• #[must_use] attribute indicates that the user or the method must do something with the return type.
• Functions that Result<> should document when the functon Err's

You could also choose to tell Clippy not to complain about it, but either way, I find the squigllies distracting.

As of right now, the Connection::event_stream() function will provide a stream of events for use by client programs (or other libraries). Right now, it can only serialize two types of events: EventBody (sent by all applications except QT), and EventBodyQT (the type sent by QT applications). QT should be updating their implementation to be compliant with all the others if what I read in the at-spi2-core issues is correct.

Now, there are other kinds of events that may come through that same connection, for example Cache::AddAccessible and Cache::RemoveAccessible return data with types ((so)(so)(so)iiassusau) and (so) respectively.

Other signals on interfaces whoes signature is currently not supported by atspi: DeviceEventListener::KeystrokeListener(De)registered, Socket::Available, Registry:EventListener(De)registered.

Ok, so here's the issue: do we use an enum to transport all of these events. Something like this:

enum A11yEvent {
    RegistryEvent(RegEventBody),
    UserEvent(EventBody),
    DeviceListenerEvent(DeviceListenerBody),
    CacheAddAccessible(AddAccessibleBody),
    CacheRemoveAccessible(RemoveAccessibleBody),
}

Or is there some other way to handle this that makes more sense?

Paging @albertotirla , @mcb2003 , @DataTriny for options on this since it will directly effect projects they work on. (Thanks again for all y'alls effort and contributions you've brought so far!)

@luukvanderduim has been able to produce a working example by slimming down our problem into a smaller binary instead of a huge code-base like a screenreader. With this, we've found that lack of more diverse serialization types has caused these other types of events to be unable to be processed by assistive technologies at the higher level.

Returning an enum like this would be a breaking change since the event type returned by event_stream() will be different. Any other suggestions or is everyone ok with the breakage?

I'm personally fine with it, since that part of Odilia's code can be easily modified, but I just want to be sure that others haven't done too much with it, especially DataTriny.

We don't want to force tokio on users of AccessKit. The atspi crate doesn't either.

Introduces new crate features for the atspi-macros crate and use these inside the main Cargo.toml to propagate the information.

I haven't looked much into atspi-codegen because it didn't seem to be used directly, but it also uses tokio unconditionally. I just don't know if this is something to worry about.

Following in the tradition of Luuk's fully-typed-events https://github.com/odilia-app/atspi/pull/28

Now adding Luuk as a contributor so we can work together on improving this branch.

Set the IsEnabled property in the session bus.

Assistive Technology provider applications (ATs) should set the accessibility IsEnabled status on the users session bus on startup as applications may monitor this property to enable their accessibility support dynamically.

See: The freedesktop - AT-SPI2 wiki

Crates that added atspi as a dependency do not need all of the repository. Adding an include or exclude directive allows for minimizing the projects published size.

This PR reduces the published version of atspi by 27% or 120.3 KB in 26 files (of 447.5 KB and 75 files in entire crate)

Achieved with the help of cargo-diet

While fixing #10 on f220104a, the zbus version requirement was not updated accordingly. This creates an issue for everyone depending on an older version of zbus that doesn't yet support the new attributes and outputs confusing compilation errors. With this fix Cargo should try to find an appropriate zbus version to satisfy everyone, or clearly say that a version can't be selected.

Signals are broadcasted Events. So, to test for events, it is sufficient to test if the Message is MessageType::Signal. This test is likely a little bit cheaper than the string compare, but that is not the point. Point is that we allow 5 more streams to come through RegistryProxy:

$ rg ' signal' src/events/ -trust | wc --lines 56

$ rg ' signal' src/ -trust | wc --lines 61

The difference is made up by these events:

src/registry.rs 31: /// EventListenerDeregistered signal 35: /// EventListenerRegistered signal

src/cache.rs 59: /// AddAccessible signal 63: /// RemoveAccessible signal

src/socket.rs 30: /// Available signal

Please create tags for published versions of the atspi crate. The following commands should do most of the work:

git tag -a atspi-v0.0.1 f0dc8abfd3d71e4462ee3767055f78df37a0b3f7
git tag -a atspi-v0.1.0 92f744cb616e9ff6bcd00c2795f87889138067e0
git tag -a atspi-v0.1.1 bf5260e9b03efe7b1f79d4781988097ce708a357
git tag -a atspi-v0.1.2 77b926357e0eb4b07bf2a9dc063fc4354ea288f2
git tag -a atspi-v0.2.0 c1e865d421a9617ca6bbcc95a83be0ac408acd9f
git tag -a atspi-v0.2.1 c5b360abe29faf99b563c40b9038b1d19e9a4f3f
git tag -a atspi-v0.2.2 7266f83ee9fa77aecea0c2b954603e14a2c878a3
git tag -a atspi-v0.3.1 038b59cd6ca36fcab93035773c92add775db08e7
git tag -a atspi-v0.3.2 5be639425bf4efa92ee58e46ec6bea25d862167a

