Path: csiph.com!usenet.pasdenom.info!weretis.net!feeder4.news.weretis.net!usenet.ukfsn.org!not-for-mail
From: Martin Gregorie <martin@address-in-sig.invalid>
Newsgroups: comp.lang.java.programmer
Subject: Re: floating point
Date: Tue, 24 Jul 2012 20:34:47 +0000 (UTC)
Organization: UK Free Software Network
Lines: 50
Message-ID: <jun0t7$bh5$1@localhost.localdomain>
References: <d94d1dd0-deb5-4259-b5a0-fb3d57e99c55@googlegroups.com> <jumuu7$6np$1@dont-email.me>
NNTP-Posting-Host: 84.45.235.129
Mime-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
X-Trace: localhost.localdomain 1343162087 11813 84.45.235.129 (24 Jul 2012 20:34:47 GMT)
X-Complaints-To: usenet@localhost.localdomain
NNTP-Posting-Date: Tue, 24 Jul 2012 20:34:47 +0000 (UTC)
User-Agent: Pan/0.135 (Tomorrow I'll Wake Up and Scald Myself with Tea; GIT 30dc37b master)
Xref: csiph.com comp.lang.java.programmer:16313

On Tue, 24 Jul 2012 13:01:11 -0700, markspace wrote:

> On 7/24/2012 12:57 PM, bob smith wrote:
>> I can never remember this.
>>
>> Let's say you have an integer and a float in an operation.  Is the
>> result always a float?
> 
> 
> You can just try this to see what you get, or read the JLS.  From
> memory, I'm 90% sure that ints will be promoted to floats.  I'm not sure
> how order of operations affects this.  In the expression (1+1)/2.0f,
> does the addition use floats or ints?  I'm not sure, but I think ints.

I think this:

$ cat MathTest.java
public class MathTest
{
   public static void main(String[] args)
   {
      double x = (9 / 2) / 5;
      System.out.println("x = " + x);
      double y = (9 / 2) / 5.0;
      System.out.println("y = " + y);
      double z = (9 / 2.0) / 5.0;
      System.out.println("z = " + z);
   }
}
$ javac MathTest.java 
$ java MathTest
x = 0.0
y = 0.8
z = 0.9


shows pretty conclusively that widening from int to double isn't done 
until the last possible moment. 

The calculation of x is entirely integral, with widening to double only 
occurring at the assignment. Calculating y can only produce that result 
if the first division was integral and its result was then widened to 
double before the second division. And, of course, the z calculation was 
entirely done as floating point
  

-- 
martin@   | Martin Gregorie
gregorie. | Essex, UK
org       |