Skip to content

Mediarium v4.1.0

Latest

Choose a tag to compare

@github-actions github-actions released this 18 Nov 18:31

Install or update: the Docker image is ghcr.io/rdborg/mediarium:4.1.0 (also :latest), for linux/amd64 and linux/arm64. A running Mediarium offers this version under Settings > Basics > General, where Update now installs it. New here? Start with the install guide or mediarium.app.

Added

  • Recyclarr can sync the TRaSH Guides straight into Mediarium. Add Mediarium to your Recyclarr file as a Radarr instance (/radarr) and a Sonarr instance (/sonarr) with an administrator's API key, and recyclarr sync puts the guides' quality profiles, custom formats and their scores, quality sizes and naming formats where you can see and edit them: the Quality profiles page, the Custom formats card, the size table and the Custom naming style. A second sync changes nothing and makes no copies. It was tried with Recyclarr 8.8 against a test copy; a sample file is in docs/quality-profiles.md.
  • Naming formats can cut a name to a length. {Episode CleanTitle:90} is at most 90 characters, as in Sonarr and Radarr, and the show names the TRaSH Guides use ({Series CleanTitleWithoutYear} {(Series Year)}) work as they are.
  • An assistant's find_releases says which releases are freeleech or at a discount. Each release has freeleech and, for any discount, discount_percent, so an assistant can prefer them on a private tracker. docs/mcp.md.
  • Extra languages don't change the ranking. A new switch under Settings > Finding releases > Quality > Audio language, off by default so nothing changes unless you turn it on. With several languages in your list, a multi-language release was placed above a plain release in your language, which pushed aside releases that match your preferred terms. With the switch on, the other languages a release carries are ignored: only which language of your list it has counts (the first one best), and then the quality and your preferred terms decide. qualityLanguageIgnoreExtra in GET and PUT /api/settings. Reported by mirisbowring on GitHub, #91.

