The VUCC grid-detail dialog listed every station location's QSOs even when a specific operated grid was selected; only 'All' should combine them. The 台址 grid-set key derivation is now shared between the selector's grouping and the dialog filter (core/domain extensions + unit tests).
The radar circle is a square, so it is limited by the shorter side of
the card it is drawn in: the small-window layout gave the radar card a
height of ~156dp in a 411x450 window, which shrank the circle to ~36% of
the page width while the full-screen layout shows ~84%.
- measure the real window with BoxWithConstraints instead of trusting the
window size class, which can still report the full screen while the app
runs in a split/floating window
- when the classic 1:1 split would give the radar less than 75% of the
page width, use a fill layout: the radar takes the largest square that
fits, a usable pager block is reserved under it while that keeps the
circle at its full-screen share, otherwise the pager collapses into a
48dp translucent tab strip with a tappable panel
- keep the normal two-row top area in the vertical layout: TimerRow and
NextPassRow both carry weight(1f) internally, so putting them in a
single row splits that row into two 50% blocks
- extract the geometry into pure Dp functions (useFillRadarLayout,
radarFillSizes) covered by unit tests
The previous compact-mode trigger used currentWindowAdaptiveInfo(),
which reports the full-screen size class in split-screen / multi-window,
so the radar never enlarged there. Measure the actual window bounds with
BoxWithConstraints: compact (radar weight 1f fills all space, pager fixed
132dp strip) when maxWidth < 600dp && maxHeight < 480dp.
build: release version 4.4.7-ba7opf.17.6 (529)
When the window is vertical (narrow) and too short for split-screen or a
small floating window, give the radar card 1.6/0.6 weight instead of the
usual 1:1 so the square radar plot stays large. The pager card shrinks
but all its pages are scrollable, so nothing is clipped. Full-screen
vertical layout (enough height) is unchanged.
P2 — markUploaded now only marks the records that actually made it
into the accepted TQ8 batch (preview.submittedIds), never the whole
candidate list. Previously, un-signable records skipped by prepare()
were still marked UP after any accepted upload, falsely showing them
as uploaded.
P4 — surface why records did not upload:
- When audit finds nothing pending, distinguish unavailable records
(invalid call/date — check logbook) from unknown-result records
(not retried automatically) instead of a blanket 'no pending'.
- Upload preview dialogs (Log tab + logbook) show a breakdown line:
N already uploaded/duplicate · N unknown result · N un-uploadable.
- Remove LOCATION_MISMATCH check: the recorded myGrid never enters the
TQ8 (the tSTATION grid is signed from the current certificate station),
so a portable/ADIF-imported record with a different grid must not block
upload. Logging and uploading are now fully decoupled from the
certificate location.
- Fingerprint now covers the CONTACT only (call, date, band, mode,
satellite, freq), excluding station fields: changing grid/zones/county/
IOTA no longer invalidates ledger entries, so already-uploaded contacts
are not re-queued and rejected as duplicates.
- prepare(): catch un-signable contacts per-record instead of aborting
the whole batch; count them as skipped and upload the rest.
- audit(): count lotwUploaded records as uploaded (ledger-loss fallback).
The LoTW upload card read uiState.lotwCertificate which was only
populated when the config dialog was opened (LoadLoTWUploadStatus).
On first entry to the settings screen it stayed null, showing
'certificate not imported' even though a certificate was stored.
Load certificate + station once in init so the card is correct
from the first frame.
atsunatsu