-
Notifications
You must be signed in to change notification settings - Fork 5
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Update dependencies and add an example #1
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks, for working on it. Some nits.
src/lib.rs
Outdated
// Wayland-backend should probably document some guarantees here to make this sound, | ||
// but this is unlikely to change |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
the last part is not clear, what is unlikely to change?
src/lib.rs
Outdated
// In theory, we would need to call the process_events handler on fd. However, | ||
// we know that Generic's process events's call is a no-op, so we can just handle the event ourselves |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I find it hard to understand what you want to say with this comment, since it's not attached to anything.
read_guard: Option<ReadEventsGuard>, | ||
fake_token: Option<Token>, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We should state what and why it's here.
src/lib.rs
Outdated
// If getting the guard failed, that means that there are | ||
// events in the queue, and | ||
// The readiness value is never used, we just need some marker | ||
// If getting the guard failed, we need to process the events 'instantly' | ||
// tell calloop this |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The formatting of the comments on the patch seems off in general. Could you properly format them?
src/lib.rs
Outdated
// Otherwise, drop the guard if we have it, as we don't want to do any | ||
// reading when we didn't get any events |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it's always dropped? There's no otherwise.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Well it's always dropped, sure. But it's not always dropped by us - if the read
call happens, that takes ownership and drops it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If we call the read
it still sort of dropped by us. The comment is a bit misleading imho, because read
should take ownership, no?
src/lib.rs
Outdated
if let Some(guard) = guard { | ||
if let Err(WaylandError::Io(err)) = guard.read() { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
You can probably merge these two and probably, but I don't really care.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't really see how without let chains, which I don't think are stable yet
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've suggested in the diff how to merge them.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I see. It's not the code style I would use, but I'll apply it anyway.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I mean, it's less nested and that's how usually written. More readable code should be preferred.
Looks good from what I can say. Some formatting diff, which you may want to adjust. diff --git a/src/lib.rs b/src/lib.rs
index 8527e18..3384192 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -79,11 +79,11 @@ impl<D> WaylandSource<D> {
pub fn new(queue: EventQueue<D>) -> Option<WaylandSource<D>> {
let guard = queue.prepare_read()?;
let fd = Generic::new(
- // Safety: `connection_fd` returns the wayland socket fd,
- // and EventQueue (eventually) owns this socket
- // fd is only used in calloop, which guarantees
- // that the source is unregistered before dropping it, so the
- // fd cannot be used after being freed
+ // SAFETY: `connection_fd` returns the wayland socket fd, and
+ // EventQueue (eventually) owns this socket fd is only used in
+ // calloop, which guarantees that the source is unregistered before
+ // dropping it, so the fd cannot be used after being freed
+ //
// Wayland-backend should probably document some guarantees here to make this sound,
// but we know that they are unlikely to make the queue have a different socket/fd
// - there is no public API to do so
@@ -145,10 +145,10 @@ impl<D> EventSource for WaylandSource<D> {
// Take the stored error
std::mem::replace(&mut self.stored_error, Ok(()))?;
- // Because our polling is based on a `Generic` source, in theory we might want
- // to to call the process_events handler on fd. However,
- // we know that Generic's `process_events` call is a no-op, so we can just
- // handle the event ourselves
+ // NOTE: Because our polling is based on a `Generic` source, in theory we might
+ // want to to call the process_events handler on fd. However, we know
+ // that Generic's `process_events` call is a no-op, so we can just
+ // handle the event ourselves.
// Dispatch any pending events in the queue
Self::loop_callback_pending(queue, &mut callback)?;
@@ -188,33 +188,31 @@ impl<D> EventSource for WaylandSource<D> {
self.read_guard = self.queue.prepare_read();
match self.read_guard {
Some(_) => Ok(None),
- // If getting the guard failed, that means that there are
- // events in the queue, and so we need to handle the events instantly
- // rather than waiting on an event in polling. We tell calloop this
- // by returning Some here. Note that the readiness value is
- // never used, so we just need some marker
+ // If getting the guard failed, that means that there are events in the queue, and so
+ // we need to handle the events instantly rather than waiting on an event in polling.
+ // We tell calloop this by returning Some here. Note that the readiness value is never
+ // used, so we just need some marker.
None => Ok(Some((Readiness::EMPTY, self.fake_token.unwrap()))),
}
}
fn before_handle_events(&mut self, events: calloop::EventIterator<'_>) {
- // It's important that the guard isn't held whilst process_events calls occur
+ // It's important that the guard isn't held whilst process_events calls occur.
