Repository navigation
v1.9.0
This release adds a network status that tells you when your own connection is the problem,
records connection trouble in the Activity Log, and puts every screen within reach of the
keyboard. It also brings Mirall's new brand: a refined logo and a redrawn app icon.
Highlights
- Mirall now tells you when your network is the problem — it used to call a connection
healthy on any network that let it out, so a machine nobody could reach still looked
fine. It now judges whether people can actually reach you, says so when they can't, and
offers a page with what to try. - The Activity Log records connection problems — a new Network filter keeps when this
device went offline, when Mirall couldn't reach anyone, and when members dropped out or
came back, so the log explains a quiet stretch instead of falling silent through it. - Every screen can be reached from the keyboard — Profile and the Activity Log have
shortcuts of their own, and Network status, the Settings pages, What's new and Send
feedback are in the command palette (⌘K / Ctrl-K). - The empty screens explain how Mirall works — the spaces list, an empty space, and the
wait for approval each carry a card that says what happens next and links to the
documentation.
Also fixed
- Downloads survive a peer's upload cap — a transfer from a limited peer could stop with
the file shown as paused, and Resume moved it only a little further. A capped peer now
says it is still there, and interrupted downloads retry on their own. - A download says when its folder is gone — a deleted, renamed, or ejected download folder
produced a generic "Transfer failed", and a deleted folder was quietly recreated
mid-download. Mirall now names the folder as the problem, offers to change it, and stops
retrying. - Clearing the Activity Log frees its disk space — clearing emptied the log from view but
left every entry on disk, so the store came out larger than before. It is now cleared
properly and the space comes back, with your retention setting kept. - Speed limits are shared out fairly — with a bandwidth cap set, a transfer that started
while nothing else was running could take more than its share and stall a transfer
running alongside it. Both now draw from the cap evenly.
Full Changelog: v1.8.0...v1.9.0