Add scrolling capture of the window under the cursor - #2097
Conversation
…notification setting, avoid revealing main window after OCR, restore macOS diagnostics bindings
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| let new_rows = offset.min(frame_height); | ||
| let strip = frame.crop_imm(0, frame_height - new_rows, frame_width, new_rows); | ||
| stitched.push(strip); | ||
| stitched_height += new_rows; | ||
| prev = cur; | ||
|
|
||
| if stitched_height >= MAX_STITCHED_HEIGHT { | ||
| debug!("Stitched image reached the height cap; stopping"); | ||
| break; | ||
| } |
There was a problem hiding this comment.
When a capture is near 16,000 pixels, the full next strip is appended before the cap is checked, allowing the result to exceed the screenshot editor's 16,384-pixel limit and fail to open.
| let new_rows = offset.min(frame_height); | |
| let strip = frame.crop_imm(0, frame_height - new_rows, frame_width, new_rows); | |
| stitched.push(strip); | |
| stitched_height += new_rows; | |
| prev = cur; | |
| if stitched_height >= MAX_STITCHED_HEIGHT { | |
| debug!("Stitched image reached the height cap; stopping"); | |
| break; | |
| } | |
| let remaining_height = MAX_STITCHED_HEIGHT.saturating_sub(stitched_height); | |
| let new_rows = offset.min(frame_height).min(remaining_height); | |
| if new_rows == 0 { | |
| debug!("Stitched image reached the height cap; stopping"); | |
| break; | |
| } | |
| let strip = frame.crop_imm(0, frame_height - new_rows, frame_width, new_rows); | |
| stitched.push(strip); | |
| stitched_height += new_rows; | |
| prev = cur; | |
| if stitched_height >= MAX_STITCHED_HEIGHT { | |
| debug!("Stitched image reached the height cap; stopping"); | |
| break; | |
| } |
Knowledge Base Used: Desktop Tauri App (Rust Backend)
Prompt To Fix With AI
This is a comment left during a code review.
Path: apps/desktop/src-tauri/src/scrolling_capture.rs
Line: 74-83
Comment:
**Height cap checked too late**
When a capture is near 16,000 pixels, the full next strip is appended before the cap is checked, allowing the result to exceed the screenshot editor's 16,384-pixel limit and fail to open.
```suggestion
let remaining_height = MAX_STITCHED_HEIGHT.saturating_sub(stitched_height);
let new_rows = offset.min(frame_height).min(remaining_height);
if new_rows == 0 {
debug!("Stitched image reached the height cap; stopping");
break;
}
let strip = frame.crop_imm(0, frame_height - new_rows, frame_width, new_rows);
stitched.push(strip);
stitched_height += new_rows;
prev = cur;
if stitched_height >= MAX_STITCHED_HEIGHT {
debug!("Stitched image reached the height cap; stopping");
break;
}
```
**Knowledge Base Used:** [Desktop Tauri App (Rust Backend)](https://app.greptile.com/cap/-/custom-context/knowledge-base/capsoftware/cap/-/docs/desktop-tauri-app.md)
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.| #[allow(unused_variables)] | ||
| fn inject_scroll(notches: i32) { | ||
| #[cfg(windows)] | ||
| { | ||
| use ::windows::Win32::UI::Input::KeyboardAndMouse::{ | ||
| INPUT, INPUT_0, INPUT_MOUSE, MOUSEEVENTF_WHEEL, MOUSEINPUT, SendInput, | ||
| }; | ||
|
|
||
| const WHEEL_DELTA: i32 = 120; | ||
|
|
||
| let input = INPUT { | ||
| r#type: INPUT_MOUSE, | ||
| Anonymous: INPUT_0 { | ||
| mi: MOUSEINPUT { | ||
| dx: 0, | ||
| dy: 0, | ||
| mouseData: (notches * WHEEL_DELTA) as u32, | ||
| dwFlags: MOUSEEVENTF_WHEEL, | ||
| time: 0, | ||
| dwExtraInfo: 0, | ||
| }, | ||
| }, | ||
| }; | ||
|
|
||
| unsafe { | ||
| SendInput(&[input], std::mem::size_of::<INPUT>() as i32); | ||
| } | ||
| } | ||
|
|
||
| #[cfg(target_os = "macos")] |
There was a problem hiding this comment.
When a Linux user invokes the exposed scrolling-capture hotkey, inject_scroll performs no operation, so the loop waits once and saves only the initially visible frame instead of capturing off-screen content.
Knowledge Base Used: Desktop Tauri App (Rust Backend)
Prompt To Fix With AI
This is a comment left during a code review.
