Re: How high can wc go?

[Elie De Brauwer <eliedebrauwer%gmail.com@localhost>]

On 06/17/10 10:14, Ray Phillips wrote:
> The other day on a NetBSD/i386 5.0.2 machine I made a 1833 GiB file
> consisting of 30752636928 64-character lines (including the new line
> char). Later I ran wc on it, expecting its output to be incorrect as a
> result of overflows in the variables it uses, but the answers were right:
>
> % ls -l /mnt2
> total 3845018400
> -rw-r--r-- 1 root wheel 1968168763392 Jun 14 08:28 1833GiB.txt
> % /usr/bin/time wc /mnt/1833GiB.txt
> 30752636928 30752636928 1968168763392 /mnt2/1833GiB.txt
> 125997.76 real 121238.03 user 3731.70 sys
> %
> % dc
> 4k
> 1833 2 30^ * 64 /p
> 30752636928.0000
>
> 30752636928 64 * p
> 1968168763392
> q
> %
>
> I was wondering if anyone knows how many lines, words, and characters wc
> can keep track of? It seems that it uses some kind of tricky technique
> like dc to avoid overflows occurring?
>
>
> Ray


If you look in src/usr.bin/wc/wc.c this contains the wc_count_t typedef

#ifdef NO_QUAD
typedef u_long wc_count_t;
# define WCFMT  " %7lu"
# define WCCAST unsigned long
#else
typedef u_quad_t wc_count_t;
# define WCFMT  " %7llu"
# define WCCAST unsigned long long
#endif

And this shows that it will support either a 64bit counter or a 32bit 
counter.  This in a typical case you will notice a wrap when you count 
more than 2^64 characters.

>>> print 2**64
18446744073709551616

Or this amount in byte.

E.