Changed

  • A custom format name is now unique for each app, not for the whole list. The TRaSH Guides use many of the same names in the Radarr and the Sonarr guide (HDR, x265 (HD), BR-DISK), with different conditions, so each app now has its own. Nothing you have is changed. The Custom formats card shows both when both exist.
  • A new install now keeps looking for better versions. The setup wizard's Quality & naming step has a Keep looking for better versions switch, on at first: it turns better versions on for the quality profile you pick and makes new titles start with it ticked, so Look for a better copy and the quality cutoff do something out of the box (they used to need two switches in two places, and a title added from an import list never took part). The same choice is now under Settings > Finding releases > Automatic searching > When you add a title, and the add dialog starts from it until you tick or untick the box yourself (it then remembers your choice in that browser). Nothing changes for an install that already exists: its profiles, its titles and the box in the add dialog stay as they are. Switch it off in the wizard, or later under Automatic searching, to keep the first good download of everything.
  • add_movie, add_series and search_now over MCP now need the downloads switch. Adding a title or searching for it starts a search that downloads by itself, so these three were a way past Let assistants pick releases and start downloads while find_releases and grab_release were behind it. They are now behind the same switch, off by default: they are no longer listed for an assistant and a call is refused with a plain message that nothing was added, searched for or downloaded. Turn the switch on under Settings > Media servers and apps > Other apps if you use them. Everything else that changes something (profiles, preferred terms, retrying a failed download) is unchanged. docs/mcp.md.
  • Big libraries load faster. The Library lists asked the database for the info language once for every film and show (thousands of small reads on a big library). Settings that are not secrets are now remembered after the first read and refreshed the moment they are saved. On a test library of 5,000 films the film list took about 1.75 seconds to build after a change and now takes about 0.1 second.
  • Wanted opens quicker on a big library. The list looked up every show again for each episode (and the grouped view for each show). It now takes what it needs from the same read: the episode list on 800 shows went from about 104 ms to 46 ms.
  • The blocklist answers faster when it is long. Asking for the whole blocklist (not a page) looked up the title of every entry one by one. The titles are now read together. Removing a film, show, album or book is also quicker with a long blocklist, because the blocklist, playlists, reading progress and the queue now have the indexes that removal needs.
  • The page loads less code at start. The made-up data of demo mode and the two add dialogs of the search box (artist and book) are now fetched only when they are used, which takes about 20 KB (compressed) off what every page load downloads (207 KB instead of 227 KB of scripts). The app also stops redrawing everything each minute when the list of switched-on media types has not changed, and coming back to a tab asks the server once instead of twice.
  • Messages, hints and the docs name the settings pages the way the menu does now. Many error messages, hints, help texts and docs pages still sent you to places that no longer exist (Settings > Library > Folders, Settings > Info, lists and subtitles, Settings > Account & Access, Settings > Accounts, Settings > Profiles) or wrote "Library and files and files". They now give the full path as the menu shows it, for example Settings > Library and files > Folders, Settings > Finding releases > Quality, Settings > Basics > Media types, Settings > People and access > All accounts or Settings > Media servers and apps > Notifications, in every language. This covers turning on passkeys, reports and 4K requests, linking Plex, making an API key, setting up Soulseek, the message on the search page when there is no TMDB key, the footer of notification emails sent without a link, the setup step for quality and file names (quality is under Finding releases > Quality, file names under Library and files > File names and importing) and the docs for the settings file, TVmaze air times, TheTVDB, OMDb scores, artwork, ListenBrainz, the recycle bin and the support copy.
  • The screenshots in the docs show the current menus and pages (the Settings groups, Library, Discover, Upcoming, books, the import pages and the quality profile editor) instead of the old ones, with the Usenet and Torrents pages as two pictures. They all use the demo account, made-up titles and no version number.
  • The README, the contributing notes and the bug report forms say where Copy for support really is (Settings > System > Logs and errors), and the docs list a row for Recyclarr.
  • The Docker image is built with Go 1.27.4. The 1.27.2 it was built with has known security problems in the web server part of Go (a request could crash the server, and a certificate check could be wrong).
  • The environment variable list in the documentation now includes APP_PORT and UMASK, and the image size and the compose file notes (ebooks and audiobooks now work) are correct.
  • Wording: "user name" is now "username" everywhere in the app, like the sign-in form.
  • Less chatter, a little faster in other languages. The app's own files are now fetched at the same time as the language file, not after it (on a slow connection in German about 10% sooner). The same question is no longer asked twice when a page opens (problems, update check, import status), and the import banner asks every 20 seconds instead of every 3 when no import is running: 20 requests a minute became 3 on an idle dashboard.

