This adds very basic GPU support for getting process usage for Intel GPUs on Linux. Note this PR does not handle memory/VRAM usage for now. The implementation is basically a clone of the AMD GPU one.
Doing this as refactoring work before I start working on supporting Intel GPUs, which are expected to share the same collection code. This should have no functional change.
This PR makes it so that we should hopefully avoid waking up NVIDIA GPUs while gathering info (on Linux only). It mostly follows how we were doing it for AMD devices, which required a bit of refactoring to get it to work with the NVML library I was using.
Note that, as mentioned above, it does not change behaviour for Windows (or other OSes that work with NVIDIA GPUs) at the moment, as I need to use some fairly OS-specific logic to get it to work on each platform from the looks of it.
Unfortunately, I no longer have any Nvidia-based machines to test with, let alone laptops with Optimus... so this may be hard to verify whether it works on my end.
This adds the ability to run with `--network_start_zeroed` or the `start_zeroed` config option, which will zero out the total network usage so that bottom starts with it at zero, rather than the total network usage reported by the OS since boot.
There was an if-condition that only ran if the sort state was false. But this would never be hit since `use_sort_table_value` always set it to false beforehand, so it would be stuck focusing on a hidden menu.
Wanted to prune Cargo.toml a tiny bit and noticed I could just use humantime to do this instead of time for logging purposes. Of course I still pull it in via Cargo.lock though.
A looooong time ago, we had to manually hijack stdio for FreeBSD to suppress a bogus stderr warning (see
#798). I filed an upstream report at GuillaumeGomez/sysinfo#875, but the real issue was actually with FreeBSD itself. From the upstream upstream bug report at https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=228432, it looks like this has been fixed since 2022 for FreeBSD 13, so I think it's safe to remove this hack at this point on our end, especially given that I don't build binaries for older than FreeBSD 13 anymore (oldest is 15 now).
This PR allows us to entirely remove the filedescriptor crate for FreeBSD since we no longer use it entirely.
I made the assumption that the max length of the comm field is 16 bytes. This is wrong though - it's now 64.
Note we may want to change it in the future to work regardless of potential length.
Because of the way I was initializing things, this caused it to report a huge spike at the start since it was doing a subtraction against 0. This fixes it by just using a hashmap (which also fixed another FIXME) and initializing the prev comparison structure on first view + skipping the first instance.
The bug would cause something like `(sd-pam)` to be missing, because what would happen is this:
1. We have something like `"123 ((sd-pam)) blah blah blah"`
2. We parse `start_paren` to be index 5.
3. We parse `end_paren` to be index 13.
4. Our `comm` is `"(sd-pam"`, and the rest of our string is `") blah blah blah"`... which is problematic, as then the rest of the parsing will be incorrect, and the entry gets rejected!
Deals with a possible issue with kitty and its `cursor_trail` functionality, where it supposedly does a bit of extra calculation work when the cursor's location after drawing is too far from the previous location when kitty is checking whether it should recalculate things for the trail, even if the cursor is hidden (at least this is my guess by looking around in https://github.com/kovidgoyal/kitty/blob/c1d507dbe8cd12830d8b97b0d350d9dc2e4d383f/kitty/cursor_trail.c). The fix is (apparently) to just park the cursor at the origin after drawing each time, and this seemingly helps as a workaround.
If this is right (and it very well could be wrong, I'm making a guess here), IMO the actual fix is that kitty should _probably_ skip cursor trail calculation work if the cursor is hidden I guess?
This change should fix the issue around `cargo install` being confused by the schema generation binary by just moving it entirely into a separate tool "crate".
Fixes a bug with cgroups v1, where the "total" if there isn't a limit set may be some crazy high number which would be interpreted as many terabytes large. This is obviously incorrect, so we should not use that as the total, and instead, use the min of the actual total and the cgroups value (so if a lower cgroup v1 value is set, we use that instead).
This PR adds ufs disk io counters on freebsd by using sysinfo. This PR also updates the Cargo.toml to use the dep syntax for the unused feature flags starship-battery and nvml-wrapper.
Comment out flags header in default config file + fix bug causing deprecated warnings if `[flags]` was uncommented even when the fields themselves were commented out.
This change moves some options to a more appropriate section for the config file. This only deprecates the old options for now, so they're still valid, but we will remove them later (maybe after the next stable release, to give warning).
One thing I want to do for the legend position options is maybe expand it, and now is the best time to do so, but for now I'm just copying it over verbatim.
The deprecated options will also emit a warning to stderr.
This PR also contains a driveby to alias `[network]` to `[network_graph]`.
Clement Tsang