Pin the component library at 9036c92: RaTeX formulas, Mermaid diagrams, pictures from the document's folder
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
1 parent
4183f729d1
commit
53bcf09b0b
8 files changed
+267
-11
No files matched your search
+1
-1
@@ -99,7 +99,7 @@
|
||||
|
||||
- [x] C14.0 前置:`reference/` 加 markion 与 markdown-rs 并钉提交;`scripts/review.sh` 照用户新定(gpt-6.1-sol,effort high,fast,每条最多三轮) — R2 ok
|
||||
- [x] C14.1 组件库:`MarkdownRenderer`、`MarkdownEditor`、Markdown 着色、块范围、大纲与链接查找改用 markdown-rs(`documents` 的 `tree`、`markdown`、`render`);`BlockEditor` 的导入导出不属 ElyMark,仍用 pulldown-cmark;原有测试照过,改了语义的测试写明 markdown-rs 的读法;推送,ElyOffice 钉新提交 — R2 ok
|
||||
- [ ] C14.2 组件库:公式与 Mermaid(MARK-04)。`typography::Latex` 改由 RaTeX 排出;Mermaid 由 mermaid-rs-renderer 画成图;画不出即显示错误与原文;`MarkdownRenderer` 用二者,本地图片按属主给的文件夹读,网络图片不自动载入
|
||||
- [~] C14.2 组件库:公式与 Mermaid(MARK-04)。`typography::Latex` 改由 RaTeX 排出;Mermaid 由 mermaid-rs-renderer 画成图;画不出即显示错误与原文;`MarkdownRenderer` 用二者,本地图片按属主给的文件夹读,网络图片不自动载入 — R1 FAIL 六项已修:版面失败把公式写进日志(只记固定文字);资源与图集从不释放(`primitives::held` 持有,换内容或离开即释放);公式失败丢了原文与位置(显示原因、位置与原文);零边距裁掉字形(按墨迹重取画框);`file:` 地址把片段当文件名、把主机当相对路径(解析主机,去查询与片段);Mermaid 节点字色未取 Ely 前景。R2 FAIL 一项已修:一个持有者离开即逐出共享资源(按资源计数)。R3 FAIL 两项已修(不再送审):最后离开的持有者若未收到图,图集里的图不释放(图存于共享记录,由最后离开者释放,单测 `the_last_holder_frees_the_image_another_painted`);一行注释 11 词(已缩短)
|
||||
- [ ] C14.3 组件库:文件级 Markdown 文档组件。一个 `CodeEditor` 持全文与撤销;可视(变高虚拟列表,逐块渲染,按下的块开成源码,写回为一步撤销)、源码(`CodeEditor` 加 Markdown 着色)、预览(只读);属主可读写文本、模式、插入点行列、滚动,可跳到行;改动发事件
|
||||
- [ ] C14.4 ElyMark 文件会话改为原生(MARK-01,§14.3,QA-21):视图换成 C14.3 的组件,文本即文件原文,无序列化;BOM、CRLF 照旧;新建、打开、保存、另存、重命名照旧;应用不再依赖 `ely-web-content`
|
||||
- [ ] C14.5 删去 CEF:`crates/ely-web-content`、`scripts/fetch-cef.sh`、`scripts/build-mark-content.sh`、`apps/ely-mark/web`、打包里的 CEF 辅助应用与许可、CEF 的测试脚本;PRD 与 `agents.md` 写明新路线,ElyMail 正文不用 CEF 与 Electron
|
||||
|
||||
@@ -0,0 +1,13 @@
|
||||
FAIL
|
||||
|
||||
1. `../Ely-GPUI-Components/src/typography/math.rs:122` — The guarded layout failure logs the entire formula source, violating the rule against logging document content and PRD §18.2. A parseable formula containing confidential text reaches this branch whenever RaTeX layout panics. A fault-injected copy reproduced `latex: RaTeX failed to lay out "\\text{PRIVATESENTINEL}"` in the captured diagnostic log; this verifies the error path, not a naturally panicking input. Remove the source from diagnostics and add a regression asserting that layout failures cannot log document text.
|
||||
|
||||
2. `../Ely-GPUI-Components/src/typography/math.rs:141` — Formula assets have no release path; the same problem affects `diagram.rs:116` and `picture.rs:158`. GPUI's application asset cache retains their source keys and decoded images until explicitly removed, and painted images also require atlas cleanup. The exact-source view probe loaded a formula, removed it from the view, and still found its asset cached. Opening and closing distinct documents in one window therefore accumulates resources, contrary to PERF-08. The new gap row documents this defect but does not resolve it. Give these assets a bounded or owner-managed lifetime and release unused cache entries and atlas images; test repeated removal without breaking shared users.
|
||||
|
||||
3. `../Ely-GPUI-Components/src/documents/render.rs:356` — Invalid math replaces the original formula with only an error message; asynchronous failures in `typography/math.rs:146` do the same. Additionally, `math.rs:39` and `math.rs:94` discard RaTeX's error location by keeping only `error.message`. For `$$x^2^3$$`, the renderer shows only `Double superscript`, losing both `x^2^3` and the parser's position 3. This violates C14.2 and PRD §14.2. Preserve the source alongside parse/layout/raster errors and retain the available error position; cover the rendered failure state, not just parser rejection.
|
||||
|
||||
4. `../Ely-GPUI-Components/src/typography/math.rs:113` — Zero SVG padding clips outlined glyphs that extend beyond RaTeX's layout metrics. The exact-source probe for `\text{私密}` at 16 px produced an SVG viewport height of 19.626667 with glyph ink extending to 21.28897, so rasterization cuts off the characters' bottom strokes. Even `x` extends above and below its viewport. Account for the actual glyph bounds or sufficient padding, include that space in the returned logical size, and test that complete glyphs survive rasterization.
|
||||
|
||||
5. `../Ely-GPUI-Components/src/documents/picture.rs:65` — The `file://` branch treats URI components as filename text. `file:///C:/notes/a.png#preview` resolves to `C:/notes/a.png#preview`, so an existing `a.png` fails to load. `file://server/share/a.png` becomes the relative path `server/share/a.png`, potentially reading an unrelated working-directory file instead. Both cases failed exact-source probes. Parse file URLs before converting them to paths: exclude query/fragment components, preserve supported authorities correctly, and explicitly reject unsupported hosts. Add regressions for these cases.
|
||||
|
||||
6. `../Ely-GPUI-Components/src/documents/diagram.rs:71` — Assigning only `theme.text_color` does not apply Ely's foreground to ordinary diagram labels. With ink `#eeeeee`, `flowchart LR\n A --> B` emits Mermaid's default `#e0dfdf` and no `#eeeeee`; the exact-source probe failed. This contradicts the last decision's claim that diagram text uses Ely's palette. The existing test named `a_flowchart_draws_with_the_ground_and_ink` checks only the background. Map the foreground to the theme fields actually used by unstyled labels and assert their emitted colors.
|
||||
@@ -0,0 +1,3 @@
|
||||
FAIL
|
||||
|
||||
1. `../Ely-GPUI-Components/src/primitives/held.rs:14` — Releasing one holder unconditionally evicts the shared asset and removes its image from every window's atlas, although other holders can still display it. The exact-source primitive probe renders two users of one image, with the surviving user inside an `Entity::cached` view. Remove the other user and render the parent again: both the asset and the surviving image's atlas entry become absent, while the cached child still references that image and never reloads it. The single-user control retains both entries. Thus the round-1 resource fix invalidates live cached scenes; the existing shared-formula test masks this by repeatedly refreshing the entire window. Retain shared assets and atlas images until their last holder leaves, and test a cached survivor without forcing its refresh, together with cleanup after the final holder leaves. Evidence: `.work/C14.2-r2-held-probe.log` and `.work/C14.2-r2-held-probe-source.log`. The unmodified library's 1748 tests and `cargo clippy --all-targets --features test-support -- -D warnings` passed.
|
||||
@@ -0,0 +1,5 @@
|
||||
FAIL
|
||||
|
||||
1. `../Ely-GPUI-Components/src/primitives/held.rs:44` — A departing holder discards its image when another holder remains, but the last holder may still have `image: None`. For example, two windows start loading the same asset; one stays covered while the other receives and paints the result. Remove the painted element, then close the covered window before it redraws: the last release removes the asset but never calls `drop_image`, leaving the first window's atlas entry allocated. The exact-source lifecycle probe seeds that pending-holder state and reproduces `asset=false, atlas=true` after every holder leaves; this is a primitive reproduction, not a native-window visibility test. Retain the loaded image with the shared asset record until the last holder leaves, and cover this release order with an atlas assertion. Evidence: `.work/C14.2-r3-pending-source.log` and `.work/C14.2-r3-pending.log`. Round 2's cached-survivor defect is fixed: its original probe now retains both asset and atlas entries. The library's 1748 tests and `cargo clippy --all-targets --features test-support -- -D warnings` passed.
|
||||
|
||||
2. `../Ely-GPUI-Components/src/primitives/held.rs:56` — The newly changed comment contains 11 whitespace-delimited words: "`source` as `A` for element `id`; the previous one let go." Applying the required ten-word comment rule to this revision fails on that line. Shorten it to ten words or fewer. Evidence: `.work/C14.2-r3-checks.log`.
|
||||
Reference in new issue
Block a user