bin/157177: primes(1) prints non-prime for numbers &gt; 2^32

Robert Lorentz robert.lorentz at gmail.com
Tue Jun 14 04:00:24 UTC 2011

The following reply was made to PR bin/157177; it has been noted by GNATS.

From: Robert Lorentz <robert.lorentz at gmail.com>
To: bug-followup at FreeBSD.org,
kcwu at kcwu.csie.org
Cc:
Subject: Re: bin/157177: primes(1) prints non-prime for numbers &gt; 2^32
Date: Mon, 13 Jun 2011 23:29:33 -0400

I regression tested this on FreeBSD 8.2-RELEASE against FreeBSD =
9.0-CURRENT r221981, using primegen-0.97 on amd64 platform and came up =
with some interesting results.

First off, from primes(1) man page, syntax is:

primes [ low [high] ]

So your bc line effectively says:

primes 4294967296=20

Where low =3D 4294967296 and high is not explicitly stated.  According =
to the man page high defaults to 1000000000 where the last prime =
possible is 999999937.  Since 1,000,000,000 < 4,294,967,296, the correct =
output of "primes 4294967296" should be nothing.

In FreeBSD 8.2-RELEASE, primes correctly does output nothing.

However on FreeBSD 9.0-CURRENT, primes incorrectly prints primes =
starting at 4,294,967,296 and seems to go forever (not sure where it =
will stop).  This is contrary to what the manual page says and is a =
problem.

Your original problem I did regression test and confirm to be working in =
8.2-RELEASE and broken in 9.0-CURRENT.  I isolated one of your examples =
using:

primes 4295360520 4295360522 | xargs -n 1 factor

On FreeBSD 9.0-CURRENT: 4295360521: 65539 65539
On FreeBSD 8.2-RELEASE: No output=20

On FreeBSD 9.0-CURRENT I debugged the source in =
/usr/ports/math/primegen/work/primegen-0.97 a bit and realized that if I =
ran the compiled version in =
/usr/ports/math/primegen/work/primegen-0.97/primes I got the correct =
expected results.  However, if I run the installed version in =
/usr/games/primes, I get the incorrect results.  The binaries in those =
two places aren't the same (verified using md5). =20

This appears to be an issue with the port building, probably building in =
32 bit.  If the inputs to primes are interpreted as 32bit then a "low" =
of (2^32 + 1) is interpreted as 1, therefore being less than 1000000000, =
therefore the code would continue to generate primes, and if this is the =
case then I wouldn't be surprised that the prime generation code also =
would misbehave.