Fixed

  • A Usenet release or a torrent that is bigger than the free space is now refused before it starts. Before, it ran until the disk was full and failed then. It now says what the release needs and how much is free, and the download is tried again by itself later, like any full-disk problem.
  • A Usenet article that arrives short or empty is now asked again from your next server. An article with no checksum of its own that held fewer bytes than its header said was accepted as it was, and left a hole for PAR2 to find.
  • A Soulseek download no longer waits for ever when only some of the songs came. If the rest stayed in the other user's queue, or slskd lost them (a restart), the download sat there with no end. Songs slskd no longer lists are now asked for again, and a release that gets nothing for hours is given up on like one that never started.
  • Real-Debrid season packs keep their folders. Real-Debrid names each file by its bare name, so two files with the same name in different folders (an English.srt in every episode's folder) wrote over each other. They now keep the paths of the torrent.
  • A debrid download that stopped half way starts over cleanly when the part on disk is bigger than the file or the service refuses to carry on, instead of failing and asking you to press Retry. Errors from Real-Debrid, TorBox and Premiumize now say plainly when the service is asking you to slow down or is having trouble, and a TorBox plan limit is no longer shown as a wrong API key.
  • An MDBList list with movies and shows together is read to the end. A full page of movies and shows was taken for the last page, so lists over 200 titles lost the rest.
  • The path mapping of a download client ignores spaces around the two folders, and a path from the client that climbs out of its folder with .. is no longer mapped into a folder it was not meant for.
  • A magnet link with spaces or a line break around it is now recognised when the torrent is looked up in your download client.
  • A background job with no interval at all can no longer run in a loop that never rests. It now waits a minute.
  • Emptying the recycle bin by age carries on past one entry it cannot delete (a file in use, a permission), so the older ones behind it are no longer kept for ever.
  • More release names are read right. A copy written Blu-Ray or Blu.Ray was not taken for a Bluray, so a 1080p Blu-Ray release was ranked as a WEB-DL. The same went for WEB.DL, WEB-Rip, BDRemux, AV1, DoVi, HDR10Plus and a size written 1920x1080. Show names with the season spelled out (Season 5 Episode 14, Staffel 5, Saison 2), episode numbers of four digits (S01E1085), dated daily shows with the year last (15.03.2024), anime batches written (01-12), Complete.S01-S08 packs and Multi-Subs are understood too.
  • A name with a single letter before the year, like WALL.E.2008, and a name that reads "Web" before a bracketed year, like Charlotte's Web (2006), keep their whole title. A release such as Show - 366.720p is no longer taken for episode 366.5 of something.
  • Other titles with an apostrophe or dots now match a release that leaves them out, so Le.Fabuleux.Destin.dAmelie.Poulain finds the film named "Le Fabuleux Destin d'Amélie Poulain". A release named Love.Island.US now matches a show called "Love Island USA".
  • The automatic pick no longer depends on which indexer answered first. When the same release name sat on two indexers some hours apart and a third release was equal in quality, the pick could change from one search to the next. It now gives the same release every time.
  • Music: Artist - 2013 - Album, scene names with WEBFLAC and a bare 320 between dashes are read right. Books: releases that carry a series note such as (Dune Chronicles #1), a 50th Anniversary Edition or .fb2 match their book, and Dune 02 - Dune Messiah no longer counts as the book Dune.
  • Look for a better copy on a movie that is set to leave what you have alone now says that (Better versions is off for this movie) instead of "Nothing suitable was found right now". Movies start with that setting, so it was the usual reason nothing happened.
  • A title with no file of its own no longer downloads the same release again and again. When a finished download was kept as a decision because a file was already at its place (a movie or show you removed with its files left on the disk and then added again, or a file copied there by hand), the title stayed wanted with no file, so every later search fetched the release once more. The file that is there is now linked to the movie or episode and the history says so.
  • Adding an NZB or a torrent by hand without a release name now gets the quality right. A file called m2.nzb used to be named "m2" and show an Unknown quality. Mediarium now reads the names of the files inside the NZB (or the biggest file in a torrent) when the file name says nothing, and uses the one that tells more about the release. A name you type still wins.
  • If a file already exists now says in its help text, and in the choice itself, that Ask me each time only waits for you with movies. For shows and season packs Mediarium keeps the file you have, like Skip. The docs say the same.
  • A download that fails because one article is gone now says "1 article couldn't be found" instead of "1 articles".
