14 releases 14 tags

RSS feed

• v2026.8.17 78347065d9

  Compare

  v2026.8.17 Stable

  robbertbos released this 2026-08-17 19:38:04 +00:00 | 1 commits to main since this release

  Replying in the thread or in the channel

  • The chevron next to "Verzend" opens one panel: the destination (thread or
    channel) above the scheduling options. The button then reads "Verzend in
    kanaal", and the choice returns to the thread every time.
  • "Markeer kaart als voltooid na verzenden" now applies to "Verzend" too, not
    only to a scheduled send. The card completes once the send has gone out, so a
    failure or an undo leaves it open.
  • A scheduled message keeps its destination, and Uitgaande post names it.
  • A channel is named in words rather than with a sigil: "In kanaal Town
    Square", "Aan @noor". The card breadcrumb drops its "#" too, so a
    screenreader hears "kanaal X" instead of "#x".
  • Fixed: Escape in the scheduling popover closed the whole card.

  Preview for text attachments

  • A markdown attachment opens in a preview instead of downloading, with a
    "Weergave"/"Broncode" switch.
  • Other text files (.txt, .csv, .json, code, logs) open straight in the source
    view, syntax-highlighted where a grammar exists. Classified on the extension
    too, because Mattermost reports many of them as octet-stream.
  • Above 1 MB the preview asks first and names the size; past 5 MB it fetches
    only the beginning, so a big file is never transferred whole.
  • HTML, PDF and text previews now share one window instead of three
    hand-rolled ones.

  Cards load faster, and show a skeleton

  • A loading card shows placeholder bars instead of a half-filled card, and its
    quick actions and status tags wait for the message.
  • Takes it longer than a second, an activity indicator covers and dims the
    list. Both follow NLDD's loading guidelines, which rule out a shimmer.
  • A message that fails to load says "Bericht kan niet worden geladen" instead
    of waiting on skeleton bars forever; the card's actions stay usable.
  • Channel and team labels come from two collection endpoints instead of one
    call per channel: four round trips per linked server, whatever the number of
    cards. A few hundred cards took over ten seconds to paint before.
  • Those labels are now cached for an hour across requests; the old per-client
    cache never hit, because the client is built fresh per request.
  • Fixed: a Mattermost token expiring during team lookup read as a server error
    rather than a token problem.

  Reading the messages around a card

  • Opening the panel no longer hides the message you were reading.
  • One centred "Berichten laden..." indicator instead of a spinner and a label
    scattered across the width, and the panel settles once instead of twice.
  • A load-more button names its direction while it loads instead of shrinking to
    "Laden...".

  Under the hood

  • An unreachable Mattermost no longer costs one hanging avatar request per
    author: the first failure is remembered for a minute, the timeout drops from
    15 to 3 seconds, and an upstream failure reads as 502 rather than 500.
  • An avatar reply that is not an image is refused rather than replaced by a
    transparent pixel, which showed as an empty circle where initials belong.
  • The dev Mattermost mock is installed at startup instead of being reached for
    per request: waggle no longer names devtools in the request path.
  • The mock can fake a per-call latency (WAGGLE_DEV_MM_MOCK_LATENCY_MS),
    answers posts/ids for ids it never created, serves /users/me/channels and
    honours Range.
  • The dev seed writes refs-only cards like production, instead of a snapshot
    nothing reads that showed as one repeated message on every card.
  • Renovate 44.31.0; fresh nginx-unprivileged digest for the frontend image.

  Container images

  • backend: code.overheid.nl/robbertbos/waggle/waggle-backend:2026.8.17
  • frontend: code.overheid.nl/robbertbos/waggle/waggle-frontend:2026.8.17