The campaign banner previously stretched edge-to-edge with
`object-cover` and a fixed aspect ratio (4:3 mobile, 21:9 sm, 3:1 lg),
which cropped landscape banners hard left/right on phones and clipped
any baked-in text near the image edges.
Replace it with a full-bleed banner that respects the image:
- Sharp foreground uses `object-contain` capped to the same
`max-w-6xl` reading column and `max-h-[70vh]`. The banner's
height is dictated by the image, never by an arbitrary aspect
ratio. Source pixels are never cropped.
- The bleed gutters around the contained image are filled by a
blurred, scaled-up copy of the same image (`object-cover scale-110
blur-2xl brightness-75`) on a `bg-black` base, with a soft
horizontal vignette (`from-black/55 via-transparent to-black/55`)
so the bleed recedes and the centered image reads as the subject.
- A subtle `shadow-lg shadow-black/25` lifts the banner off the
page below.
- Clicking the banner opens the cover in the shared
`Lightbox` from `@/components/ImageGallery` — same fullscreen
component already used by NoteContent, NoteCard, PostDetailPage,
etc. (portal, body-scroll lock, keyboard nav, swipe, pinch zoom,
Capacitor-safe download).
Move the back / edit / delete chrome to behave differently per
breakpoint. On mobile (where the banner image fills nearly the full
viewport width and the bleed gutters are too narrow to safely host
overlays) the controls live in a black band above the banner. On
`sm:` and up they overlay the banner inside the same `max-w-6xl`
column as glass chips. The black band wrapper collapses to zero
height on desktop so the banner sits flush with the page top.
Add a new `campaignsDetail.openCover` translation key (aria-label
for the cover button) to en.json and translate it into all fifteen
non-English locales.
The campaign detail hero overlaid the title, summary, byline, country/
deadline meta, and PostActionBar on top of the banner image, behind an
80%-tall bottom scrim. Two failure modes on every breakpoint:
- Banner images often carry baked-in text or branding. The scrim
covered the bottom 80% of the image, and `object-cover` cropped
the rest hard on phone-portrait containers (mobile used
`min-h-[92svh]`, roughly 1:1.6 portrait), so banners with side-
aligned text were sliced off.
- Long titles and tl;dr summaries overflowed the fixed-height hero.
With `overflow-hidden` on the header the top of the title got
clipped behind the upper edge of the banner.
Split the hero into two components. `CampaignHero` keeps only the
banner image and the floating back/admin chip buttons; aspect ratios
stay landscape at every breakpoint (4:3 mobile, 21:9 sm, 3:1 lg) so
`object-cover` only trims a thin strip top/bottom. The top scrim is
gone — the chip buttons already carry `bg-black/30 backdrop-blur-md`
backdrops that read on any image without darkening the banner.
`CampaignHeading` renders the title, summary, byline, meta, divider,
and `PostActionBar` in the normal page flow inside the existing
`max-w-6xl` column on `bg-background`. Title and summary no longer
need a text-shadow or line-clamp; the action bar drops the glass-chip
overrides and inherits its default styling against the page surface.
The Trans component's values={{ count: formatNumber(N) }} was spreading
after the numeric count prop, overriding it with a string. i18next v26
requires a numeric count for plural resolution — a string causes it to
return the raw key path (e.g. 'campaignsDetail.repost') instead of the
resolved plural form.
Rename the interpolation variable from count to formattedCount so it no
longer collides with the count prop that Trans uses for plural selection.
Update all 16 locale files to use {{formattedCount}} in the repost, quote,
and like translation strings.
Regression-of: 86a084f3
When FollowPage was removed in the unreachable-page cleanup, the
QR code on the profile kept encoding `${origin}/follow/<npub-or-nip05>`,
which falls through to NotFound because there is no `/follow/...`
route. Encode the bare identifier instead — Agora's universal
NIP-19 dispatcher at /:nip19 already resolves both npub and
`user@domain.com` to ProfilePage.