  • Unpacking a .7z archive no longer fails with Cannot set file attribute ... Operation not permitted on a download folder that does not let programs set file attributes (a network share, a NAS or an NTFS/exFAT mount). 7z writes the file first and then fails on the attributes; when that is the only complaint, the file is now used. Any other 7z error still fails the unpack. Reported by JoshuaNCSU on GitHub, #189.
  • A half-finished copy no longer stays in your library for good. If Mediarium was stopped or the power went while a file was being copied into the library (another drive or a network share), the half-copied temporary file stayed behind, and every retry added another one. The next import of that title now clears the ones nothing has written to for an hour. The same goes for the temporary file of a hand-picked replacement.
  • Disk and file problems read as sentences. A full or read-only database, a file another program is holding open and a drive that reports a read or write error now say what to do instead of showing the database's own words. When Mediarium can't start because the port is taken, the config folder can't be written or the disk is full, the log says what to do about it too.
  • A very large page number or limit in the Radarr, Sonarr and SABnzbd style APIs, and on the music Discover page, no longer breaks the request. It is now just a page past the end.
  • A crash in one background job or one bad line from an IRC announce server now ends that job or that connection only, not the whole program. A huge Retry-After from an indexer is capped at the longest pause instead of overflowing.
  • Updating a library with a long blocklist no longer sits for minutes. One of the update steps (the one from 4.0.x that ties old album and book blocklist entries to their title) looked through every download and every history line again for each entry, so a blocklist of 20,000 entries with a few years of history took far longer than a restart should. It now uses small temporary indexes and takes under a second on that size, with exactly the same result.
  • An update that can't be applied says so in plain words. When one step of an update fails, Mediarium stops instead of running on a half-updated database, and now says which step it was, that the step was undone, that nothing is lost, and where the copy it saved before the update is.
  • Home Assistant and other dashboards that read SABnzbd's queue now get its status word (Downloading, Paused or Idle) next to the speed and the size, as SABnzbd sends it.
  • Apps that send the API key as Authorization: Bearer <key> now work with the Radarr, Sonarr and Lidarr style API, as they do with the real apps. The key still also works as the X-Api-Key header or ?apikey=.
  • Two albums of one very long name no longer share a folder. A long album title was cut at the end to fit the file system's limit, which took the year with it, so two editions of one long-titled album ended up in the same folder. The title is now shortened and the year kept.
  • Two bands with the same name, or an album and a single of the same name and year, no longer fight over one folder. The second one failed to import because "the files already exist". It now gets a folder of its own: the band's disambiguation or the start of its id after the name, and "[Single]" or "[EP]" after the title. Folders that already exist keep their names.
  • Indexer replies that were not clean XML no longer lose the whole page of results. A reply compressed without saying so, in Latin-1 or windows-1252, with a stray byte or a control character in a title, with a bare &, or cut off part way is now read (what arrived whole is kept). An empty reply, a JSON error or a line of plain text from a proxy now says what it was instead of "EOF". A Torznab item that lists several categories (2000 and 2040) keeps all of them.
  • Books: two works of the same title, author and year no longer share a folder, and Open Library, Google Books and MusicBrainz say what is wrong in plain words. A busy service, one that asks for fewer requests, one down for maintenance that sends a web page, and a Google Books request limit (which used to read as "your key was refused") each get their own sentence, and Open Library's answers are read up to a size limit.
  • A media server address that lands on a login page or a default web page no longer passes the test as an empty server. Silo and Audiobookshelf accepted any web page as an answer, so a wrong address or a proxy's sign-in page showed "connected" with no libraries. They now say the address does not look like that server. The same goes for an slskd address that answers with a web page.
  • A notification sent to an address that redirects to a login page is no longer reported as delivered. When the service answers a message with a redirect that turns it into a plain page request, Mediarium now says nothing was delivered and why.
  • Books follow "Start searching when a title is added". Adding a book always started a search, whatever Settings > Finding releases > Automatic searching said. The book dialog now has Start searching for it now, ticked or not by that setting like the movie, show and album dialogs, and the quick Add on the Books page follows the setting too.