+ //
// This can use arbitrary user-provided code, which may want to use the wayland
- // socket For example, creating a Vulkan surface needs access to the
- // connection
+ // socket. For example, creating a Vulkan surface needs access to the
+ // connection.
let guard = self.read_guard.take();
if events.count() > 0 {
// Read events from the socket if any are available
- if let Some(guard) = guard {
- if let Err(WaylandError::Io(err)) = guard.read() {
- // If some other thread read events before us, concurrently, that's an expected
- // case, so this error isn't an issue. Other error kinds do need to be returned,
- // however
- if err.kind() != io::ErrorKind::WouldBlock {
- // before_handle_events doesn't allow returning errors
- // For now, cache it in self until process_events is called
- self.stored_error = Err(err);
- }
+ if let Some(Err(WaylandError::Io(err))) = guard.map(|guard| guard.read()) {
+ // If some other thread read events before us, concurrently, that's an expected
+ // case, so this error isn't an issue. Other error kinds do need to be returned,
+ // however
+ if err.kind() != io::ErrorKind::WouldBlock {
+ // before_handle_events doesn't allow returning errors
+ // For now, cache it in self until process_events is called
+ self.stored_error = Err(err);
}
}
}
@@ -223,16 +221,16 @@ impl<D> EventSource for WaylandSource<D> {
fn flush_queue<D>(queue: &mut EventQueue<D>) -> Result<(), calloop::Error> {
if let Err(WaylandError::Io(err)) = queue.flush() {
+ // WouldBlock error means the compositor could not process all our messages
+ // quickly. Either it is slowed down or we are a spammer. The error is
+ // not critical on its own and we should simply retry later.
if err.kind() != io::ErrorKind::WouldBlock {
// in case of error, forward it and fast-exit
log_error!("Error trying to flush the wayland display: {}", err);
return Err(err.into());
}
- // WouldBlock error means the compositor could not process all
- // our messages quickly. Either it is slowed
- // down or we are a spammer. Should not really
- // happen, if it does we do nothing and will flush again later
}
+
Ok(())
}
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The patch unfortunatelly doesn't work and busy loops inside the new calloop logic.
With the calloop's patch sctk works now. Were you able to verify that your project works as well and the issues are really resolved? I'm not sure about the example here, since it adds extra maintainance and the calloop wayland source is used within sctk, so it's not that hard to test with it. And I tested with the sctk because it has more complex configurations with the calloop. |
src/lib.rs
Outdated
let guard = queue.prepare_read()?; | ||
let fd = Generic::new(guard.connection_fd().as_raw_fd(), Interest::READ, Mode::Level); | ||
let fd = Generic::new( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hm, what's the reason we try to acquire the guard when creating the source? I know that it was like that before, but what is the reason, @elinorbgr ? We won't do any polling here, so it's a bit weird, just to check that the FD is alive?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In this repository, it's entirely because wayland-backend
chooses not to expose the fd through any mechanism other than the guard, as far as I can tell. The fd
should still be valid any time the event queue (and hence the Connection
) is around, although we're not given that guarantee.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Indeed, though Smithay/wayland-rs#656 is adding the necessary APIs
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yep, Smithay/wayland-rs#656 lets us do this in a clean and safe way with simply let queue = Generic::new(queue, Interest::READ, Mode::Level);
. Then using self.queue.file
to access the contained EventQueue
.
https://github.com/Smithay/client-toolkit/pull/403/files#diff-8d59229f4c6aea492f0cbceaafc305fcade8dbe513b86e6466feed5a0b2b3107 has a version doing that.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We'd need to wait for a patch release, but I'd rather do so.
Not having an example makes for a very poor development experience. If the recommended way to test is to hope that sctk doesn't have conflicting dependencies, and test there, then I would expect the outcome to be that testing won't occur. Adding the example allows this crate to be tested by itself. It isn't completely necessary, though |
Maybe we should change the way example works by creating one more thread and trying to poll on it as well? So we could try to test the fallback code path? I'm just used to having all the relevant crates around so testing is very simple, but I understand that it might not be the case for everyone. |
FYI calloop |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
And the docs/etc should be resolved.
src/lib.rs
Outdated
/// `queue` must be from the connection `Connection`. | ||
/// This is not a safety invariant, but not following this may cause | ||
/// freezes or hangs | ||
#[must_use] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
must_use
is not required anymore.
src/lib.rs
Outdated
|
||
Ok(PostAction::Continue) | ||
})?; | ||
// Safety: We don't replace the |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This comment looks out of place.
src/lib.rs
Outdated
if let Some(guard) = guard { | ||
if let Err(WaylandError::Io(err)) = guard.read() { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've suggested in the diff how to merge them.
src/lib.rs
Outdated
// WouldBlock error means the compositor could not process all | ||
// our messages quickly. Either it is slowed | ||
// down or we are a spammer. Should not really | ||
// happen, if it does we do nothing and will flush again later |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This comment still seems a bit weird, should be above the WouldBlock
error.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This was the existing style, but I'll move it up
src/lib.rs
Outdated
@@ -28,8 +28,8 @@ | |||
//! } | |||
//! ``` | |||
|
|||
#![forbid(unsafe_op_in_unsafe_fn)] |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
#![forbid(unsafe_op_in_unsafe_fn)] | |
#![deny(unsafe_op_in_unsafe_fn)] |
This now includes the changes in Smithay/calloop#144
This should be sufficient to fix some of the weirdness that
winit experiences, as well as resolve the hangs that Glazier finds