top: refined memory graphs two abreast summary display
When more than two cpus are displayed per summary area line in graph form, those memory graphs were scaled to that same width for consistency & aesthetics. However, they probably shouldn't have been reduced to less than terminal width due to a resulting loss of information. [ after all, detailed memory stats are never reduced ] So now, supporting logic was refactored to behave just as it did before the 4 toggle was expanded beyond '1'. [ the changes impact the 2 memory graphs exclusively ] Reference(s): https://www.freelists.org/post/procps/top-enhancements-2-bugs-swatted,1 Signed-off-by: Jim Warner <james.warner@comcast.net>
This commit is contained in:
parent
3543172d4b
commit
5c5bff392b
@ -229,8 +229,14 @@ static int Numa_node_sel = -1;
|
||||
#define GRAPH_prefix_std 25 // '%Cpunnn: 100.0/100.0 100[' or 'nnn-nnn: 100.0/100.0 100['
|
||||
#define GRAPH_prefix_abv 12 // '%Cpunnn:100[' or 'nnn-nnn:100[' or 'GiB Mem 100[' or 'GiB Swap 99['
|
||||
#define GRAPH_suffix 2 // '] ' (bracket + trailing space)
|
||||
static float Graph_adj; // bars/blocks scaling factor
|
||||
static int Graph_len; // scaled length (<= GRAPH_length_max)
|
||||
// first 3 more static (adj_geometry), last 3 volatile (sum_tics/do_memory)
|
||||
struct graph_parms {
|
||||
float adjust; // bars/blocks scaling factor
|
||||
int length; // scaled length (<= GRAPH_length_max)
|
||||
int style; // rc.graph_cpus or rc.graph_mems
|
||||
long total, part1, part2; // elements to be graphed
|
||||
};
|
||||
static struct graph_parms *Graph_cpus, *Graph_mems;
|
||||
static const char Graph_blks[] = " ";
|
||||
static const char Graph_bars[] = "||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||";
|
||||
|
||||
@ -2073,15 +2079,23 @@ static void adj_geometry (void) {
|
||||
if (Curwin->rc.double_up) {
|
||||
int num = (Curwin->rc.double_up + 1);
|
||||
int pfx = (Curwin->rc.double_up < 2) ? GRAPH_prefix_std : GRAPH_prefix_abv;
|
||||
Graph_len = (Screen_cols - (ADJOIN_space * Curwin->rc.double_up) - (num * (pfx + GRAPH_suffix))) / num;
|
||||
Graph_cpus->length = (Screen_cols - (ADJOIN_space * Curwin->rc.double_up) - (num * (pfx + GRAPH_suffix))) / num;
|
||||
Graph_mems->length = (Screen_cols - ADJOIN_space - (2 * (GRAPH_prefix_std + GRAPH_suffix))) / 2;
|
||||
} else {
|
||||
Graph_len = Screen_cols - (GRAPH_prefix_std + GRAPH_length_max + GRAPH_suffix);
|
||||
if (Graph_len >= 0) Graph_len = GRAPH_length_max;
|
||||
else Graph_len = Screen_cols - GRAPH_prefix_std - GRAPH_suffix;
|
||||
Graph_cpus->length = Screen_cols - (GRAPH_prefix_std + GRAPH_length_max + GRAPH_suffix);
|
||||
if (Graph_cpus->length >= 0) Graph_cpus->length = GRAPH_length_max;
|
||||
else Graph_cpus->length = Screen_cols - GRAPH_prefix_std - GRAPH_suffix;
|
||||
Graph_mems->length = Graph_cpus->length;
|
||||
}
|
||||
if (Graph_len < GRAPH_length_min) Graph_len = GRAPH_length_min;
|
||||
if (Graph_len > GRAPH_length_max) Graph_len = GRAPH_length_max;
|
||||
Graph_adj = (float)Graph_len / 100.0;
