to_line_index loops indefinitely when a visual line starts with an empty text run
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 50/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- rust
- 领域
- accessibility, backend
调研方向
Start in accesskit_consumer/src/text.rs at TextPosition::to_line_index() and backward_to_line_start(), then trace the preceding_text_runs iterator and its line-boundary metadata. Reproduce with the supplied empty-run example and add regressions for leading empty runs, document-end positions, hard breaks, soft wraps, and empty lines. Done when the method terminates with the correct line number and the consumer and macOS adapter tests pass.
由索引模型根据 Issue 内容生成。
描述
TextPosition::to_line_index() can repeatedly return to the same position when a visual line begins with an empty text run linked to a populated run. It never returns a line number. In the macOS adapter, this method serves accessibilityLineForIndex: and accessibilityInsertionPointLineNumber, so an affected request can block the calling thread.
Affected code and verification
Reproduced with published accesskit 0.25.1 / accesskit_consumer 0.39.1 and with AccessKit revision a39329d281a9381510496c17e81e87763edbb8bb. The failure is in platform-independent consumer code. Verification ran on macOS ARM64 with Rust 1.87.0; the reproduction uses no platform accessibility APIs or screen reader.
Minimal reproduction
One visual line contains two runs, "" and "a", with reciprocal next_on_line / previous_on_line links. Their character lengths are [] and [1], so the metadata matches the text. There is no hard line break or stored selection.
Cargo.toml:
[package]
name = "accesskit-empty-text-repros"
version = "0.1.0"
edition = "2024"
rust-version = "1.87"
publish = false
[dependencies]
accesskit = "=0.25.1"
accesskit_consumer = "=0.39.1"
src/main.rs:
use accesskit::{Node, NodeId, Role, TextDirection, TreeId, TreeInfo, TreeUpdate};
use accesskit_consumer::Tree;
fn main() {
let mut field = Node::new(Role::MultilineTextInput);
field.set_children([NodeId(1), NodeId(2)]);
let mut first = Node::new(Role::TextRun);
first.set_value("");
first.set_character_lengths([]);
first.set_text_direction(TextDirection::LeftToRight);
first.set_next_on_line(NodeId(2));
let mut last = Node::new(Role::TextRun);
last.set_value("a");
last.set_character_lengths([1]);
last.set_text_direction(TextDirection::LeftToRight);
last.set_previous_on_line(NodeId(1));
let tree = Tree::new(
TreeUpdate {
nodes: vec![(NodeId(0), field), (NodeId(1), first), (NodeId(2), last)],
tree: Some(TreeInfo::new(NodeId(0))),
tree_id: TreeId::ROOT,
focus: NodeId(0),
},
true,
);
let node = tree.state().root();
let pos = node.text_position_from_global_utf16_index(0).unwrap();
eprintln!("Computing line index; expected 0:");
println!("{}", pos.to_line_index());
}
Run cargo run. It prints Computing line index; expected 0: and then does not terminate.
Expected behavior
The position is on the first visual line, so the method should promptly return 0.
Cause
to_line_index repeatedly calls backward_to_line_start() until is_document_start() is true. The UTF-16 lookup in this example returns character index 0 of the "a" run. backward_to_line_start then:
- Biases backward to character index 0 of the preceding empty run.
- Finds that empty run as the line start.
- Biases forward from the empty run's end to character index 0 of the
"a"run again.
The result has the same node and character index as the original position. It still has a preceding text run, so is_document_start() is false and the loop repeats. All character indices are in bounds; no tree update or stale selection is needed.
Proposed approach
Compute the zero-based line number by counting preceding visual-line boundaries with the finite preceding_text_runs iterator. Under the existing same-line metadata, each preceding run without a next_on_line link terminates a visual line. This avoids depending on normalized cursor movement making progress and preserves the public signature.
Add regressions for one or several leading empty runs on the first line and after a preceding line, including document-end positions. Preserve the existing hard-break and soft-wrap behavior, and cover empty lines and multiple linked runs per line.
I have a local candidate which fixes the reproduced loops and passes the existing default workspace and macOS adapter tests. The consumer implementation remains no_std. The change affects a shared method, whose current in-repository callers are in the macOS adapter; other-platform runtime testing and interactive accessibility testing have not been performed as part of this verification.
For comparison, Chromium's CreatePreviousLineStartPosition uses boundary traversal with explicit line-start/end predicates and movement options. The proposed AccessKit implementation instead counts its existing visual-line boundaries directly, avoiding a broader redesign of shared line navigation.
- 主要语言
- Rust
- 星标
- 1.5k
- 派生
- 127
- 平均合并
- 16 小时 44 分钟
- 30 天内合并 PR
- 12
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
AccessKit/accesskit 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 75/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 76/100
维护者通常 1 天内回复
-
难度 4/5 1-2 天 新手友好度 32/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 25/100
维护者通常 1 天内回复
查看 AccessKit/accesskit 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 72/100
dsx-ai-factory/infra-controller#8341 ·
维护者通常 1 天内回复
-
enhancement triage:deciding
难度 2/5 1-3 小时 新手友好度 70/100
open-telemetry/otel-arrow#4540 ·
维护者通常 2 天内回复
-
benchmark: CSV report files contain blank records on Windows可能已有人在做 关联的 PR 仍在进行中或已合并。 未关闭component:benchmark
难度 2/5 1-3 小时 新手友好度 76/100
agentic-os-org/ANOLISA#9177 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
oracle/rust-oracledb#61 ·
-
bug priority-medium to-agent
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复