Reconsider adaptive touch correction? Slimmer proposal with details
维护者通常 1 天内回复
评估
调研方向
This is a design proposal, not a ready task. Read the maintainer's earlier concerns in #671 and the open pull request #680 before anything else. The payload names InputLogic as the core file the change touches, plus a separate statistics package with 17 tests. Done means the maintainers have agreed on a design, not that a patch exists. Wait for that decision before starting code.
由索引模型根据 Issue 内容生成。
描述
Thanks again for reviewing #671. I'd like to explain the feature in more detail and offer a much slimmer version, because I think it addresses your concerns.
Why it matters
I'm an experienced, heavy phone typist. Still, over ~1,250 recorded touches about 13% landed on a neighboring key — roughly every 8th letter. And it's systematic, not random: "a" hit as "s" is consistently my most frequent miss. A fixed, per-user offset like this is exactly what adaptive touch correction fixes, and as far as I know it has been standard in mobile keyboards for many years.
What it does
The keyboard learns where this user actually touches each key, based on the words they finally commit (including reverted autocorrections and self-corrections). Once a key has enough data (30+ touches), a touch near the border between two keys is assigned to the key the user most likely meant. The shift is capped to a fraction of a key width, and the dictionary receives the corrected coordinates too.
Your concerns
- Determinism: same data, same key. It only differs from today near key borders and only after enough data. With the feature off, the exact current code path runs.
- Latency: per touch a few arithmetic operations on cached offsets, no I/O.
- Muscle memory: it adapts to the user's muscle memory instead of changing it.
- Code footprint: of the ~1.5k lines, ~1.45k are in a separate package (model, statistics screen, 17 tests). Core changes are ~115 lines.
Slimmer proposal
- One single switch ("Adapt to my typing"), fixed strength, no other settings.
- Move all recording logic out of InputLogic into the separate package, leaving ~60 lines of hooks in core.
- Optional: drop the statistics screen if you prefer.
If you're still not convinced, no problem — I'll keep it in my fork.
- 主要语言
- Kotlin
- 星标
- 1k
- 派生
- 59
- 平均合并
- 13 小时 21 分钟
- 30 天内合并 PR
- 48
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 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。
LeanBitLab/LeanType 的其他 Issue
-
Voice input fails in browser address bars due to incorrect password field detection (v4.2.9, follow-up to #516)可能已有人在做 @LeanBitLab 今天认领。 未关闭bug
难度 2/5 1-3 小时 新手友好度 76/100
LeanBitLab/LeanType#713 · 1 条评论 · 已指派 1 人 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 65/100
LeanBitLab/LeanType#423 · 1 条评论 ·
维护者通常 1 天内回复
-
bug
难度 3/5 1-2 天 新手友好度 54/100
LeanBitLab/LeanType#717 ·
维护者通常 1 天内回复
-
Crush Report可能已有人在做 @LeanBitLab 今天认领。 未关闭
难度 3/5 1-2 天 新手友好度 20/100
LeanBitLab/LeanType#715 · 1 条评论 · 已指派 1 人 ·
维护者通常 1 天内回复
-
Text Expander/cursor-handling bug可能已有人在做 @LeanBitLab 于 3 天前认领。 未关闭bug
难度 3/5 1-2 天 新手友好度 62/100
LeanBitLab/LeanType#703 · 1 条评论 · 已指派 1 人 ·
维护者通常 1 天内回复
查看 LeanBitLab/LeanType 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
Kotlin/kotlinx-io#529 ·
-
难度 2/5 1-3 小时 新手友好度 70/100
mcnaveen/health-connect-webhook#110 ·
维护者通常 1 天内回复
-
Delete all above未关闭enhancement
难度 2/5 1-3 小时 新手友好度 68/100
mjaakko/NeoStumbler#1307 · 2 条评论 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 65/100
tanujnotes/Olauncher#770 ·
-
难度 2/5 1-3 小时 新手友好度 66/100