|
||||
if (Graph_cpus->length < GRAPH_length_min) Graph_cpus->length = GRAPH_length_min;
|
||||
if (Graph_cpus->length > GRAPH_length_max) Graph_cpus->length = GRAPH_length_max;
|
||||
Graph_cpus->adjust = (float)Graph_cpus->length / 100.0;
|
||||
Graph_cpus->style = Curwin->rc.graph_cpus;
|
||||
|
||||
if (Graph_mems->length < GRAPH_length_min) Graph_mems->length = GRAPH_length_min;
|
||||
if (Graph_mems->length > GRAPH_length_max) Graph_mems->length = GRAPH_length_max;
|
||||
Graph_mems->adjust = (float)Graph_mems->length / 100.0;
|
||||
Graph_mems->style = Curwin->rc.graph_mems;
|
||||
|
||||
fflush(stdout);
|
||||
} // end: adj_geometry
|
||||
@ -3658,6 +3672,9 @@ static void before (char *me) {
|
||||
error_exit(fmtmk(N_fmt(X_THREADINGS_fmt), __LINE__, strerror(errno)));
|
||||
pthread_setname_np(Thread_id_tasks, "update tasks");
|
||||
#endif
|
||||
// lastly, establish support for graphing cpus & memory
|
||||
Graph_cpus = alloc_c(sizeof(struct graph_parms));
|
||||
Graph_mems = alloc_c(sizeof(struct graph_parms));
|
||||
#undef doALL
|
||||
} // end: before
|
||||
|
||||
@ -6132,7 +6149,7 @@ struct rx_st {
|
||||
* A *Helper* function to produce the actual cpu & memory graphs for |
|
||||
* these functions -- sum_tics (tertiary) and do_memory (secondary). |
|
||||
* (sorry about the name, but it keeps the above comment commitment) | */
|
||||
static struct rx_st *sum_rx (long total, long part1, long part2, int style) {
|
||||
static struct rx_st *sum_rx (struct graph_parms *these) {
|
||||
static struct {
|
||||
const char *part1, *part2, *style;
|
||||
} gtab[] = {
|
||||
@ -6141,30 +6158,30 @@ static struct rx_st *sum_rx (long total, long part1, long part2, int style) {
|
||||
};
|
||||
static __thread struct rx_st rx;
|
||||
char buf1[SMLBUFSIZ], buf2[SMLBUFSIZ], buf3[MEDBUFSIZ];
|
||||
int num1, num2, width;
|
||||
int ix, num1, num2, width;
|
||||
float scale;
|
||||
|
||||
scale = 100.0 / total;
|
||||
rx.pcnt_one = scale * part1;
|
||||
rx.pcnt_two = scale * part2;
|
||||
scale = 100.0 / these->total;
|
||||
rx.pcnt_one = scale * these->part1;
|
||||
rx.pcnt_two = scale * these->part2;
|
||||
if (rx.pcnt_one + rx.pcnt_two > 100.0 || rx.pcnt_two < 0)
|
||||
rx.pcnt_two = 0;
|
||||
rx.pcnt_tot = rx.pcnt_one + rx.pcnt_two;
|
||||
|
||||
num1 = (int)((rx.pcnt_one * Graph_adj) + .5),
|
||||
num2 = (int)((rx.pcnt_two * Graph_adj) + .5);
|
||||
if (num1 + num2 > Graph_len)
|
||||
num2 = Graph_len - num1;
|
||||
num1 = (int)((rx.pcnt_one * these->adjust) + .5),
|
||||
num2 = (int)((rx.pcnt_two * these->adjust) + .5);
|
||||
if (num1 + num2 > these->length)
|
||||
num2 = these->length - num1;
|
||||
|
||||
width = Graph_len;
|
||||
width = these->length;
|
||||
buf1[0] = buf2[0] = buf3[0] = '\0';
|
||||
--style; // now relative to zero
|
||||
ix = these->style - 1; // now relative to zero
|
||||
if (num1) {
|
||||
snprintf(buf1, sizeof(buf1), gtab[style].part1, num1, gtab[style].style);
|
||||
snprintf(buf1, sizeof(buf1), gtab[ix].part1, num1, gtab[ix].style);
|
||||
width += 2;
