www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Printing floating point numbers

reply Yui Hosaka <hos hos.ac> writes:
The following code prints weird results on my machine.

----
import std.stdio;
void main() {
   real a = 0.16;
   real b = 0.016;
   writefln("%.1f", a);
   writefln("%.2f", b);
}
----

Output:
---
0.2
0.01
----

I am using dmd on Windows. It doesn't happen when compiling with 
-m32.
----
$ dmd
DMD32 D Compiler v2.088.1-dirty
----

Do you have any idea for this issue?
Oct 24 2019
next sibling parent Adam D. Ruppe <destructionator gmail.com> writes:
On Thursday, 24 October 2019 at 20:48:02 UTC, Yui Hosaka wrote:

 Do you have any idea for this issue?


my suspicion would be that .016 is actually represented as 
.0159999 and the .2 round ignores all those 9's...
Oct 24 2019
prev sibling parent reply berni44 <dlang d-ecke.de> writes:
On Thursday, 24 October 2019 at 20:48:02 UTC, Yui Hosaka wrote:

 Do you have any idea for this issue?


I added a bug report: 
https://issues.dlang.org/show_bug.cgi?id=20320

Internally the conversation from the binary representation of the 
value to the printed one is done by a call to a C function called 
snprintf. Probably the error is inside of this function call. But 
it could also happen before. Maybe the internal representation of 
0.016 is allready wrong.

It would be helpful if you could run the following program and 
post the output:

import std.stdio;
import std.format;
void main()
{
     real b = 0.016;
     writefln!"%.2f"(b);
     foreach (c;format("%r",b)) writef("%x ",c);
     writeln();

     char[6] sprintfSpec = "%*.*Lf";
     char[512] buf = void;
     import core.stdc.stdio : snprintf;
     immutable n = snprintf(buf.ptr, buf.length, sprintfSpec.ptr, 
0, 2, b);
     writeln(buf[0..n]);
}
Oct 24 2019
parent reply Yui Hosaka <hos hos.ac> writes:
On Friday, 25 October 2019 at 06:53:38 UTC, berni44 wrote:

 On Thursday, 24 October 2019 at 20:48:02 UTC, Yui Hosaka wrote:

 Do you have any idea for this issue?


 I added a bug report: 
 https://issues.dlang.org/show_bug.cgi?id=20320

 Internally the conversation from the binary representation of 
 the value to the printed one is done by a call to a C function 
 called snprintf. Probably the error is inside of this function 
 call. But it could also happen before. Maybe the internal 
 representation of 0.016 is allready wrong.

 It would be helpful if you could run the following program and 
 post the output:

 import std.stdio;
 import std.format;
 void main()
 {
     real b = 0.016;
     writefln!"%.2f"(b);
     foreach (c;format("%r",b)) writef("%x ",c);
     writeln();

     char[6] sprintfSpec = "%*.*Lf";
     char[512] buf = void;
     import core.stdc.stdio : snprintf;
     immutable n = snprintf(buf.ptr, buf.length, 
 sprintfSpec.ptr, 0, 2, b);
     writeln(buf[0..n]);
 }


Thank you for forwarding the issues.
The outputs for your program are as follows:

Without compiler options (32-bit):
---
0.01
3b df 4f 8d 97 6e 12 83 f9 3f
0.01
---
and some trailing characters. "./a | od -c" shows:
---
0000000   0   .   0   1  \r  \n   3   b       d   f       4   f   
     8
0000020   d       9   7       6   e       1   2       8   3       
f   9
0000040       3   f      \r  \n   0   .   0   1 377 377 206 003  
\r  \n
0000060
---

With "-O" (32-bit), the output changes in every run, such as:
---
0.01
3b df 4f 8d 97 6e 12 83 f9 3f
0.01S
---
0.01
3b df 4f 8d 97 6e 12 83 f9 3f
0.01b
---
0.01
3b df 4f 8d 97 6e 12 83 f9 3f
0.01c
---

With "-m64" or "-m64 -O":
---
0.02
3b df 4f 8d 97 6e 12 83 f9 3f
0.00
---
Oct 25 2019
parent berni44 <dlang d-ecke.de> writes:
On Friday, 25 October 2019 at 14:51:33 UTC, Yui Hosaka wrote:

 The outputs for your program are as follows: [...]


Thanks for the output. I added it to the bug report.

The internal representation of the value is correct in all 
versions. While the result of the direct call of snprintf always 
gives the wrong answer. Therefore I think, the bug is to be found 
in this function. Resently Robert Schadek noted, that it would be 
nice, to have this function be rewritten in D anyway. I'll have a 
look if I can do this in the next weeks but do not want to 
promise anything yet.


 With "-m64" or "-m64 -O":
 ---
 0.02
 3b df 4f 8d 97 6e 12 83 f9 3f
 0.00
 ---


The "0.00" is strange.
Oct 25 2019