 'true')

--anders

PS.
It does not help that the spec says it *should* return a "bool":

 The result type of an OrOr expression is bool,
 The result type of an AndAnd expression is bool, 
 Equality expressions ... The type of the result is bool.
 The is compares for ... The type of the result is bool.
 The result type of a relational expression is bool.


http://www.digitalmars.com/d/expression.html
Mar 16 2005
prev sibling parent "Ben Hinkle" <ben.hinkle gmail.com> writes:
Walter, can we get warnings that don't cause the compiler to error? I can't 
use -w for anything due to this int/bit conversion warning. I could change 
my code to avoid bit and pass ints around so that I can use the comparison 
operators with causing the compiler to freak out, but that seems like the 
wrong way to go. What is the recommended way to deal with -w and int/bit?

"Anders F Björklund" <afb algonet.se> wrote in message 
news:d13hhd$126q$1 digitaldaemon.com...

 There is a bug with the brand new warning system in D.
 Bit is treated like an integer type, which it is *not*.

 Simple test case:


 bit t() { return 1; }
 bit f() { return 0; }

 void main()
 {
   bit b = t() && f();
 }


 "implicit conversion of expression (t() && f()) of
 type int to bit can cause loss of data"

 This code works fine:


 void main()
 {
   bit b = true && false;
 }


 (most likely because it is a compile-time expression?)


 Converting any integer to a bit is *defined* as returning
 0 for zero integers and 1 for non-zero integer values...

 There is no "loss of data", as with going from int->short.

 --anders
Mar 24 2005