Trying to watch an already watched video will make the video start 15 seconds
before the end. This is not very comfortable when listening to music or
watching/listening playlists over and over.
This can be easily tested on any instance with the "Save playback position"
enabled in the Preferences.
Closes issue 3976
Before this PR, Invidious assumed that every playlist had at least one video.
When a playlist had no videos, Invidious was throwing an "Index out of bounds"
exception.
The following API endpoints were impacted:
* api/v1/playlists/:plid
* api/v1/auth/playlists/:plid
Fixes issue 4679
Videos of a playlist cataloged as podcast are called "episodes" therefore
Invidious was not able to find video in the text value inside the stats array.
Test case: "/playlist?list=PLDu-Eh5lUs1a4irCbnxMIB6FrUMaTXgVF"
Fixes issue 4688
This pull request fixes that bug that was causing the query parameters to get
doubled in the streaming URLs when '?local=true' is passed to the
'/api/v1/videos/{id}' API endpoint.
Before: host/path?parameters?parameters
After: host/path?parameters
No associated open issue
At the moment Invidious will return hardcoded data for the 'size',
'qualityLabel' and 'fps' fields for streams, when such hardcoded data is
available, otherwise it just omits those fields from the response (e.g. with
the AV1 formats). Those issues are especially noticable when Invidious claims
that 50fps streams have 60fps and when it claims that the dimensions for a
vertical video are landscape. The DASH manifests that Invidious generates
already use the correct information.
This pull request corrects that issue by returning the information that
YouTube provides instead of hardcoded values and also fixes the long
standing bug of Invidious claiming that audio streams have 30 fps.
Here are two test cases:
50/25/13fps: https://youtu.be/GbXYZwUigCM (/api/v1/videos/GbXYZwUigCM)
vertical video: https://youtu.be/hxQwWEOOyU8 (/api/v1/videos/hxQwWEOOyU8)
Originally these problems were going to be solved by the complete refactor
of stream handling in 3620, but as that pull request got closed by the stale
bot over a month ago and has such a massive scope that it would require a
massive amount of work to complete it, I decided to open this pull request
that takes a less radical approach of just fixing bugs instead of a full
on refactoring.
FreeTube generates it's own DASH manifests instead of using Invidious' one,
so that it can support multiple audio tracks and HDR. Unfortunately due to
the missing and inaccurate information in the API responses, FreeTube has
to request the DASH manifest from Invidious to extract the height, width and
fps. With this pull request FreeTube could rely just on the API response,
saving that extra request to the Invidious instance. It would also make it
possible for FreeTube to use the vp9 streams with Invidious, which would
reduce the load on the video proxies.
Closes issue 4131
Before this PR, setting the modified code repo URL through the preferences
page in Invidious was broken:
* the HTML input tag for this field had invalid type "input"
(though browser falls back on text input)
* the URL was used to set the "checked" property and not as a plain value,
which makes no sense for a text-based input (and resulted in a blank field)
* when the submitted field is empty, the retrieved value was an empty 'String'
instead of 'nil', causing the "modified source code URL" to be an empty
'href' link which just pointed to the current page
No associated open issue
When visiting /api/v1/popular and popular endpoint is disabled
Before:
500 {"error":"Closed stream"}
After
403 {"error":"Administrator has disabled this endpoint."}
The current Content Security Policy does not allow to embed videos
inside local HTML files which are viewed in the browser via the file
protocol. This commit adds the file protocol to the allowed frame
ancestors, so that the embedded videos load correctly in local HTML
files.
This behaviour is consistent which how the official YouTube website
allows to embed videos from itself.
Closes issue 4448
Some opengraph implementations don't support a URL without the domain
therefore failing to fetch the video thumbnail and channel image.
This pull request basically fixes that.
The transcript logic in Invidious was written specifically as a workaround for
captions, and not transcripts as a feature.
This PR genericises the logic as so it can be used to implement transcripts
within Invidious.
The most notable change is the added parsing of section headings when it was
previously skipped over in favor of regular lines.
This PR changes the current master based container to use "master" tag instead
of "latest" tag and adds a new workflow to build a container on each new
release which has the "latest" tag, and a tag based on the current released
version.
Trying to watch an already watched video will make the video start 15
seconds before the end of the video. This is not very comfortable when
listening to music or watching/listening playlists over and over.
The transcript logic in Invidious was written specifically
as a workaround for captions, and not transcripts as a feature.
This commit genericises the logic a bit as so it can be used for
implementing transcripts within Invidious' API and UI as well.
The most notable change is the added parsing of section headings
when it was previously skipped over in favor of regular lines.
Invidious assumes that every playlist will have at least one video
because it needs to check for the `index` key. So if there is no videos
on a playlist, there is no `index` key and Invidious throws
`Index out of bounds`
Videos of a playlist cataloged as podcast are called episodes therefore
Invidious was not able to find `video` in the `text` value inside the
stats array.