Mirror of forum.nim-lang.org

5 :: Float format and unsigned casting

• ventor3000 (orginal) [2012-05-08T21:34:37+02:00] view original

  Hi.

  I've been looking into nimrod lately and really likes what i see. Im trying to port a BigInteger class i made in c# just to learn, and i'd need to cast from an int to int64 as if they were unsigned.

  Problem: var t:int=0xffffffff #same as -1 if signed var q:int64=t #q is now -1. How can i cast the to the unsigned value? echo(q) #outputs '-1'

  Another question I have is the formatting of floats: echo(formatFloat(1.2,ffDefault,6)) outputs 1.20000 (Why 5 digits after dot by the way?) I want to format 1.2 to .... "1.2" but i cant find how to do it.

  /Robert

---

• Araq (orginal) [2012-05-09T02:18:12+02:00] view original

  1. Check out system.toUxx and system.ze: toU32
  2. It was a bug, I fixed it in the github version; this code should do what you want:

  echo(formatFloat(1.2,ffDefault,0))
  Run

  PS: Yes the forum's wiki syntax parser is about to be improved...

---

• ventor3000 (orginal) [2012-05-09T07:55:57+02:00] view original

  toU32 looks nice for my purposes.

  Is there a way to convert unsigned integers to strings?

  /Robert

---

• Araq (orginal) [2012-05-09T21:31:21+02:00] view original

  Proper support for unsigned is planned (don't hold your breath though). For now, you need to wrap some C function to parse unsigned numbers. Or you parse them as hexadecimal numbers. Or you parse them as int64 and convert then toU32...

---

• Alex12 (orginal) [2012-09-13T19:16:57+02:00] view original

  Sorry for my English. I am recently familiar with the nimrod language and it was pleasant to me. I have simple test the formatting of floats:

  
    var f: float
    var s: string
    var t = cpuTime()
    for i in 0..99999:
      f = float(i)
      s = f.formatFloat(ffDecimal,4)
    t = cpuTime() - t
    stdout.writeln t
  Run

  Average result for nimrod: 6.8700000000000006e-001 (0.687) sec. The similar test for Delphi passes for: 0.038 sec. The similar test for C# passes for: 0.047 sec.

  Why nimrod formatFloat too slow?

---

• Araq (orginal) [2012-09-13T19:25:41+02:00] view original

  parseFloat and formatFloat have known performance problems. formatFloat ends up calling C's sprintf with a format string that can't be optimized by the C compiler. Patches are welcome. ;-)

  PS: This thread is old, the github version now has support for unsigned.

---

• Araq (orginal) [2012-09-18T22:17:02+02:00] view original

  While parseFloat and formatFloat are indeed slow, your example program creates lots of temporary GC-ed strings and so spends most of its time in GC. I get better results with --gc:boehm for this program. Nimrod's GC is no copying collector so it can't be as fast as C#'s when you create lots of garbage. (Delphi/FPC uses no GC.) Nimrod's GC is optimized for responsiveness, not throughput. A copying GC might be implemented one day but we also have other features planned to improve string handling.

---

• Alex12 (orginal) [2012-09-19T01:48:52+02:00] view original

  I understand that this test creates lots of temporary GC-ed strings. Before this test, i used parseFloat and formatFloat in program for read and save csv files. Results for reading csv files are comparable with Delphi, but save to file much more slowly. Then i create his restricted variant of formatFloat and save to csv files only slightly slowly Delphi. In Delphi analog formatFloat implemented with asm code. Yes Delphi/FPC uses no GC, but they also allocate and free memory for strings. What is --gc:boehm? After run program i get could not load: boehmgc.dll

---

• leledumbo (orginal) [2012-09-19T17:33:53+02:00] view original

  What is --gc:boehm?

  It's a compiler switch to use Boehm GC instead of Nimrod's own GC. It relies on Boehm GC library.

---