www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - is it bug?

reply Alexandr Druzhinin <drug2004 bk.ru> writes:
on win7 64 bits using dmd 2.062 32 bits this code http://dpaste.dzfl.pl 
/6cca43b5 failed with assert failure: 
core.exception.AssertError std.range(5288): Assertion failure
but on ubuntu 12.04 64 bits, dmd 2.062 64 bit it compiles and works 
rather well. May be I missed something important with it?
Apr 04 2013
parent reply =?UTF-8?B?QWxpIMOHZWhyZWxp?= <acehreli yahoo.com> writes:
On 04/04/2013 06:27 AM, Alexandr Druzhinin wrote:

 on win7 64 bits using dmd 2.062 32 bits this code http://dpaste.dzfl.pl
 /6cca43b5 failed with assert failure:
 core.exception.AssertError std.range(5288): Assertion failure
 but on ubuntu 12.04 64 bits, dmd 2.062 64 bit it compiles and works
 rather well. May be I missed something important with it?


The problem is reproducible on Linux as well if you compile with -m32.

The following assert fails in range.d:

   assert(start + count * step >= end);

start: 0
count: 83
step: 12.0386
end: 999.2

Even though count * step is seemingly 999.2 as well, due to the inexact 
nature of floating point calculations the assertion is false.

The problem seems to be in your code because your step calculation 
ignores a remainder. This works:

     // OLD: step = width / factor;
     step = (width  + factor - 1) / factor;

Ali
Apr 04 2013
parent reply Alexandr Druzhinin <drug2004 bk.ru> writes:
04.04.2013 21:24, Ali Çehreli пишет:

 On 04/04/2013 06:27 AM, Alexandr Druzhinin wrote:

 on win7 64 bits using dmd 2.062 32 bits this code http://dpaste.dzfl.pl
 /6cca43b5 failed with assert failure:
 core.exception.AssertError std.range(5288): Assertion failure
 but on ubuntu 12.04 64 bits, dmd 2.062 64 bit it compiles and works
 rather well. May be I missed something important with it?


 The problem is reproducible on Linux as well if you compile with -m32.

 The following assert fails in range.d:

    assert(start + count * step >= end);

 start: 0
 count: 83
 step: 12.0386
 end: 999.2

 Even though count * step is seemingly 999.2 as well, due to the inexact
 nature of floating point calculations the assertion is false.

 The problem seems to be in your code because your step calculation
 ignores a remainder. This works:

      // OLD: step = width / factor;
      step = (width  + factor - 1) / factor;

 Ali


that works, thank you very much! but provide some link to some good 
article that explains my mistake, I didn't realize it clearly
Apr 04 2013
parent reply =?UTF-8?B?QWxpIMOHZWhyZWxp?= <acehreli yahoo.com> writes:
On 04/04/2013 08:45 AM, Alexandr Druzhinin wrote:

 04.04.2013 21:24, Ali Çehreli пишет:



 The problem seems to be in your code because your step calculation
 ignores a remainder. This works:

      // OLD: step = width / factor;
      step = (width  + factor - 1) / factor;

 Ali


 that works, thank you very much! but provide some link to some good
 article that explains my mistake, I didn't realize it clearly


I was wrong. What I found is just a workaround. Created a bug for iota:

   http://d.puremagic.com/issues/show_bug.cgi?id=9877

The following program fails when compiled with -m32:

import std.range;

void main()
{
     float st = 0.000000000000000;
     float step = 12.038554191589355;
     float en = 999.200012207031250;

     iota(st, en, step);
}

Ali
Apr 04 2013
next sibling parent Alexandr Druzhinin <drug2004 bk.ru> writes:
05.04.2013 0:26, Ali Çehreli пишет:

 I was wrong. What I found is just a workaround. Created a bug for iota:

    http://d.puremagic.com/issues/show_bug.cgi?id=9877

 The following program fails when compiled with -m32:

 import std.range;

 void main()
 {
      float st = 0.000000000000000;
      float step = 12.038554191589355;
      float en = 999.200012207031250;

      iota(st, en, step);
 }

 Ali


Thank you, Ali.
will be waiting it will be fixed.
Apr 04 2013
prev sibling parent reply Alexandr Druzhinin <drug2004 bk.ru> writes:
05.04.2013 1:26, Ali Çehreli пишет:

 I was wrong. What I found is just a workaround. Created a bug for iota:

    http://d.puremagic.com/issues/show_bug.cgi?id=9877

 The following program fails when compiled with -m32:

 import std.range;

 void main()
 {
      float st = 0.000000000000000;
      float step = 12.038554191589355;
      float en = 999.200012207031250;

      iota(st, en, step);
 }

 Ali


Ali, please, take a look at this http://dpaste.dzfl.pl/44e64eb0
it's code of ctor of iota float point specialization. And it's strange 
for me, that pastEnd and (start + count * step) have the same value, but 
give different comparing results.
Apr 10 2013
parent Alexandr Druzhinin <drug2004 bk.ru> writes:
11.04.2013 12:13, Alexandr Druzhinin пишет:

 Ali, please, take a look at this http://dpaste.dzfl.pl/44e64eb0
 it's code of ctor of iota float point specialization. And it's strange
 for me, that pastEnd and (start + count * step) have the same value, but
 give different comparing results.


it fixes bug (I guess) http://dpaste.dzfl.pl/b9577e70
Apr 10 2013