Two-channel graph (Graph2) meter style#1415
Two-channel graph (Graph2) meter style#1415Explorer09 wants to merge 13 commits intohtop-dev:mainfrom
Conversation
Explorer09
force-pushed
the
graph-with-neg
branch
3 times, most recently
from
f47a975 to
745f394
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
745f394 to
af6b927
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
3 times, most recently
from
31cb513 to
bedca8f
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
4 times, most recently
from
ade4922 to
54d3d1c
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
3 times, most recently
from
728ef87 to
3033b1d
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
2 times, most recently
from
aa96aa4 to
10bb617
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
10bb617 to
8ab21e9
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
8ab21e9 to
30691c9
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
2 times, most recently
from
f57affb to
5e3eeb1
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
2 times, most recently
from
b5560f5 to
8a8cfda
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
8a8cfda to
eac20b6
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
3 times, most recently
from
c7c11bc to
b446269
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
b446269 to
a84d790
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
2 times, most recently
from
8ce2c96 to
4ae1d01
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
3 times, most recently
from
95052ef to
d370bf3
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
from
d370bf3 to
c8d8c19
Compare
Explorer09
force-pushed
the
graph-with-neg
branch
2 times, most recently
from
5e7bb49 to
0d092e5
Compare
|
The traditional graphical meters have a gray baseline. Does it make sense to have that here as well? |
|
Oh, and another note... Personally I do not like just adding numbers to names, at least for new functionality. How about something a little bit more descriptive? Perhaps "NGraph" (for negative) or "TCGraph" (for two-channel)? |
I've tried it. It would visually distract too much, as the "baseline" on the middle would be two times as thick. And I didn't like it.
"Graph2C" ? |
Ah, right... But I have some systems that have just little to no disk io activity, especially for reading. Some random blue dots floating around is distracting, too. 😝 Not sure what's better.
Oh, I read the name "Graph2" as "Graph, increment by one". If you take it as an abbreviation for "Graph two-channel" it makes kind of sense. 😉 Anyway, I would prefer your suggestion "Graph2C" if it is possible and fine for you. Any other opinions? |
|
This screenshot is made in a machine started from live media. 
Nothing in the disk meter (as it is running from optical media nothing is ever written), and just very little activity in network meter. It's really hard to tell where the (virtual) base line is. |
|
... and created a mockup. 
That's what a base line could look like. Still not sure which one I would prefer... |
The baselines could look bad if you have a terminal emulator that displays braille matrix with gaps between characters. See this screenshot for what I mean: 
In the screenshot I also use an alternative to two baselines in the middle. |
|
Actually it's not that bad. I think that's the version I would prefer, and with proper spacing (I use font Terminus here, and my screenshot were made with that) this would look even better. |
When both the options "Add guest time in CPU meter percentage" and "Detailed CPU time" are turned off, the CPU meter used to show the "virtual" CPU time in the bar display as a glitch. Now fix it. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
…d mode" This reverts commit 47aeb0a.
The time displayed is (steal+guest), not just guest CPU time. Regression from 3d8fa0b Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Linux and PCP platform code used to write the 'virtualized' (steal+guest) CPU time into the CPU_METER_IRQ item when the "detailed CPU time" option is off, which would result in a wrong color painted for virtual CPU time in bar mode. Write the 'virtualized' CPU time to CPU_METER_STEAL instead, which is more appropriate for the job. Also update the "virtualized" CPU meter item in the help screen: (1) The color is now CPU_STEAL for consistency. (2) In monochrome mode, two dummy items are displayed before the "virtualized" word so that the bar meter symbol mapping is correct. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Round the graph meter's dynamic scale to a power of two and print the graph scale. For a percent graph, a "%" character is printed in place of the scale. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
This is a prerequisite for the feature "Graph meter coloring (with GraphData structure rework)". powerOf2Floor() will utilize __builtin_clz() or stdc_bit_floor_ui() (__builtin_clz() is preferred) if either is supported. popCount8() will utilize ARM NEON instructions and x86 POPCNT instruction if the machine supports either of them. I am not adopting the C23 standard interface stdc_count_ones_uc() yet, as I am not sure C libraries would implement it as fast as our version. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Rewrite the entire graph meter drawing code to support color graph drawing in addition to the dynamic scaling (both can work together because of the new GraphData structure design). The colors of a graph are based on the percentages of item values of the meter. The rounding differences of each terminal character are addressed through the different numbers of braille dots. Due to low resolution of the character terminal, the rasterized colors may not look nice, but better than nothing. :) The new GraphData structure design has two technical limitations: * The height of a graph meter now has a limit of 8191 (terminal rows). * If dynamic scaling is used, the "total" value or scale of a graph now has to align to a power of 2. The code is designed with the anticipation that the graph height may change at runtime. No UI or option has been implemented for that yet. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Specifically 'PIXPERROW_*' and 'GraphMeterMode_dots*' constants. They were commented out rather than removed in the previous commit (for ease of code reviewing). Now this commit removes the constant defines for good.
Introduce a new meter mode (meter style) called "Graph2". The "Graph2" name stands for "two-channel graph" or "two-quadrant graph". This meter mode is not in the list of default supported modes. This "Graph2" mode is intended for meters with two item values. It plots the first item above the x-axis and the second item below the x-axis. Items other than the first two are always hidden. This code is written with the intention to share as much code with the Graph meter mode as possible. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Specifically it's the DiskIORateMeter part that supports "Graph2" mode. For the combined DiskIOMeter, the "IO rate" part can show as two-channel graph while the "IO time" part remains a one-channel graph. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
This special case of "details" value would print as ':' in the ASCII display mode. It is used in "Graph2" meter drawing only. Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
3 participants
This pull request implements the "two-channel" graph style as the new "Graph2" meter mode. This is based on #714 so that the graphs will be shown with colors.
This new mode is designed for a few meters only— only
DiskIORateMeter(#1763, merged) andNetworkIOMetercan correctly utilise this mode by showing the input and output rates as separate channels.The feature was previously discussed in #1390.
Below is a preview screenshot taken in macOS Sequoia.
I included both Unicode mode and
--no-unicodemode.