Regression-of: b975e557
The Create Campaign page initialised `walletSource` to `'custom'` and
relied on a post-mount effect to flip it to `'mine'` once
`hdWalletAvailable` became true. With Radix Select's controlled value
that left users staring at the "Choose a wallet" placeholder on the
initial paint and forced them to open the dropdown to see that their
Agora wallet was even an option.
Seed both `walletSource` and `mineAccept` lazily from the
synchronously-available HD-wallet availability so the trigger already
reads as the user's Agora wallet on first render. The existing
availability-change effects still cover the (rare) case where the hook
resolves a tick later or where the wallet disappears mid-session.
Opens the HD wallet's send dialog prefilled with the campaign's on-chain
address — same flow used at /wallet, so donors don't have to learn a
second send UI. Sits above 'Open external wallet' inside the donate
panel; the external button downgrades to the secondary variant when
both are present so only one orange CTA stacks.
When the campaign declares both an on-chain (bc1…) and a silent payment
(sp1…) endpoint, a swap link appears under the privacy disclaimer so
donors can flip to SP without leaving the modal. The toggle hides
itself if the donor manually edits the recipient field, so we never
trash their typed input.
Gated on useHdWalletAccess (nsec logins only) — extension and bunker
logins fall through to the external-wallet QR. The button is also
hidden for SP-only campaigns (no on-chain address to prefill) and for
the campaign owner themselves.
Add Kosovo (XK) and Western Sahara (EH) to the country list. Kosovo
has no Unicode emoji flag, so it follows the Tibet pattern with a
bundled SVG asset that CountryFlag swaps in.
Surface Tibet (CN-XZ) as a search-list entry so it can be picked from
country autocompletes and pickers. The on-wire identifier stays
iso3166:CN-XZ; only the picker pretends.
Route every remaining raw country.flag span through CountryFlag so
bundled SVGs render in autocomplete dropdowns, organizer selects, the
ComposeBox destination switcher, and the world stats dialog.
Picked the noun each locale uses in its existing last3h / last24h
preset translations so the standalone label stays consistent with
the dropdown options it sits underneath.
The Math.max(resume, target) clamp inside resolveWindowFromHeight
turned the Since dropdown into a no-op whenever the wallet's
scanHeight had caught up to the chosen window. Picking 'Last week'
one minute after a successful scan resolved to scanHeight + 1, not
to a week ago, so the user couldn't actually rescan history from
the primary control — they had to drop into Advanced -> From block.
Removed the clamp so Since presets honor the wall-clock window
literally. Re-scanning blocks we've already scanned is cheap (the
indexer is just iterating tweak data) and is exactly what the user
asked for. Also dropped the isPresetUpToDate predicate that, after
the original clamp landed, was the mechanism by which Start went
disabled after a scan completed. isManualUpToDate is kept so the
override path still flags 'block number past the tip'.
Regression-of: 38946cbc
Adds a Custom… entry to the Since select that, when picked, reveals
an hours number input directly under the select. The hours value is
parsed as a positive (fractional allowed) number, multiplied to
seconds, and fed through the same mempool.space resolver as the
fixed presets. The Start button stays disabled until a valid hours
value is entered so there's no ambiguous empty-equals-zero submit.
Power users who want sub-hour granularity (e.g. 0.5) get it without
having to drop into the Advanced block-height path.
Bitcoin block timestamps obey BIP-113's median-time-past rule — they
must exceed the median of the previous 11 timestamps but don't have
to be strictly monotonic. Empirically, 1-2-block inversions show up
in the live chain regularly; 3-block rewinds covered the common case
but could still miss a payment landed in a block whose timestamp ran
backwards near the user's selected cutoff. 11 is the principled
upper bound (no inversion can extend further under MTP) and costs
only ~8 extra block fetches per scan.
Regression-of: 46d9952a
mempool.space's timestamp-to-block endpoint is the sole source of
truth now. The Blockbook binary search fallback was sequential and
could stall the UI for ~20 round-trips with no progress feedback
when mempool.space was down. When the lookup fails, surface a toast
that points the user at the Advanced -> From block override and
auto-open the disclosure so the escape hatch is immediately visible.
Regression-of: d6e6d616
lemon