Path: apps/desktop/src-tauri/src/scrolling_capture.rs
Line: 158-187
Comment:
**Linux scroll injection is absent**
When a Linux user invokes the exposed scrolling-capture hotkey, `inject_scroll` performs no operation, so the loop waits once and saves only the initially visible frame instead of capturing off-screen content.
**Knowledge Base Used:** [Desktop Tauri App (Rust Backend)](https://app.greptile.com/cap/-/custom-context/knowledge-base/capsoftware/cap/-/docs/desktop-tauri-app.md)
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.| async takeScreenshot(target: ScreenCaptureTarget) : Promise<string> { | ||
| return await TAURI_INVOKE("take_screenshot", { target }); | ||
| }, | ||
| async captureOcrText(target: ScreenCaptureTarget) : Promise<string> { | ||
| return await TAURI_INVOKE("capture_ocr_text", { target }); | ||
| }, | ||
| async captureScrollingWindow(windowId: WindowId) : Promise<string> { | ||
| return await TAURI_INVOKE("capture_scrolling_window", { windowId }); | ||
| }, | ||
| async importCurrentDesktopBackground(projectPath: string) : Promise<string> { | ||
| return await TAURI_INVOKE("import_current_desktop_background", { projectPath }); | ||
| }, |
There was a problem hiding this comment.
This directly modifies the generated tauri.ts IPC surface and includes unrelated generated type churn. These changes can be overwritten during regeneration and leave the checked-in bindings out of sync; regenerate the file from the Rust command definitions instead.
Context Used: AGENTS.md (source)
Knowledge Base Used: Desktop Frontend (apps/desktop/src)
Prompt To Fix With AI
This is a comment left during a code review.
Path: apps/desktop/src/utils/tauri.ts
Line: 59-70
Comment:
**Generated bindings changed directly**
This directly modifies the generated `tauri.ts` IPC surface and includes unrelated generated type churn. These changes can be overwritten during regeneration and leave the checked-in bindings out of sync; regenerate the file from the Rust command definitions instead.
**Context Used:** AGENTS.md ([source](https://github.com/capsoftware/cap/blob/main/AGENTS.md))
**Knowledge Base Used:** [Desktop Frontend (apps/desktop/src)](https://app.greptile.com/cap/-/custom-context/knowledge-base/capsoftware/cap/-/docs/desktop-frontend.md)
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
There was a problem hiding this comment.
These bindings match what specta regenerates from the Rust command definitions - the dev build regenerated tauri.ts after the capture_scrolling_window command was added and the checked-in file is byte-identical to the regenerated output, so it's in sync, not hand-drifted.
There was a problem hiding this comment.
That makes sense — if the file is byte-identical to the specta output after adding capture_scrolling_window, checking it in is the right move. The concern was about hand-drifted bindings, which doesn't apply here. This comment is resolved.
Tip: You can customize Greptile's behavior for this repo with .greptile/rules.md and .greptile/config.json.
1 participant
Summary
Adds scrolling capture: a new hotkey action ("Scrolling capture of window under cursor", unbound by default) that captures an entire scrollable window — not just the visible part — by scrolling it and stitching the frames into one tall screenshot, saved as a normal Cap screenshot project.
How it works (
scrolling_capture.rs):imagebuffers.Showcase
A 60-section page captured into one 1280x7404 image:
Bottom of the stitch — capture stops at the page footer with no repeated content:
Testing (Windows, end-to-end on a live dev build)
.capscreenshot project on every runmacOS compiles the same capture/stitch path with a CoreGraphics scroll-event injector; runtime-tested on Windows only.
Builds on #2095 (OCR text capture — shares its capture/save helpers); the scrolling-capture change itself is the top two commits.
Greptile Summary
This PR adds window scrolling capture, OCR area capture, associated hotkeys and settings, and refactors screenshot capture/saving into reusable backend helpers.
Confidence Score: 3/5
The PR should not merge until scrolling captures are capped before appending oversized strips and the Linux action is either implemented or disabled.
Near-limit captures can produce images rejected by the screenshot editor, while Linux exposes an action whose scroll injector performs no operation and therefore cannot capture off-screen content.
Files Needing Attention: apps/desktop/src-tauri/src/scrolling_capture.rs, apps/desktop/src/routes/(window-chrome)/settings/hotkeys.tsx, apps/desktop/src/utils/tauri.ts
Important Files Changed
Prompt To Fix All With AI
Reviews (1): Last reviewed commit: "Tune scrolling capture stepping and matc..." | Re-trigger Greptile
Context used (3)