An example on how to build a tree of all objects using atspi.
Root
├── Child1
│ ├── Grandchild1
│ └── Grandchild2
│ ├── Great-Grandchild1
│ └── Great-Grandchild2
└── Child2
├── Grandchild1
└── Grandchild2
The A11yNode::from_accessible_proxy_recursive
method currently requires Rust 1.77.2 or higher.
Cargo-msrv reports:
error[E0733]: recursion in an `async fn` requires boxing │
│ --> src/main.rs:67:5 │
│ | │
│ 67 | async fn from_accessible_proxy(ap: AccessibleProxy<'_>) -> Result<A11yNode> { │
│ | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ recursive `async fn` │
│ | │
│ = note: a recursive `async fn` must be rewritten to return a boxed `dyn Future` │
│ = note: consider using the `async_recursion` crate: https://crates.io/crates/async_recursion
The A11yNode::from_accessible_proxy_iterative
method requires Rust 1.75.0 or higher.
processes | objects | recursive method | iterative method |
---|---|---|---|
25 | 241 | 11.87871ms | 31.792694ms |
26 | 2900 | 288.716071ms | 474.438153ms |
28 | 4960 | 992.864247ms | 808.82431ms |
27 | 6948 | 1.447283253s | 1.077571422s |
28 | 7509 | 1.515564922s | 1.08042886s |
30 | 7813 | 1.589614858s | 1.119880292s |
It appears 'recursive' is faster with a small number of accessible applications, however 'iterative´ is faster when a larger number of objects is exposed on the bus. Note that it does not take that many applications to reach the point where iterative is faster.
xdg-dbus-proxy
does not seem to implement all methods on the Accessible
interface.
This results in an error if applcations require it:
Error: MethodError(OwnedErrorName("org.freedesktop.DBus.Error.UnknownMethod"), Some("Method \"GetRole\" with signature \"\" on interface \"org.a11y.atspi.Accessible\" doesn't exist\n"), Msg { type: Error, serial: 53, sender: UniqueName(":1.106"), reply-serial: 48, body: Signature("s"), fds: [] })
It appears LibreOffice Calc exposes 2^31 accessible objects (the table cells), which leads to an impractical and impossible situation.
When fama11y-tree calls GetChildren
on the Accessible
interface of their parent's frame, Calc will try and send them all. Which results in Calc freezing.
Not sure what is going on, maybe Calc tries to gather data to construct a reply message containing all children which - even if it succeeds at that - it would not send a message that size because D-Bus protocol prohibits sending messages exceeding 128 megabyte.
The problems:
- Calc unresponsive and the accessible technology waits indefinitely for a reply.
- eg. A screen-readers' cache could never handle a list of children that large.
- D-Bus messages are not allowed to exceed 128 megabytes.
Exposing a subset, eg. exposing 'visible cells' only could make this more practical and sending the reply feasible.