- Fixed building on some systems where
c_chardoes not correspond toi8. This was a waybackend bug, so it can be fixed my merely bumping our dependencies over here.
- Fixed a big issue when animating wallpapers that was leading to ghosting. Basically, we were still assuming the fourth channel was irrelevant, despite the fact we moved from xrgb to argb. And so the previous frames would blend into the current ones.
- IMPORTANT:
--format xrgbis no longer supported. Use--format argbinstead. However, this is now the default, so, if you want, you can simply not set anything instead. We still accept--format xrgbso as to not break any configurations. Previously, the daemon automatically would check to see whichwayland_shmformat was the most convenient for us to use. However, many people have had weird bugs when using any format other thanxrgb. Therefore, we have made that the default, and only change it when the user specifically asks for it. swww querynow returns information in the following format:<namespace>: OUTPUT: SIZE, scale: SCALE, currently displaying: IMAGE_OR_COLOR- we are creating the socket in a slightly different location now:
${XDG_RUNTIME_DIR}/${WAYLAND_DISPLAY}-swww-daemon.<specified namespace>.socket
- every command (except
clear-cache) now accepts a-n|--namespaceargument. This namespace will then be appended toswww-daemonto make up the wayland namespace of the daemon's layer surface. - as a convenience, every command (except
clear-cache) also accepts a-a|--allflag, that lets the user send commands to all namespaces at once. - support
avifdecoding throughdav1d, with the--features=avifcompilation option. - we finally support static
SVGs! swww-daemonnow correctly renders static images with transparency.swww clearnow accepts colors with an alpha value.
- Fixed a division by 0 when animating with no frames
- Fixed destroying wayland buffers too early
- Better fractional-scale implementation (we shamelessly tried copying the
swaybgimplementation). This will hopefully prevent newer issues withhyprland.
@shumvgolove fixed a very dumb mistake I made when detecting the output current mode. This could potentially break a few setups, so I am releasing it now in a minor release. Apologies for the inconvenience.
Another small update with very important bug fixes.
- We prioritize fractional-scale events over preferred-buffer-scale ones. This should fix many of our long-standing issues with fractional scaling. Thanks a lot to @AryanRarestand who hard-coded a solution that led me to find the problem.
- do not accept
wl_outputmodes that not current, as they are considered deprecated by the protocol. - No longer spaming standard output with warnings that we are ignoring
preferred-buffer-transformevents.
A small update with important bug fixes.
- Make
swwwwork on systems wherec_longis is not mapped toi64, by @triallax - Fix fractional scaling, by @cyuria
- @nullcubee and @iynaix helped me understand fix the nix build (thanks, guys)
- using
tiny-benchinstead ofcriterionreduces our dependencies indebugbuilds
Okay, this one's a doozy.
- new
--layeroption forswww-daemon - new
--resize stretchoption, by @rexept - cache now stores the image filter, by @kerty0
- we can now read images from standard input, by @iynaix
- new
swww_randomize_multi.shexample script, by @cplir-c
Note: because there were many changes between the previous version and this one, some of these things may resurface as problems. Unfortunately, many of these issues I either cannot reproduce in my system, or the whole application has changed so much that testing them in isolation would be challenging.
- fixed output transforms
- animation timings could be slightly off
- examples scripts are more POSIX-friendly, by @Fawn
- fixed memory leak and halted animation when connecting/disconnecting monitor, by @kerty0
- some stuff wasn't working when files were behind symlinks, fixed by @mendess
- fixed connecting to Wayland via WAYLAND_SOCKET, by @mstoeckl
- reimplemented the communication from the client and the daemon from scratch, allowing for much more efficient data transfer
- the above was then made type-safe by @rkuklik
- using a custom log implementation that looks much nicer
- rewrote the transitions code to be more SIMD friendly -- will improve
animation stuttering in some systems. Furthermore, and more importantly,
the animations are now fast enough that we no longer need
rayonas a dependency. - more generally, we have completed nuked the multi-threaded code.
swww-daemonnow only spawns a single thread, letting us get rid of everyArcwrapper, and other annoying, fragile, synchronization code. - we are using
rustixinstead of thestdas much as possible, for lower syscall overhead - remove dependency on
spin-sleep flake-lockby @JohnRTitorwaybackendbased implementation -- this is much more lightweight thanwayland-rs, and lets us make the daemon very resource efficient.- can now use hexcodes instead of image path, which we detect automatically, by @flick0
- fixes to some internal documentation by @JaKooLit, @Axlefublr, @filip-rs, @rexept, @Not-Glunk
- @rkuklik did a big refactor on the project's overall organization
- BREAKING CHANGE: dropped the
initsubcommand for the client, by @WhyNotHugo
Note there have also been other contributions that were rendered obsolete in the meantime, since so much has changed from the last released version (in fact, the last release was roughly a whole year ago!). The daemon codebase itself has gone through 3 very large refactors, and now I am finally mostly satisfied with how it looks. Unfortunately, this does mean some things will inevitably break. I apologize in advance for any inconveniences.
