Hi,

I'm having trouble understanding why in some cases a double value will be rounded up and other times down, for the same code. Here's a snippet with code I tried to debug:

  int getNumberOfStitchesForRowLength(double rowLength)
  {
    writeln("input ", rowLength, " ", currentGauge.stitch_gauge, " ", currentGauge.gauge_length);

    writeln("stitches: ", (rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length,  " -> " , ((rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length).roundTo!int);

    double end = 2.5;
    double start = 0;
    writeln("I expect: ", ((abs(end-start)*10)/10).roundTo!int);

    return ((rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length).roundTo!int;

  }

And here's some output from that:
input 2.5 10 10
stitches: 2.5 -> 2
I expect: 3

OR, similarly, I get

input 3.5 10 10
stitches: 3.5 -> 3
I expect: 4

However, it works as I would expect for one value:

input 1.5 10 10
stitches: 1.5 -> 2
I expect: 2

I would appreciate some ideas for why I see this seeming inconsistency. Or if someone can share how I can further debug this.

Thanks!

I'm using DMD32 D Compiler v2.074.1 on Windows

On 07/07/2017 11:29 AM, alex_ca wrote:

> input 2.5 10 10
> stitches: 2.5 -> 2
> I expect: 3

That's because what is printed as 2.5 is actually a little less than that. (Try printing with a format string like "%.20f".)

The common way of dealing with this issue is to add 0.5 before the conversion:

import std.stdio;
import std.conv;

void main() {
    for (double d = 1; d < 2; d += 0.1) {
        writefln("%s (%.20f): %s", d, d, (d + 0.5).to!int);
    }
}

1 (1.00000000000000000000): 1
1.1 (1.10000000000000008882): 1
1.2 (1.20000000000000017764): 1
1.3 (1.30000000000000026645): 1
1.4 (1.40000000000000035527): 1
1.5 (1.50000000000000044409): 2
1.6 (1.60000000000000053291): 2
1.7 (1.70000000000000062172): 2
1.8 (1.80000000000000071054): 2
1.9 (1.90000000000000079936): 2

Ali

On 07/07/2017 08:29 PM, alex_ca wrote:
> I'm having trouble understanding why in some cases a double value will be rounded up and other times down, for the same code. Here's a snippet with code I tried to debug:
> 
>    int getNumberOfStitchesForRowLength(double rowLength)
>    {
>      writeln("input ", rowLength, " ", currentGauge.stitch_gauge, " ", currentGauge.gauge_length);
> 
>      writeln("stitches: ", (rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length,  " -> " , ((rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length).roundTo!int);
> 
>      double end = 2.5;
>      double start = 0;
>      writeln("I expect: ", ((abs(end-start)*10)/10).roundTo!int);
> 
>      return ((rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length).roundTo!int;
> 
>    }
> 
> And here's some output from that:
> input 2.5 10 10
> stitches: 2.5 -> 2
> I expect: 3
> 
> OR, similarly, I get
> 
> input 3.5 10 10
> stitches: 3.5 -> 3
> I expect: 4
> 
> However, it works as I would expect for one value:
> 
> input 1.5 10 10
> stitches: 1.5 -> 2
> I expect: 2
> 
> I would appreciate some ideas for why I see this seeming inconsistency. Or if someone can share how I can further debug this.

Works for me when I plug in the exact values:

----
import std.stdio;
import std.math: abs;
import std.conv: roundTo;
import std.math;

struct Gauge
{
    double stitch_gauge;
    double gauge_length;
}

Gauge currentGauge = Gauge(10, 10);

int getNumberOfStitchesForRowLength(double rowLength)
{
    writeln("input ", rowLength, " ", currentGauge.stitch_gauge, " ",
        currentGauge.gauge_length);

    writeln("stitches: ",
        (rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length,
        " -> " ,
        ((rowLength * currentGauge.stitch_gauge) /
            currentGauge.gauge_length).roundTo!int);

    return ((rowLength * currentGauge.stitch_gauge) / currentGauge.gauge_length).roundTo!int;
}

void main()
{
    foreach (x; [1.5, 2.5, 3.5])
    {
        getNumberOfStitchesForRowLength(x);
    }
}
----

Prints:

----
input 1.5 10 10
stitches: 1.5 -> 2
input 2.5 10 10
stitches: 2.5 -> 3
input 3.5 10 10
stitches: 3.5 -> 4
----

As expected, right?

Could it be that the inputs aren't exactly the numbers you think they are? For example, when you change one the 10s to 9.9999999, it still prints as "10", but the results are different.

Thanks, that was what was happening.