Unfortunately, the following versions do not point to a commit that still exist in this repository:

• atspi-v0.2.3 (commit 7ab0c5a204856acc63c39d6677ee3783f4047ccd)
• atspi-v0.3.0 (commit bfa075e5fd4eb703bc0cdacd12cd213fc9341417)
• atspi-v0.3.3 (commit 49cdfdebfe330a3a6c5b4ee46d985522d4bd2081)
• atspi-v0.3.4 (commit 4fdcba0d4334b2782b0ffb62a7a0b990c234a721)

It would be nice if you could do this for new versions as it helps people figure out what code they are using, and to detect potential supply-chain attacks.

Thanks.

It looks like atspi-macros doesn't actually need the dependency on serde_with; it compiles fine without it. I bring this up because serde_with, at least using default features, adds many more dependencies, including such apparently nonsensical ones as android_system_properties (that one came indirectly via chrono). These may all be build-only, and some may be platform-specific dependencies that don't actually come up on any build platform, but some people will find the addition of so many extra dependencies in their Cargo.lock file annoying anyway.

Set the IsEnabled property in the session bus.

Assistive Technology provider applications (ATs) should set the accessibility IsEnabled status on the users session bus on startup as applications may monitor this property to enable their accessibility support dynamically.

See: The freedesktop - AT-SPI2 wiki

• rebased against main
• removed #[must_use]. It does something different than I intended. Useful, but not here.

Whether the StateChangedEvent contains a boolean or not over the DBus wire, it appears that right now it works with an i32 (DBus: "i") return type.

Should this be manually changed in the identify.rs file, then add a test for it returning a boolean, or should we leave it up to the library users figure it out? It's not like checking ==1 is super complicated.

Some concerns:

1. I'm getting "module name reiterated as suffix" sort of clippy error on all the Enums in identify.rs; this is because the module name is separate to distinguish between object::PropertyChangeEvent and window::PropertyChangeEvent. Should I ignore these errors, or is there some better way to format that file (note the identify.rs is auto-generated, so please suggest the change here and I will put it in atspi-codegen).
2. Some methods have "too many parameters", but those parameters have been inserted by the dbus_proxy macro, not the code itself; it would never be visible to end-users. Should this be ignored as well?

That's just about it, I'm pretty happy with how close we are to passing clippy here.

Currently the event de-structuring looks as such:

while let Ok(ev) = event_stream.next().await.expect("None") {
    let Event::Interfaces(EventInterfaces::Object( ObjectEvents::StateChanged(state_ev))) = ev else { continue; };

    if state_ev.state() == "focused" {
        let Some(sender) = state_ev.sender()? else { continue; };
        println!("Focused by: {sender}")
     }
}

If Event would hold a copy of AtspiEvent (instead of EventInterfaces) and the TryFrom implementations are adjusted to convert from Event to respectively: the signified types or the grouped EventInterfaces, we could:

while let Ok(ev) = event_stream.next().await.expect("None") {
    let ev: StateChangedEvent = ev.try_into()? else { continue; };

    if ev.state() == "focused" {
        let Some(sender) = state_ev.sender()? else { continue; };
        println!("Focused by: {sender}")
     }
}

perhaps have current behaviour as such:

while let Ok(ev) = event_stream.next().await.expect("None") {
    let evs: EventInterfaces = ev.try_into()? else { continue; };

    /// Then do destructuring ...
}

Right now one has to use bpth atspi and zbus separately to allow events to come in on an event stream.

It should be this easy:

let conn = connection::new();
conn.register_event(Event::Document::LoadComplete);

let stream = conn.event_stream();
while let Some(event) = stream.next().await {
    ...
}

If the library user would like to add additional things to the dbus connection, then they are welcome to do so by getting the internal connection (already possible) or creating a new connection all on their own.

It should be easy to use atspi without knowing how dbus works—you should only have to know that this is hpw atspi works, and not be exposed to implementation details.

I'm in the process of doing several things more or less at the same time:

• documenting the dbus XML interfaces in at-spi2-core
• Fixing incorrect method signatures as part of making the XML the single source of truth

The result is that the dbus XML files in at-spi2-core are changing to reflect what the C code actually does. Bindings that use the XML as a basis will of course want to be notified of this.

So, this issue is me asking you for how to best communicate this.

I can file a bug report here every time I make a change, but this doesn't help other projects (e.g. GTK4, WebKit) that also use DBus directly instead of atk.

(Concretely, my latest fix is in this commit - not merged yet, hopefully in a merge request soon.)