feat(frontend): avatar crop dialog (viewport 320, wheel/pinch/keyboard, guides+shade, previews, dirty guard)
Rebuilds the "pick a photo" flow around an Edit Image modal, matching every product surveyed in the 2026-10 avatar-settings research (Discord, Google, Slack, Figma, Cropper.js playground). Before: 240 px circular viewport with no visible crop frame, a single range slider, and no way to close with the keyboard — the direct reason the previous version felt unfinished. cropMath: VIEWPORT=320, OUTPUT_SIZE=512 (matches Slack min), plus wheelZoom (exp so scroll up then down returns to the same zoom, unlike linear), pinchZoom (ratio of two-pointer distance), PAN_STEP/ZOOM_STEP for keyboard, PREVIEW_SIZES for the "how others see it" column. AvatarCropper is now a pure canvas: pointer capture for pan, two-pointer tracking for pinch, wheel handler with exponential zoom, tabIndex=0 + onKeyDown for arrows/+/-/0, a Reset button, and a visual frame with rule-of-thirds guides (40 % opacity) and an inset box-shadow that dims the corners outside the round target — so the user actually sees what falls in the file. Cancel/Confirm live in the dialog; dirty signal and current CropState bubble up via props. AvatarEditDialog is a portal: bg-black/60 backdrop with click-to-close, role=dialog + aria-modal, initial focus into the card, ESC closes, X button in the header. If the user has panned/zoomed (dirty), closing prompts for confirmation. Header shows file.name + size + the accepted-format hint previously only surfaced after a failed upload. Right column renders 3 live previews at 128/64/32 px so the small HUD nameplate treatment is visible before upload. Progress bar and error notice moved inside so the user does not have to look away during a slow upload. AvatarUpload picks the dialog instead of the raw cropper; two-step "Remove avatar" from the previous commit is unchanged. Tests: 5 new cropMath cases (wheel symmetric, wheel clamps both ends, pinch ratio, pinch guards zero baseline). Existing 140 stay green (145 total), tsc clean, oxlint 0/0. Not yet: no dep on react-easy-crop — own code as planned. No focus trap beyond initial focus + ESC (dialog traps implicitly via portal + backdrop covering). Rotate is still missing — deliberately, since Discord does not ship it either and extending sourceRect for theta would complicate the preview pass.
This commit is contained in:
parent
5e76b6176a
commit
4fce425dcd
7 changed files with 491 additions and 75 deletions
|
|
@ -3,6 +3,34 @@
|
|||
export const MIN_ZOOM = 1
|
||||
export const MAX_ZOOM = 4
|
||||
|
||||
/**
|
||||
* Cropper viewport edge (px). Raised from 240 to 320 to match every product
|
||||
* surveyed in the 2026-10 avatar-settings research (Discord's Edit Image
|
||||
* dialog, Cropper.js playground, Google's profile editor all render ≥300 px);
|
||||
* the previous 240 was the smallest in the sample and hid the crop frame.
|
||||
*/
|
||||
export const VIEWPORT = 320
|
||||
|
||||
/** Output edge (px) written to the file on confirm. Matches Slack's 512x512 minimum. */
|
||||
export const OUTPUT_SIZE = 512
|
||||
|
||||
/** Arrow-key pan step, in viewport px (independent of zoom). */
|
||||
export const PAN_STEP = 12
|
||||
|
||||
/** `+` / `-` zoom step; roughly 10 % of the range per press. */
|
||||
export const ZOOM_STEP = 0.15
|
||||
|
||||
/**
|
||||
* Wheel-to-zoom factor. Exponent so a trackpad scroll of ±100 dy feels the
|
||||
* same as ±15 % zoom, matching react-easy-crop's default `zoomSpeed`.
|
||||
*/
|
||||
export const WHEEL_ZOOM_FACTOR = 0.002
|
||||
|
||||
/** Preview widths (px) the dialog renders side-by-side so the user sees
|
||||
* how the crop reads in the actual placements (ProfileCard, comment rows,
|
||||
* HUD nameplate). Mirrors Discord's "Preview" panel. */
|
||||
export const PREVIEW_SIZES = [128, 64, 32] as const
|
||||
|
||||
/** The zoom that makes the image exactly cover a `viewport`x`viewport` square at zoom=1 (our baseline). */
|
||||
export function coverScale(naturalWidth: number, naturalHeight: number, viewport: number): number {
|
||||
return viewport / Math.min(naturalWidth, naturalHeight)
|
||||
|
|
@ -34,6 +62,21 @@ export function clampZoom(zoom: number): number {
|
|||
return Math.min(MAX_ZOOM, Math.max(MIN_ZOOM, zoom))
|
||||
}
|
||||
|
||||
/**
|
||||
* Zoom for a wheel event: exponential in deltaY so a light scroll and a hard
|
||||
* fling feel proportional, and symmetric (scrolling up then down by the same
|
||||
* total returns to the starting zoom, which a linear `zoom + k*delta` does not).
|
||||
*/
|
||||
export function wheelZoom(current: number, deltaY: number): number {
|
||||
return clampZoom(current * Math.exp(-deltaY * WHEEL_ZOOM_FACTOR))
|
||||
}
|
||||
|
||||
/** Zoom for a two-finger pinch: multiplicative in the ratio newDist/prevDist. */
|
||||
export function pinchZoom(current: number, prevDistance: number, nextDistance: number): number {
|
||||
if (prevDistance <= 0) return current
|
||||
return clampZoom(current * (nextDistance / prevDistance))
|
||||
}
|
||||
|
||||
/**
|
||||
* Given the image's natural size, the crop viewport size, the current pan offset (viewport px,
|
||||
* center-relative) and zoom, compute the source rectangle (in natural image pixels) that should
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue