| Age | Commit message (Collapse) | Author |
|---|
|
|
|
This corrects my (wrong) assumption that the timer coordination thread
will only fail if there's an error in the code, and in that case any
attempt to recover is futile. It turns out that the thread does fail
recoverably in one notable case: when running in the non-threaded RTS,
registerDelay fails immediately. And we probably still wish for xmobar
to support the non-threaded RTS.
One way to solve this issue is to add a bunch of #ifdefs and compile the
code only in the threaded case. This would double the number of
configurations that need to be tested, though.
Instead, let's make the code robust against all kinds of exceptions in
the timer coordination thread, and get non-threaded RTS support for
free.
|
|
The first implementation assumed all timers (monitors) are fast and
frequent (which happens to be the case in my configuration). This meant
that a single on-line weather monitor could block the entire xmobar
instance for a long time due to the refresh pausing (meant to reduce
power consumption).
This commit attempts to fix that by limiting the refresh pause time and
using the old periodic sleep method for these slow timers (monitors).
|
|
xmobar currently runs every monitor in its own thread. Monitors that do
periodic updates simply sleep and loop. This unfortunately leads to
these threads coming out of sync, and xmobar ends up waking up and
redrawing for every periodic monitor. In my case, that is 7 times per
second, which is enough for xmobar to be at the top of "top" with more
than 1% CPU usage, and to have a noticeable impact on battery life.
This commit adds a central timer coordination thread which makes sure
that periodic updates happen together and that we only redraw once
they're all done.
Together with PR #409, I managed to lower the idle power draw of my
laptop from 4W to 3W.
|
|
A preparation for timer coalescing: tenthSeconds is just a sleep whereas
doEveryTenthSeconds enables using a central timer and waiting for all
monitors to update before refreshing the window. This commit is just a
simple refactor, the actual timer coalescing code comes later.
|
|
src/Xmobar/X11/Draw.hs:51:11-12: warning: [-Wunused-matches]
Defined but not used: ‘wr’
|
51 | drawInWin wr@(Rectangle _ _ wid ht) ~[left,center,right] = do
| ^^
src/Xmobar/App/EventLoop.hs:50:1-36: warning: [-Wunused-imports]
The import of ‘Xmobar.X11.Events’ is redundant
|
50 | import Xmobar.X11.Events(nextEvent')
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
|
The calculations are based on the assumption that current_now (or
power_now) are always positive. However, power_supply documentation in
the kernel sources say nothing about it, and so some drivers provide a
signed value (e.g. bq27xxx_battery_current).
Discovered and fixed on ac100/paz00 netbook.
|
|
|
|
|
|
|
|
|
|
|
|
This reverts commit 1f1f0bd8b811740c84215f9ed4fa5ebd8309a990.
|
|
This reverts commit efb6d6817c092fe08e9b0f1b8a17bddd29d97cdb.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Exposed via -f with_freebsd flag, uses sysctl to query battery status.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The first option applied here is a default value for a field that's not always
reported to be there, namely the 'weather' field. It now defaults to saying
"normal" instead of displaying an empty string.
|
|
|
|
|
|
|
|
|
|
The DateZone plugin calls `getTimeZoneSeriesFromOlsonFile` using the
hard-coded path /usr/share/zoneinfo. While that may work just fine on
most Linux distros, it does not work on NixOS since that directory is
always locates somewhere under /nix/store.
Based on mild research, it seems the environment variable TZDIR is
commonly set to the absolute path to `zoneinfo` (but without a trailing
slash).
This change modifies the DateZone plugin to first try getting
the zoneinfo path from the TZDIR environment variable, falling back
to the hard-coded path /usr/share/zoneinfo
|
|
|
|
- use strict ByteString as the Lazy version of readFile allocates a 32k
buffer even though we usually need much less (isUp needs a few bytes)
- refactor NetDev datatype and use unsafeInterleaveIO in isUp to avoid
reading the operstate file entirely if we're not interested in that
device
- postpone ByteString unpacking in netParser to shave off some cycles,
and avoid ByteString unpacking in isUp entirely
On my system with 8 network devices (and more if docker is up), this
seems to reduce xmobar's CPU usage noticeably. I have two "Run Network"
in xmobar configuration, for eth and wlan, so without these changes,
xmobar would evaluate isUp 16 times a second, and each evaluation would
allocate a buffer for the IO Handle and then another buffer for the lazy
ByteString readFile. Now it only does isUp once for every device I'm
interested in, and the only large buffers allocated are the IO Handle
ones (getting rid of these isn't worth the code complexity).
|
|
|
|
|
|
* add filtering option for Net devices
* relate to comments
* upd readme
* add few more words to readme
|
|
This time taking into account that ~/.config/xmobar could be populated
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
