Consider these 2 cases:
* A window is mapped while fading out
* A window is mapped and damaged while fading out
From the perspective of map_win_start, these 2 cases look the same. The
window will has ever_damage = true in both cases. However, map_win_start
has to distinguish between these 2 cases, because at the end of
map_win_start, window in the first case cannot have ever_damage = true,
while window in the second case should have ever_damage = true.
Currently, map_win_start always clears ever_damage in both cases
(indirectly through win_skip_fading), which causes windows in the second
case to not be rendered when they should be.
This commit move clearing of ever_damage from unmap_win_finish to
unmap_win_start, so when map_win_start sees ever_damage = true, it's
always to second case. And to make sure windows which are fading out are
still rendered, we relax the ever_damage check in paint_preprocess to
also accept windows that are fading out. (see code comment for
explanation why this is fine)
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Set WIN_FLAGS_MAPPED instead, it will be handled later. Previously, we
call map_win_start because we need the geometry of the window, which is
updated in map_win_start. Now, we get the geometry in fill_win, so
map_win_start is not needed anymore.
Eliminate a case where destroy_backend could see IMAGES_STALE flags set
on windows.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Make unmapped window events work mostly like a mapped window, except
flags set on unmapped windows aren't processed until the window is
mapped.
Hopefully this unifies some of the code paths and reduce corner cases.
Should fix#525
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
When the screen is unredirected, no window have a shadow image. So the
assertions in win_set_shadow don't hold. But in that case, we don't want
to add damages anyway. So we put them behind a check of whether the screen
is redirected.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
- Fix non-critical memory-leak in `picom.c` and `options.c` where we
don't free all allocated memory before dieing.
- Explicitly allocate new branch in `c2.c` to silence false-positive
memory-leak.
Of course, we still use GLX, so we can't completely remove Xlib yet. But
this removes all Xlib uses outside of the backends.
This drops support for COMPOUND_TEXT Xorg strings, so people how wants
multilingual support has to use UTF8, which should be fine since most of
the applications support that.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Partially revert 32754b0262.
We start with mapping the window (`win_process_update_flags()`). Then check
if focus has changed and process focus updates. Finally refresh stale images
(`win_process_image_flags`) because rules based on focus may have invalidated
them or require them to be created.
Fixes#465 with the following rule:
```
shadow-exclude = [
"focused" != 1"
]
```
**Work-in-Progress**
Add `dual_kawase` to configuration and argument parsing. Allow `kawase`
for backward compatibility. Add `--blur-strength` parameter for
blur-method `dual_kawase`.
Update documentation to reflect the new blur-method and parameters.
All names start with underscore are reserved in C.
Also just testing out the language server's renaming feature.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
This is a workaround for #301. This doesn't fix the bug, but by
allocating X resources much less frequently, this should make the
bug almost never happen.
Also, it might generally be a good idea to not create/destroy X
resources so often. (XFixes Region accounts for >99% of the resource
creations/destructions)
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
xrender-sync-fence might be enabled after init because of driver
detection. Instead of creating it on demand, just always create it as
this is simpler.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
We use the __GL_YIELD=usleep workaround when we detect the NVIDIA
driver, so we could use glFinish without the NVIDIA driver taking all
the CPU.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
This reverts commit 04fe4a76b2.
This brings back the previous incomplete fix attempt for #299. See the
commit message of the revert for why it's incomplete.
A different fix is then attempted, see commit xxxxxxx for how that fix
works.
However, the second fix is incomplete by itself as well. The reason is
that i3 reparent the real window to the frame first, before destroying
the placeholder client of that frame. So briefly, that frame would have
2 client windows. And the frame is mapped before the placeholder is
destroyed. So even though we only call win_recheck_client when/if the
frame window is mapped, it can still be called when there are 2 client
windows, it would pick up the wrong client window in that case.
So what we need is to combine both fixes.
The second fix makes sure we are up to date on the client window
information when we starts to listen for events on the frame window;
while the first fix would keep us up to date afterwards.
This revert also includes a fix for assertion failure raised in #371
See #299 for root cause of the bug.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
These two flags are intended for subtly different things. I can probably
justify the distinction, but it's definitely difficult to explain. And
there is no obvious benefits to keep them separate.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Commits reverted:
b652e8b58dbdf809d039e9ab970989481ac54f67
While those commits themselves could be useful even if they don't
completely fix#299, they come with the risks of introduce more
unforeseen bugs, which outweigh their benefit, so revert them.
A brief explanation of their problem:
The fix hinges on the destroy event of the client window to work.
However, the client window could be destroyed so fast, before we even
handle the map, or even the create, event of the frame. And we won't be
listening for substructure events on the frame at the point the client
window is destroyed. Thus completely miss the client window destroy
event.
Because of the inherent racy nature of Xorg, this approach is really
difficult to make work.
Fixes#371
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
So we can pick up destroy notify of the client windows. Because we need
the destroy notify of the client window to trigger a client window
recheck on the frame window. This is to handle the case where a frame
window is later reused, and the case where multiple client windows are
momentarily attached to the same frame[1].
This will also cause events previously unseen to be generated. Most of
the events would be ignored, because they don't have a corresponding
managed window. Change has been made to ev_reparent_notify, so it would
handle the case where a client window is reparented from one frame to
another[2]. Previously such events are not received by picom.
Related[1]: e9ab970989
Related[2]: 481ac54f67
Related: #299
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Previously, root ConfigureNotify is handled immediately, by resetting
the backend, which in turn releases all the window images. This puts all
the windows into a state where they don't have images attached, which
they really should be in when the screen is redirected.
(To expand a little, a window without images should only exist if:
* It's an unmanaged window.
* Screen is unredirected.)
Normally, this kind of window could be fine, as the next render phase
will re-acquire images for them. However, if a window in this state is
destroyed with fading enabled, then the render phase won't try to
acquire images for it, causing it to go into the main rendering function
without images attached, and trigger an assertion.
This commit delays the handling of root ConfigureNotify until the render
phase. This way, the images will be immediately re-acquired after they
are released, thus prevent this problem from happening.
Also adds a testcase for this.
Fixes#357
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
x_print_error aborts the program when it sees a redirect_subwindow
failure. A function called x_print_error really shouldn't cause the
program to terminate.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
In configure_root, the root pixmap is freed by destroy_backend, and is
never acquired again.
Usually this doesn't seem to cause any problems, maybe because we
immediately receives a root damage notification. But under rare
circumstances we ended up not showing the wallpaper.
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Don't set win::flags to 0 in paint_preprocess, as that can break
assertions in other places. And there is also no need to do so as
win::flags are properly tracked regardless of the backend used.
Fixes#270
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Currently picom fails at register_cm(), which is before
print_diagnostics(), so we don't see anything.
This commit makes sure we keep going when there is another compositor
already running. However, in this case we need to be careful to not take
the overlay window, otherwise a poorly implemented compositor (for
example, picom) will stop working.
Also restructured the target window initialization logic and added some
comments explaining the logic.
Fixes#333
"notify" is the only wintype that has a name inconsistent with its Xorg
name. So fix that.
Compatibility code is added as well, so the old "notify" wintype name
still works.
Fixes#323
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
Unredirect the screen when some window sets _NET_WM_BYPASS_COMPOSITOR
to 1, but ignore the case where some window requests the screen to be
redirected by setting _NET_WM_BYPASS_COMPOSITOR to 2.
Closes#267
Signed-off-by: Yuxuan Shui <yshuiv7@gmail.com>