On the other hand, I believe the rewrite has fixed some old-standing bugs we have. Though those will have to be tested again on systems that can actually reproduce them reliably.
I also apologize if I forgot to mention anything/anyone in the above summary. If I missed your contribution, please feel free to open a PR adding it.
This is mostly just fixes and small improvements.
- fixed wallpaper never setting
configuredto 'true' - fixed fractional scaling rounding incorrectly
- fixed scaling for vertical monitors (thanks, @AhJi26)
- fixed the annoying black screen on login issue (finally)
- add --no-cache option to
swww-daemon, by @lucasreis1
- specialized transition for
--transition-type none(previously it was an alias tosimplewith special values) - remove an extra call to
thread::sleepwhen loading the cache - no longer using an
event_fdto wake up the main thread
Fractional Scaling is finally implemented! Woooo!!
That also fixes a nasty problem some people were having with fractionally scaled outputs.
Apologies for spaming small releases like these in short order. Hopefully this will be the last one for a while.
Quick release to fix a scaling error that might affect a lot of people.
- fix wrong scale calculation
- deleted leftover
/proctraversal code in the client - no longer setting nonblocking mode for daemon socket, since we are already polling it
- better IPC structs between client and daemon
- fix stack overflow on some systems, by @iynaix
- make sure we start the daemon even when the socket file already exists
- fix build in 32bit x86, by @Calandracas606
- fix image resize when image is larger than monitor
- fix transitions performance that had regressed from versions 0.8.*
- added SIGHUP to the list of signals we catch to exit properly
- allow
swwwto run on a nested wayland environment, by @Fuyukai.
- we no longer traverse
/procto detect whether the daemon is running, we just try pinging it instead, by @Fuyukai - many internal refactors:
- client now sends images in the format requested by the daemon (previously we were transforming the images in the daemon itself)
- simplified the daemon's transitions
- using
bitcodeinstead ofrkyvfor serialization - using
rustixinstead ofnixfor unix stuff - The Big One: we've eliminated our dependency on
smithay-client-toolkit, now make calls directly towayland-client. This gives us more control over our code for the price of a little extra verbosity.
My bad everyone, 0.9.0 wasn't loading the cache, so I am publishing this quick
fix.
MSRV is now 1.74.0.
swww init is now considered deprecated. Use swww-daemon instead. To run it
in the background, simply do swww-daemon &.
- fix the
let_underscore_lockerror. Note that all 0.8.* will probably no longer build with newer Rust versions due to that error. By @akida32 - fixed webp and gifs that are only a static image
- fixed busy waiting for WlBuffers to be released. This was a big one, and it involved rewriting a ton of stuff. We've implemented our own memory pool and are using frame callbacks to know when to draw now. A big thanks to @YaLTeR, @jeLee6gi for their patience and help with debugging and testing this thing.
- properly removing all cache contents on
clean-cache - always center images that are larger than the monitor
- animations no longer overlap when sending two animated images in succession
- fix randomize script trying to use directories as images. Fix was suggested by @MRSS02
- waiting for child swww process when loading the cache, preventing zombie processes
- waiting for daemon initialization before certain requests. By @musjj
- New, better compression function implementations. We are now using some SIMD code to accelerate the frame compression functions, leading to some nice speedups in some cases. Thanks to @Akida31 for their help in verying my unsafe code.
- We are using 3 channel color formats for some nice perf and memory improvements. Unfortunately, it seems to not work for some people on some notebooks (see Known Issues).
- Implemented a way to force the use of a specific wayland_shm format, as a workaround for Known Issues. Also went ahead and implemented some cli options for the daemon.
- Support for animated pngs
- Support for animations when piping images from standard input
- Fps is now a
u16, so we can support newest monitors framerates - Created a
restorecommand to manually restore the cache. By @musjj - GitHub actions! Big thanks to @MichaelOultram!
Some people are having some problems with the 3 channel color formats (see issue
#233). Currently, initializing the daemon with swww-daemon --format xrgb is a
workaround to that.
NOTE ALL 0.8.* VERSIONS WILL PROBABLY NO LONGER BUILD WITH NEWER RUST VERSIONS.
This is because Rust promoted let_underscore_lock to a hard error, which
wasn't the case at the time we've published these versions.
