Mirror of forum.nim-lang.org

10866 :: 9999999999999999.0 – 9999999999999998.0

• jasonfi (orginal) [2024-01-11T18:55:50+01:00] view original

  I found this programming language challenge on HN: https://news.ycombinator.com/item?id=38952491

  Nim v2.0.2 fails at this, giving 2.0 as the answer.

  
  var c = 9999999999999999.0 - 9999999999999998.0
  echo $c
  Run

  I tried giving c type float64, but the answer is the same.

---

• morturo (orginal) [2024-01-11T19:14:37+01:00] view original

  it works with decimal but not with floats

  import decimal/decimal
  
  let a = newDecimal("9999999999999999.0")
  let b = newDecimal("9999999999999998.0")
  echo a - b
  Run

---

• auxym (orginal) [2024-01-11T19:30:31+01:00] view original

  Nim doesn't "fail", Nim uses machine native floats by default (double precision to be exact) and your your CPU conforms to the IEEE 754 spec for floats. Therefore, this result is expected.

  What Every Computer Scientist Should Know About Floating-Point Arithmetic : https://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html

---

• Isofruit (orginal) [2024-01-11T19:37:07+01:00] view original

  Just to emphasize what auxym just stated:

  Your browser will fuck this up as well because JS does not have integers, its "number" type is all floats. Which is why your browser is incapable of handling really large numbers like that unless you start using BigInt.

  If you want to test this out, open your browser console and enter 9999999999999999. You'll find said console not printing back out the number you typed in. Because the float spec is really screwy for representing large numbers with floats.

---

• giaco (orginal) [2024-01-11T19:53:36+01:00] view original

  This just proves that you don't know how a float works inside your CPU

---

• jasonfi (orginal) [2024-01-12T03:36:41+01:00] view original

  I did wonder if the fault wasn't at a lower level, but then Perl 6 and Wolfram both got the right answer. No need for the snarky reply.

---

• jasonfi (orginal) [2024-01-12T06:58:23+01:00] view original

  The reason I posted, was to link from HN to this forum, to give the community a little boost. I'm not alarmed by this.

---

• PMunch (orginal) [2024-01-17T16:21:15+01:00] view original

  This works with MAPM as well. As everyone else have already pointed out it's just a quirk of how IEEE floating point numbers work. MAPM has some nice convenience features as well, so this works and produces the number 1:

  import mapm
  
  var c = 9999999999999999.0'm - 9999999999999998.0'm
  echo c
  Run

---