As strange as it might sound it could indeed be improved.
I think it should list exabytes though.
We should probably change the off_t type (to unsigned?) and the calculation
using "double size".
Then listing a directory with multiple exabyte-sized files will be 0.00001%
faster also ;)
On Fri, Feb 06, 2026 at 12:32:33PM +0100, Arusekk wrote:
> While it might sound silly, who knows what happens some day.
> Individual data centers are moving beyond exabytes,
> and RISC-V already specifies a 128-bit architecture variant.
> Alternatively, someone might decide that multiple exabytes
> should be reported in terabytes instead (corectness improves
> hackability).
>
> I tested the patch by changing the constant "BKMGTPE" to just "BKMGT"
> and running:
>
> $ ./truncate -s 1E /tmp/zeros
> $ ./ls -lh /tmp/zeros
>
> Before (incorrect, 1 EiB != 1024.0 B):
> -rw-r--r-- 1 arusekk arusekk 1024.0 Feb 06 11:53 /tmp/zeros
> After (corrected, 1 EiB == 1048576 TiB):
> -rw-r--r-- 1 arusekk arusekk 1048576.0T Feb 06 11:53 /tmp/zeros
>
> A side effect is less memory accesses/calls, so listing a large
> directory could become marginally faster.
> ---
> libutil/human.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/libutil/human.c b/libutil/human.c
> index 7e39ba5..0f80b9b 100644
> --- a/libutil/human.c
> +++ b/libutil/human.c
> _AT_@ -1,7 +1,7 @@
> /* See LICENSE file for copyright and license details. */
> +#include <stdint.h>
> #include <stdio.h>
> #include <string.h>
> -#include <stdint.h>
>
> #include "../util.h"
>
> _AT_@ -13,7 +13,7 @@ humansize(off_t n)
> double size;
> int i;
>
> - for (size = n, i = 0; size >= 1024 && i < strlen(postfixes); i++)
> + for (size = n, i = 0; size >= 1024 && postfixes[i+1]; i++)
> size /= 1024;
>
> if (!i)
> --
> 2.52.0
>
>
--
Kind regards,
Hiltjo
Received on Fri Feb 06 2026 - 14:43:09 CET