I've changed my username from Horus645 to LGFae. This means you will have to
update the remote url to https://github.com/LGFae/swww. Anyone who also has
direct links to the old address should update them.
I've done this because mostly to make it more professional looking. I've considered using two github accounts instead, but that would involve setting up multiple ssh keys on my machine and included other complications. I deeply apologize for the inconvenience.
- update MSRV in README.md, by @micielski
- implemented a
--no-cacheflag forswww init - fixes to the build script, by @m4rch3n1ng
- client waits for daemon to be ready on
swww init - more accurate image fit implementation
- added MSRV to Cargo.toml, by @akida31
- fix some documentation typos and inacuracies
- fix timings for transition frames
- fix capacity of Vec in
image_padfunction, by @MichaelOultram - implement animated WebP Support, by @MichaelOultram
- some other memory optimizations, by @MichaelOultram
- implement
clear-cachecommand - we also automatically clear the cache from old
swwwversions now!
Also, we have updated all crates versions so that they all match. This should
help some package maintainers that were having difficulty setting swww up for
e.g. Debian, I believe.
Pretty a much a near-exclusive bug fix release:
- Fixed
swww clearcausing the daemon to exit - The cache is once again being correctly load during
swww init - Fixed glitches happening to animated gifs (frames were being loaded in the wrong order)
- Fixed
swww-daemonsometimes not drawing to the whole screen (forgot to set the exclusive zone to -2) - Fixed an issue where the daemon would hang if multiple images were sent in quick succession
--transition-type none, which is an alias to--transition-type simple--transition-step 255
I have changed the way we are caching the images / animations. I would recommend users to delete the previous cache directory once they install this new version:
rm -r $XDG_CACHE_HOME/swww # OR
rm -r $HOME/.cache/swwwWe now sync everything automatically, every time. So we've eliminated that flag.
Update to sctk 0.17:
Updating to sctk 0.17 unlocked
many, many improvements:
- We managed to ditch and extra
clonewhen calculating the transitions - We now draw the images directly into the wayland buffer, instead of having to send an intermidiary buffer through a channel.
- Because of the above, we can now send from the client a
bgrimage, instead of abgraone. This lets us seriaze and write roughly 3/4 of what we were doing previously.
We have changed our serialization strategy from
serde to
rkyv. This lead to even further memory usage
reductions, since rkyv does not use an intermidate buffer to deserialize its
structures.
We have reworked the way we do synchronization between monitors. It should sync all monitors animations automatically.
- New transition,
fade, that is essentiallysimplewith beziers (@flick0) invert_yflag (@flick0)- New option to resize to fit, padding only what was left (@SignalWalker)
- @RiedleroD fixed a type :)
With all of the changes above, I've managed to reduce memory consumption almost by a factor of 3. The price to pay was a full rewrite of the wayland implementation part of the daemon, a partial rewrite of the way we do ipc, and all the code adaptions necessary to make all that work.
Unfortunately, because I had to rewrite so much stuff, it is possible that old bugs will resurface. I've tried my best to test and validate it with every thing that blew up in the past, but it is probably inevitable that some stuff slipped by. Apologies in advance, and keep this in mind when upgrading.
Fixes:
- Missing
/when using$HOME/.cache/swww, by @max-ishere --transition-stepwithsimplehas saner defaults- correctly splitting outputs argument with ',', by @potatoattack
Improvements:
- we only send the image after we've finished processing the whole animation. This diminishes the weird lag that can sometimes happen when sending a large gif.
- sending a status update to systemd when daemon has initialized, by @b3nj4m1n
Internals:
- we have benchmarks for our compression functions! This will be useful for later once I start poking things to see if I can make them more efficient.
Unfortunately, there are a few bugs people have reported that I still haven't been able to fix. Sorry about that.
In any case, the immediate plan for the future is actually to update sctk to version 0.17.0. I am actually studying the possibility of using sctk with wgpu, since that should bring many improvements (most noticeably, we would be able to store an rgb vector, instead of rgba, so we could potentially cut memory usage by 3/4. In theory, of course, I still have to actually measure it to see if there's any difference. It will probably still take a while).
Improvements:
- Images (and animations) are now cached at
$XDG_CACHE_HOME#65 - We now have man-pages! They must be installed manually in your system by
moving the files in
doc/generatedto the appropriate location. Typically, you can figure out where that is by runningmanpath. The scriptdoc/gen.shwill create the above directory and all the relevant manpages. Note that the script depends onscdocbeing installed in the system. Have a look at it for more details. - We now also have automated spell checking. This let us fix a number in typos in our documentation, both internal and user-oriented.
- New option for
swww-img:--sync. This syncs the animations in all your monitors. Note that all monitors must be displaying animations in order for it to work.