|
||||
}
|
||||
if (num2) {
|
||||
snprintf(buf2, sizeof(buf2), gtab[style].part2, num2, gtab[style].style);
|
||||
snprintf(buf2, sizeof(buf2), gtab[ix].part2, num2, gtab[ix].style);
|
||||
width += 2;
|
||||
}
|
||||
snprintf(buf3, sizeof(buf3), "%s%s", buf1, buf2);
|
||||
@ -6223,7 +6240,10 @@ static int sum_tics (struct stat_stack *this, const char *pfx, int nobuf) {
|
||||
/* display some kinda' cpu state percentages
|
||||
(who or what is explained by the passed prefix) */
|
||||
if (Curwin->rc.graph_cpus) {
|
||||
rx = sum_rx(tot_frme, rSv(stat_SUM_USR), rSv(stat_SUM_SYS), Curwin->rc.graph_cpus);
|
||||
Graph_cpus->total = tot_frme;
|
||||
Graph_cpus->part1 = rSv(stat_SUM_USR);
|
||||
Graph_cpus->part2 = rSv(stat_SUM_SYS);
|
||||
rx = sum_rx(Graph_cpus);
|
||||
if (Curwin->rc.double_up > 1)
|
||||
return sum_see(fmtmk("%s~3%3.0f%s", pfx, rx->pcnt_tot, rx->graph), nobuf);
|
||||
else {
|
||||
@ -6434,27 +6454,23 @@ static void do_memory (void) {
|
||||
my_misc = MEM_VAL(mem_TOT) - MEM_VAL(mem_FRE) - my_qued;
|
||||
my_used = MEM_VAL(mem_TOT) - MEM_VAL(mem_AVL) - my_misc;
|
||||
|
||||
rx = sum_rx(MEM_VAL(mem_TOT), my_misc, my_used, Curwin->rc.graph_mems);
|
||||
if (Curwin->rc.double_up > 1)
|
||||
snprintf(row, sizeof(row), "%s %s~3%3.0f%s"
|
||||
, scT(label), N_txt(WORD_abv_mem_txt), rx->pcnt_tot, rx->graph);
|
||||
else {
|
||||
Graph_mems->total = MEM_VAL(mem_TOT);
|
||||
Graph_mems->part1 = my_misc;
|
||||
Graph_mems->part2 = my_used;
|
||||
rx = sum_rx(Graph_mems);
|
||||
prT(bfT(0), mkM(MEM_VAL(mem_TOT)));
|
||||
snprintf(row, sizeof(row), "%s %s:~3%#5.1f~2/%-9.9s~3%s"
|
||||
, scT(label), N_txt(WORD_abv_mem_txt), rx->pcnt_tot, bfT(0)
|
||||
, rx->graph);
|
||||
}
|
||||
Msg_row += sum_see(row, mem2UP);
|
||||
|
||||
rx = sum_rx(MEM_VAL(swp_TOT), 0, MEM_VAL(swp_USE), Curwin->rc.graph_mems);
|
||||
if (Curwin->rc.double_up > 1)
|
||||
snprintf(row, sizeof(row), "%s %s~3%3.0f%s"
|
||||
, scT(label), N_txt(WORD_abv_swp_txt), rx->pcnt_tot, rx->graph);
|
||||
else {
|
||||
Graph_mems->total = MEM_VAL(swp_TOT);
|
||||
Graph_mems->part1 = 0;
|
||||
Graph_mems->part2 = MEM_VAL(swp_USE);
|
||||
rx = sum_rx(Graph_mems);
|
||||
prT(bfT(1), mkM(MEM_VAL(swp_TOT)));
|
||||
snprintf(row, sizeof(row), "%s %s:~3%#5.1f~2/%-9.9s~3%s"
|
||||
, scT(label), N_txt(WORD_abv_swp_txt), rx->pcnt_two, bfT(1), rx->graph);
|
||||
}
|
||||
Msg_row += sum_see(row, 1);
|
||||
|
||||
} else {
|
||||
|
@ -760,7 +760,7 @@ typedef struct WIN_t {
|
||||
//atic void keys_window (int ch);
|
||||
//atic void keys_xtra (int ch);
|
||||
/*------ Tertiary summary display support (summary_show helpers) -------*/
|
||||
//atic struct rx_st *sum_rx (long total, long part1, long part2, int style);
|
||||
//atic struct rx_st *sum_rx (struct graph_parms *these);
|
||||
//atic inline int sum_see (const char *str, int nobuf);
|
||||
//atic int sum_tics (struct stat_stack *this, const char *pfx, int nobuf);
|
||||
//atic int sum_unify (struct stat_stack *this, int nobuf);
|
||||
|
Loading…
Reference in New Issue
Block a user