v1.6.41 - Intel/AMD hardware transcoding no longer fails on sources the GPU cannot decode

With an Intel or AMD GPU (VAAPI), a source its hardware decoder does not support — 10-bit HEVC on
older Intel graphics, AV1 before the decoder supports it — made the transcode fail, and the player
showed an error. Those sources are now decoded in software and still encoded on the GPU. Sources
the GPU can decode take the hardware path as before.

Upgrade: pull it.

v1.6.40 - Software transcodes play everywhere

Without a GPU, a 10-bit source (common for HEVC releases) was transcoded to 10-bit H.264, which
TVs and phones cannot decode and browsers decode only in software, if at all. Software transcodes
now always come out 8-bit, as hardware ones already did.

Upgrade: pull it.

v1.6.39 - Closes a gap in the check for playlists disguised as video

Since 1.6.20 the server refuses a torrent whose "video" file is really a playlist, by reading the
file's first bytes before the probe opens it. A torrent built to deliver only part of those bytes
in time could get past that check. The server now refuses any file whose first bytes it could not
read in full.

Upgrade: pull it.

v1.6.38 - A transcode survives a torrent that stalls

When a torrent still downloading stalled for longer than a piece's timeout while it was being
transcoded, the transcode took the stall for the end of the file: it stopped, and the player
treated the film as over at that point. It now asks again from where it stopped and carries on
once the data arrives.

Upgrade: pull it.

v1.6.37 - Transcoded playback no longer stops ten minutes after the transcode finishes

A file the player cannot play directly (one with embedded subtitles, for example) is transcoded
much faster than it is watched. About ten minutes after the transcode finished, its housekeeping
deleted the output while the film was still playing, and playback stopped at the next segment. A
transcode a player is still reading is now kept until it stops reading.

Upgrade: pull it.

v1.6.36 - Statistics match the stock server for streams without a file index

Since web player 5.0.0-beta.41, a stream that names no file asks the server for the torrent's
statistics instead of one file's. The server answered with a file progress of zero, so the player's
statistics showed "Completed 0 %" for the whole stream. It now leaves those per-file fields out, as
the stock server does, and the player hides the row.

Upgrade: pull it.

v1.6.35 - Keeps reporting where the TV certificate came from

1.6.34's certificate status (certStatus in /health) is kept in a small file beside the cache.
Once the cache filled up, the cache cleaner could delete that file first, and /health then said
nothing about the certificate until the next restart. The cleaner now leaves it alone.

Upgrade: pull it.

v1.6.34 - Gets its TV certificate back by itself

With IPADDRESS set, a server that could not get the trusted *.stremio.rocks certificate at
startup now asks again every 30 minutes and switches to it the moment it arrives — no restart from
you. Meanwhile it says why, in one plain log line and in /health (certStatus), and reports
itself degraded, since TVs cannot connect until then.

Upgrade: pull it.

v1.6.33 - Starts cleanly while the TV certificate service is down

With IPADDRESS set, the server asks Stremio's certificate service for the trusted *.stremio.rocks
certificate at startup. When that service failed quickly, the container could stop during start and
restart over and over. It now starts every time: a trusted certificate it already has is kept for as
long as it is valid, and only a server with none falls back to a self-signed one until the service
is back (restart the container then).

Upgrade: pull it.

v1.6.32 - Keep no longer asks for the whole cache budget twice

Keep held back the full cache budget plus 10% beside the title, even when the cache already filled
most of that budget on disk. On a small disk it refused titles that fitted — a 3 GB film on a disk
with 26 GB free, under a 30 GB budget. It now holds back only the room the cache can still grow
into, plus the 10%. An empty cache gets the same reserve as before, and a title that would fill the
disk is still refused. Thanks to jaison for the report.

Upgrade: pull it.