  • Searching without a TMDB key says so. Without a TMDB API key, every search in the top box and on the Search page sent a request for movies and shows that was bound to fail, even when you were on the Books or Music tab. Mediarium now knows there is no key (tmdbKey in GET /api/modules), does not send it, and says where an administrator adds the key. Books and music search is not held up.
  • A failed release is still blocked when its id is missing later. A release that failed was blocklisted under the id its indexer gave it (its guid). When the same upload came back without an id, for example from another kind of search or after the indexer left it out, the entry did not match and the release was tried again. The entry now also keeps the indexer and size (or the info hash of a magnet link), and a release without an id is matched by them wherever the blocklist is checked: searches, the release lists, grabs and the reason shown in Activity. A copy with an id of its own that differs is still told apart by it. docs/activity.md.
  • A paused indexer is checked again by itself, and says so on its card. An indexer that failed three times in a row is left out of searches for 5 minutes up to 6 hours, and only the next search after that asked it again, so one whose key was refused for a while could stay out for hours while its Test button worked. About every half hour a paused indexer now gets the small connection check (one request, not a search), and when it answers the pause is lifted. An indexer that said you made too many requests is left to cool down as before. The card also shows a Paused tag next to the name, with the reason when you hover it, next to the "Paused until ..." line. Reported by SanketBaraiya and mirisbowring on GitHub, #131.
  • The page name in the top bar is no longer cut short on a laptop screen ("File names and import..."): the search box gives way first. The side menu stops showing a few letters of the page you are on next to Settings.
  • Red badges and counters (for example the Required badge in setup) are readable in the dark theme: the text is dark on the light red instead of white.
  • Hints and notes no longer leave one lonely word on their last line; they wrap in even lines.
  • On a phone each episode is a small card instead of a table that scrolled sideways.
  • The "That page doesn't exist" page has proper buttons, names itself in the browser tab, and back links above lists are easy to tap on a phone.
  • The About page listed one contributor twice; their ideas now sit under one entry.
  • On a phone, ebook, audiobook and music cards no longer hide their title behind the row of buttons; the buttons sit over the bottom of the cover.
  • On the dashboard, a film with a cinema date and a digital date (listed twice in the coming-up list) no longer makes two entries share one key, which could drop or repeat a line.
  • In the light theme the white-on-colour labels (the eBook ribbon, Read buttons, the Ended ribbon) are readable.
  • Chinese no longer scrolls sideways on a phone. The help line under a Settings page title (and any label with a help button) was kept on one line, and a Chinese sentence has no spaces to break at, so Quality, Folders, File names, Download rules and a few more pages ran wider than the screen. Only the last character now stays with the button.
  • The year next to a title stays readable on a phone (it was 8 px).
  • Backup dates, the statistics bars and the page names on the Settings overview no longer get cut off on a narrow screen, in German and Russian as well.
  • The Audiobooks tab name next to Library in the side menu is no longer cut short.
  • Activity: the Blocklist tab showed nothing but an error in demo mode; it shows two sample entries now.
  • A title or artist page that cannot load now offers Try again, the library and, when the TMDB key is missing, a link to the page where it goes, instead of a lone line of red text.
  • Discover on a new install without a TMDB key said nothing under Movies and TV shows. It now says a key is needed and where to put it, and a list that could not be read offers Try again.
  • The quality profiles table no longer breaks "Stops upgrading at" and the quality names in the middle of a word on a wide screen.
  • The setup steps no longer break a button in the middle of a word on a phone (for example "Back" and "Skip for now" in Russian): the buttons wrap as whole buttons.
  • Demo mode: an artist page no longer asks the server for a discography that the demo does not have.
  • Screen readers and keyboards. A box inside a label that also holds an info button ("Videos per check (i)", "Priority (i)") had no name for a screen reader, because the button came first; the label now points at the box. Two headings that skipped a level on the Requests and Linked services pages are fixed. A notice's text now stays on screen longer when it is longer, and stays while a button inside it has focus.
  • Phones. Small toggles (the details rows), links in lists, credits and the dashboard now have a 40 px target without moving anything. The count on the bottom bar is readable in the light theme, and the NEW tag in the logs in the dark theme. Cards that rise in as a list loads now stay still when your system asks for less motion. In high-contrast mode the switches, the chosen tab and chip, and the focus ring are drawn with system colours instead of disappearing.