Swap to manually calculating the mem total and usage via procfs. The usage calculation is now:
total - (free + cached + buffers + slab_reclaimable - shmem)
This follows the same usage calculation as htop. See the PR for more details.
Fixes the accuracy of the memory widget for Linux and macOS, and uses binary prefixes instead to be more accurate.
Regarding the first part, it turns out that the way I was calculating memory usage was *slightly* incorrect for a few reasons:
- Regarding macOS, it seems like the way I was determining usage (`usage = total - available`) is not the most accurate. The better way of doing this is apparently `usage = wire + active`, where `wire` is memory always marked to stay in RAM, and `active` is memory currently in RAM. This seems to be much closer to other applications now.
- Regarding Linux, this was somewhat due to two issues - one was that I should have used heim's own built-in function call to get how much memory was *used*, and the other is that when heim reads info from `meminfo`, it reads it in *kilobytes* - however, the values are actually in *kibibytes*. As such, to get the value in kibibytes, you want to actually take it in kilobytes.
While I've filed an issue for the library, for now, I'll just manually bandaid over this. See https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/deployment_guide/s2-proc-meminfo for more info.
Both changes take more advantage of platform-specific methods, and as such, the change unfortunately adds some ugly platform-specific code blocks.
Side note, Windows Task Manager apparently (?) uses binary prefixes for the values behind the scenes, but displays decimal prefixes. As such, now that we've switched to binary prefixes, it'll "seem" like the two aren't matching anymore since the units won't match, despite the values matching.
Adds the missing hide_time and battery config option to the default config and corresponding documentation.
Should probably automate the generation of this somehow tbh, though this might change when I add in-app config (soon™)
This is just a temp change, I wanted to remove it just for clarity's
sake among dependencies, and will probably add it back in the future.
For now I'll just stick to std's beef.
A large migration of documentation over to mkdocs, and some rewrites. Some stuff (install information, basic supported systems, contributors, thanks) are still staying in README.md, and CONTRIBUTING.md is essentially duplicated right now. However, stuff like configuration and key/mouse bindings are now moved to mkdocs.
Some parts are still a bit WIP - it is definitely not done (documentation never seems to be...). However, it should be "good enough" for now, and I'm much happier working with the documentation in this form than trying to scroll through a giant endless README.md file. It also works much better for adding new documentation.
Refactor to split up data collection by OS and/or the backing library. The goal is to make it easier to work with and add new OS support, as opposed to how it was prior where we stored OS-independent implementations all in the same file.
Lowers the timer for multi-digit inputs in dd.
I'm going to eventually completely rewrite the input part for the entire application though, but this will do for now.
Did not update crossterm (and tui-rs) since it seems to have resulted in a massive CPU usage increase. Also fix minor clippy error with a duplicated to_string call.
Fixes basic mode having broken click hitboxes (they were 1 unit too long in both directions). I'm pretty sure normal mode does too, but it's less noticeable due to bounding boxes.
This PR accomplishes two things:
1. This PR aims to add mount_filter to the config file. This allows a user to filter their disk widget entries by the mount name as well; this was particularly a problem in trying to address #431.
2. A slight rework of how the filter system works due to the need of being able to manage two potentially conflicting filter sources, since the disk widget will now potentially filter on both the disk name and the mount name.
In regards to the second point, the new behaviour is as such:
1. Is the entry allowed through any filter? That is, does it match an entry in a filter where is_list_ignored is false? If so, we always keep this entry.
2. Is the entry denied through any filter? That is, does it match an entry in a filter where is_list_ignored is true? If so, we always deny this entry.
3. Anything else is allowed.
This main (breaking) change is really the third point. This would mean that temp_filter and net_filter, when set to allow listed entries with is_list_ignored = false, are kinda... useless, as a whitelist in the scenario of being the only filter is kinda pointless. But hopefully this shouldn't be a problem...?
A bit of a followup to #449, this adds decimal places for values over 1GB in regards to disk usage. This affects the disk widget (for the read/write per second) and process widgets (total read, total write, read/write per second).
Removes/tweaks some really light colours that might cause issues with a white background. For example, yellow on white didn't look so great, so I adjusted the memory/rx colours for this mode.
This change adds a decimal + single digit to memory usage values over the 1 GiB threshold. Otherwise, there is no visible change.
(Note to self: implement the per-column width system soon, this change causes some values to potentially look a bit weird in mem-non-percent mode as it is if the value is really large, like 530.2GiB pushing right up against the column width, but it's currently tied to mem-percent mode. Ugh.)
Also revert a change made by accident where I switched to a decimal prefix system (GB) for memory values. This has been reverted back to a binary prefix (GiB).
For the process widget, we now sum the resource usage of the child processes on the parent entry when collapsing in tree mode.
Note that if you search to filter, and collapse, it will not sum the pruned values (values that cannot be seen). This is partly because I'm a bit lazy, and partly because I think this behaviour makes sense.
For example, let's say I search for a process with 4 child processes "AA, AB, BA, BB", with CPU usage 0.1, 0.2, 0.3, 0.4 respectively. Assume the parent process has 0 usage.
- Without filter, it sums to 1.0
- With a filter on A, it would sum to just 0.3
- With a filter on AA, it would sum to 0.1
I think this is fine because I'm treating this as summing any child that is still *visible* somehow. Summing unseen values would probably be weird as it would look like it's not adding up.
Further note that if you had, say, a child "CC" with a usage of, say, 2.0, and its parent of "AB", and you searched for CC in our above example, you would get a sum of 2.2. This is because AB is still visible by the fact that CC was the searched process, and AB must still exist (albeit faded out) in the tree hierarchy, and as such will still be displayed.
In particular, use non-binary prefixes for disk and memory usage in a process. Ideally everything is configurable by the user, but this is fine for now IMO until I can get around to doing in-app config.
Rewrite of the y-axis labeling and scaling for the network widget, along with more customization. This still has one step to be optimized (cache results so we don't have to recalculate the legend each time), but will be done in another PR for sake of this one being too large already.
Furthermore, this change adds linear interpolation at the 0 point in the case a data point shoots too far back - this seems to have lead to ugly gaps to the left of graphs in some cases, because the left hand limit was not big enough for the data point. We address this by grabbing values just outside the time range and linearly interpolating at the leftmost limit. This affects all graph widgets (CPU, mem, network).
This can be optimized, and will hopefully be prior to release in a separate change.
Adds users into the process widget (for Unix-based systems). This shows only in non-grouped modes, similar to state. Search is also supported.
In addition, a quick fix to prevent users from being in grouped mode when they tried to enter tree mode while grouped.
Two issues were highlighted as a result of using either Void Linux with disk encryption, or just disk encryption on Linux in general:
Two fixes:
1. Fixes a failed `usage()` call in the `get_disk_usage` function from failing the entire result. Now it only returns an entry with N/A results. This occurred in some distros and disk encryption setups, for example, the one for Void Linux here: https://docs.voidlinux.org/installation/guides/fde.html.
2. Fixes a potential mapping issue with disk encryption on Linux in general. Since the disk might map to `/dev/mapper/whatever`, but the I/O harvester was using another name, the mappings would not match. As such, we now also check if a symlink exists; if it does, then we take it and work out the correct path. This also fixes the disk name being wrong.
For consistency, we now don't automatically jump to the top of the list when using a sort shortcut. This behaviour already occurred with the sort menu and sorting by mouse clicks, so this is just now more consistent (and IMO less annoying, you can also always jump to the top via gg).
Clement Tsang
ClementTsang
Zeb Piasecki
pvanheus