Internal:
- Integration tests are not run by default. You must now use
cargo test -- --ignoredto run them. This will make it possible for some people (like the ones trying to packageswwwat Nix) to run some of the tests in a sandboxed environment where they don't have access to the wayland server. If anyone is interested in running all tests, they can do that withcargo test -- --include-ignored.
Improvements:
- you can now use absolute screen coordinates with
--transition-pos(@flick0)
Fixes:
swww querynot returning the image being displayed- document
--no_resizeand--fill_coloroptions forswww img - reading img from stdin (now with a proper integration test to make sure it doesn't happen again) (#42)
Internal:
- fixed
tests/integration_tests.rscalling the wrongswww-daemonbinary
BREAKING CHANGES
- ATTENTION, PACKAGE MAINTAINERS -
swwwis now composed of two separate binaries:swwwandswww-daemon. Both must be installed on the user's system in order forswwwto work correctly. Doing this allowed for major improvements in terms of overall memory usage, among other things (#52).
Improvements:
- separate client and daemon (see above).
- we don't try to animate
giffiles that have only one frame - we can read images from stdin (not this does not work for animated gifs; we simply display the image's first frame) (#42)
--no-resizeoption (pads the outer part of the image withfill-color) (#37)- new transition:
wave, by @flick0 - reading image format properly (instead of using file extension) (#74)
Fixes:
- fixed panic with on gif that had identical frames (#68)
- fixed panic with fractional-scaling (#73) (by @thedmm)
Non-breaking Changes:
- @flick0 changed the default
transition-step
Internal:
- Many improvements to the README.md (@aouerfelli and @flick0)
BREAKING CHANGES
transition-speedno longer exists. Now, speed is controlled through a bezier curve (transition-bezier), and duration (transition-duration) flags (note this also applies to the env var, SWWW_TRANSITION_SPEED). A warning was added when we detect the presence of the SWWW_TRANSITION_SPEED environment variable. This warning will go away in the next release, it is only there as a means of making sure everyone knows the variable has been superseded, and having it in your configs no longer does anything.
Improvements:
- New grow transition. Grow and outer transition now accept a --transition-pos command line argument. By @flick0.
- Transitions
growandouternow both work with bezier curves (see breaking changes, above). This allows for finer control in animation speed than before. Also by @flick0. - Very slightly faster decompression routine
BREAKING CHANGES:
swww querynow formats its output as<output>: ..., instead of<output> = .... This will break your scripts if you relied on the output's format.
Improvements:
- Fixed
swwwgetting stuck on a futex when a new monitor was connected (#26) - New
wipetransition by @flick0 - Several small code improvements by @WhyNotHugo
- Typo fix (@thebenperson)
- Check to see if daemon is REALLY running when we see that socket already exists (#11)
- Fix dpi scaling (#22)
- Fixed #13.
- Improved error message when daemon isn't running (#18)
- Fixed regression where the image was stretched on resize (#16)
-
implemented the new transition effects
-
refactored socket code
-
refactored event loop initialization code, handling errors properly now
-
BREAKING CHANGE: we are using fast_image_resize to resize our images now. This makes resizing much faster (enough to smoothly play animations before caching is done), but it makes it so that the
GaussianandTrianglefilters no longer exist. Furthermore, the filtersBilinearandMitchellwere added. -
deleted previously deprecated
init -iandinit -coptions
- Limited image formats to:
gif,jpeg,jpeg_rayon,png,pnm,tga,tiff,webp,bmp,farbfeld - Bumped rust edition to 2021
- Our custom compression is now even faster
- I did a rewrite of the way the code that handled animations was structured. This made caching a LOT faster, but it incurs in more memory usage, since we spawn an extra thread to make a pipeline. That said, since this also greatly simplified the code itself, I considered it an overall positive change.
- Fixed a bug where the animation wouldn't stop until it had processed all the frames, even when it was told to.
- Setting a custom names and stack sizes to our threads. The custom name will help in debugging in the future, and the custom stack sizes lets us push the memory usage even lower.
- Did all the preparatory work for us to start writing new transition effects. Ideally they should come in the next version, which should hopefully also be our first release (since then I will consider swww to be pretty much feature complete).
Using unsafe to speed up decompression.
Also, swww init -i and swww init -c may now be considered deprecated.
It was originally created to bypass swww init && swww img <path/to/img> not
working. Now, however, it seems to be working properly. In hindsight, it was
probably already working for a while, but I failed to test it properly and
thought it was still a problem.
The swww init -i and swww init -c options shall remain for now, for
compatibility and just in case a regression happens. Once I am confident
enough, they will be eliminated (that will let me erase around 50 lines of
code, I think).
Initial release.