fd2ce3fa2
Denys Vlasenko
2026-01-18
Subject
vmstat: show shorter, human-readable accumulating counts if they grow large
Binary
No binary (build failed)
Architecture & Compiler Flags
gcc -Wp,-MD,applets/.usage.d -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -Iinclude -Iinclude -o applets/usage applets/usage.c
Build Log
(log yok)
Commit Diff GitHub →
procps/vmstat.c+29 -13
| @@ -372,25 +372,41 @@ static void print_row(const unsigned data[NCOLS], | |
| if ((col.mod & MOD_DECREMENT) && value) | |
| value--; | |
| + | if (!is_first) |
| + | bb_putchar(' '); |
| + | is_first = 0; |
| /* memory can easily overflow the columns: | |
| * r b swpd free buff cache si so bi bo in cs us sy id wa st gu | |
| * 3 0 0 2019152 1029788 8602060 0 0 29 211 5699 13 9 5 86 0 0 0 | |
| - | * ^ ^ ^ shortened width was here |
| - | * To improve this, shorten the requested field width if previous field overflowed. |
| + | * ^-------^-------^-attempt to shorten width here |
| + | * To improve this, shorten the next field width if previous field overflowed. |
| */ | |
| + | //TODO: more improvements: |
| + | //Shift left entire set of memory fields if it has space padding and overruns at the right, |
| + | //as in the example above (zero bytes in swap: can replace " 0" with "0" to make data fit). |
| { | |
| - | int digits_printed, width; |
| + | int chars_printed, width; |
| width = (col.width - overrun >= 0 ? col.width - overrun : 0); | |
| - | digits_printed = printf(" %*u" + is_first, width, value) - 1 + is_first; |
| - | overrun += (digits_printed - col.width); |
| - | is_first = 0; |
| - | //TODO: more improvements: |
| - | //Shift left entire set of memory fields if it is space-padded at the left and overruns at the right, |
| - | //as in the example above (zero bytes in swap). |
| - | //For large intervals such as "vmstat 60" (one minute), |
| - | //can use smart_ulltoa4() to format "si", "so", "in" and "cs"; |
| - | //smart_ulltoa5() to format "bi" and "bo": |
| - | //this way, we can show arbitrarily large accumucated counts for these. |
| + | /* For large intervals such as "vmstat 60" (one minute), or |
| + | * for cases with a lot of block I/O (try 'grep -rF noT_eXisting /usr'), |
| + | * can use smart_ulltoa4() to format "si", "so", "in" and "cs"; |
| + | * smart_ulltoa5() to format "bi" and "bo". |
| + | * This way, we can show arbitrarily large accumulated counts for these. Example: |
| + | * r b swpd free buff cache si so bi bo in cs us sy id wa st gu |
| + | * 1 0 0 947636 76548 15373248 0 0 657k 23284 2.3m 4.1m 24 9 65 0 0 0 |
| + | * Downside: incompatible output format. Do we need to make this conditional on an option? |
| + | */ |
| + | if ((col.width>>1) == 4/2) { |
| + | char buf45[6]; |
| + | if (col.width == 4) |
| + | smart_ulltoa4(value, buf45, " kmgtpezy")[0] = '\0'; |
| + | else |
| + | smart_ulltoa5(value, buf45, " kmgtpezy")[0] = '\0'; |
| + | chars_printed = printf("%*s", width, skip_whitespace(buf45)); |
| + | } else { |
| + | chars_printed = printf("%*u", width, value); |
| + | } |
| + | overrun += (chars_printed - col.width); |
| } | |
| } | |