The Linux /proc/stat fields "guest" and "guest_nice" are already
included in "user" and "nice". In Proc::collect do like Cpu::collect
does and don't double count them. The double-counting caused the CPU
usage denominator to be too large when VM guests were active, causing
the reported CPU usage percentages of all processes to be too small.
Closes: https://github.com/aristocratos/btop/issues/1608
coreCount now reflects only the schedulable CPUs. With hw.smt=0 on a
4-core/8-thread machine, coreCount becomes 4, so btop allocates 4
display slots instead of 8.
The collect loop now also checks CPUSTATS_ONLINE for each CPU and skips
offline CPUs, such as disabled hyperthreads.
btop uses i / 2 as the third MIB index when it should be i, exactly as
top does. This causes it to query the same CPU twice in a row and skip
half the CPUs entirely.
Also use the correct size for a single cpustats struct.
Process signal popup menus use the "current" selected process, which can drift
around. Fix by retaining the initial PID at popup creation.
Closes: https://github.com/aristocratos/btop/issues/846
- Add CFRef<T> template for CoreFoundation types (CFRelease)
- Add IORef wrapper for IOKit object types (IOObjectRelease)
- Refactor GPU code to use RAII wrappers for all local CF/IO objects
- Add 'apple' to shown_gpus description in btop_menu.cpp
- Wrap AppleSilicon::shutdown() call and declaration with #ifdef __APPLE__
so Linux GPU builds don't get undefined reference errors
- Add IOReport library linking in CMakeLists.txt for macOS CMake builds
Add GPU monitoring for Apple Silicon Macs using the IOReport framework
and IOHIDEventSystem for temperature readings.
Metrics supported:
- GPU utilization (from GPU Performance States residency)
- GPU power consumption (from Energy Model channel)
- GPU temperature (from IOHIDEventSystem thermal sensors)
- GPU clock speed (from DVFS frequency table + residency weighting)
- Memory usage (unified memory architecture, reports system RAM)
Implementation follows the existing Nvml/Rsmi/Intel backend pattern with
init/shutdown/collect<is_init> template functions in a new AppleSilicon
namespace under Gpu.
Build system changes:
- Enable GPU_SUPPORT on macOS in both CMakeLists.txt and Makefile
- Link IOReport library on macOS
- Add 'apple' to shown_gpus vendor filter
Closes#955
Reselection Issues with Pause and Follow
- I found an edge case I had missed before for the very end of the process
list (not view). For the process list pausing feature, if the process
selected was the very last item in the whole list then the selection was
moved up one. And vice versa when the list is paused and the very last
item is selected, unpausing would also move it up one item in the list.
For the process following feature the same thing would happen when
unfollowing (if not using the detailed view follow) but for the last few
items (upto halfway up the view)
- This was fixed by adding 1 to either `selected` or `start` when
required.
- When process following with detailed view. unfollowing by clicking the
process and then clicking again to close detail view threw the view
off because following was disengaged.
- Fixed with the modification to the code block that handles
following. It now also restores the selection and view when exiting
detailed view also. (I had a different fix for this but the fix for
the next issue made it irrelevant and was removed)
- If following a process with detailed view and it is one of the first
or last few items in the whole list (not view) then if you change the
reverse sort mode and then unfollow the process, instead of the
selection returning to the process that was being followed it returns
to where the process would have been if the sorting wasn't reversed.
- I struggled to figure out a fix for this but landed on reusing the
code block that locates the followed process to restore the
selection when detailed view closes.
- I actually quite like this fix since it also centers the selection
if possiple when closing the detailed view.
- The change that removed selection for detailed view following caused
an issue where if the list was paused holding up would cause the
selection to loop from the list position where the followed process
was after reaching the top of the list.
- This was fixed by adding some additional conditions before changing
the selection back to the followed process when selected is 0
- This gets rid of this issue so that when paused scrolling to the
top of the list doesn't loop back to the followed process location
- When following a process and the view was at the end of the list, if
the process died and following disengaged, a blank line would be left
at the bottom of the list until the next collection cycle.
- fixed this by making select_max not a const and adding 1 to it when
the following mode disengaged.
- When the followed process is the first process in the list, pressing
up doesn't redraw the screen and get rid of the following banner
until the next collection cycle.
- I fixed this by moving `bool changed = false` to the top of the
selection function and setting `changed` to true when reselecting
the followed process before moving back up and `selected` is 0
again.
- When following a process clicking on the banner with the mouse would
exit following mode and select the last line. I am not sure if this is
desirable or not.
- I have fixed this with an extra condition in the mouse input logic
- This can be reverted if you feel it is good for clicking the banner
to exit following in that way.
Improvements to follow mode list centering
- The followed process was centered using select_max / 2. this works
but would leave it off center by 2 if select_max was odd.
- it has been updated to use select_max / 2 if select_max is even
which has the followed process off center as close as it can be
but slightly closer to the top of list. if select_max is odd then
it uses select_max / 2 + 1 which places the process dead center.
- If follow-detailed is disabled and a process is being followed while
paused, opening or closing the detailed view doesn't recenter the
followed process.
- fixed this by relocating a line that sets update_following
Issue with tree expand/collapse when following
- When I made the change that set `selected = 0` when following the
detailed process. I did not realize that this would prevent the user
from being able to expand or collapse the tree of the process being
followed without first unfollowing it.
- I fixed this by slightly reworking the input code for
expand/collapse with some extra conditions
Misc improvement to follow mode
- Added a line in the F input handling code that sets the selection to
the followed process before exiting follow mode. This means that if
if you press F after opening a detailed view and following that it
will exit follow mode and move selected to the followed process
location instead of leaving selected as 0. This is equivalent to
pressing down after opening a detailed view when not following the
process.
- Since when following the detailed view process the selection can be
moved up or down from the followed process, the up arrow for the
select button should not be greyed out if following the detailed view
process and `should_selection_return_to_followed` is false.
- Added a comment in the selection function to explain what the
following mode code block does there.
Small optimization for Pause mode
- I moved `Config::getB("keep_dead_proc_usage")` outside of the for loop
that goes through all the current procs while paused.
The mouse mappings were only mapped as 'k' and did not change to 'K' when vim keys are enabled.
Change made so that they now map correctly
I also corrected the code that was clearing mouse mappings as it was using some uppercase letters
for keybinds that are lowercase.
fix mouse selection when following is active
When making the change to the process following PR that made moving the selection
disengage following mode, I missed a couple points in the mouse input code that was
preventing following mode from being disengaged when clicking on the detailed info box
or on the upper border of the proc box/label line.
This alteration fixes that.
Also made a change so that if the detailed view is open/opened and
following mode engages. If the detailed view process is the one being
followed then the selection is set to 0 and the detailed view buttons
are active instead of the bottom bar buttons.
added follow button to detailed view buttons
Running btop with valgrind shows that when the `atomic_wait` function is
called that a couple of million instructions being executed in a short
amount of time.
The replacement puts the thread to sleep and waits for a notification
which is much more efficient.
The mouse click coordinates where only mapped to the first 5 tabs, but
since width of each tab decresed to make room for the gpu tab, the
mapping was of the shown strings.
Introduce a named constant that is different with gpu support enabled.
Closes: https://github.com/aristocratos/btop/issues/1430
This will only allocate the log message if a log file exists and the log
level is enabled. The interface forwards it's arguments to fmt::format.
This gets also rid of some global state and hides it in the btop_log.cpp
translation unit.
Closes: https://github.com/aristocratos/btop/issues/1347
