hydrusnetwork/hydrus v599

Version 599

on GitHub

misc

• some users ended up getting a crazy wide duplicates page after the recent mix-up with the new auto-resolution multi-column list's ID. on update, the list will be reset to default widths, which should fix the rest of the duplicates page. let me know if you have any more trouble! (issue #1625)
• the e621 downloader is fixed to find search results again (I updated the gallery parser to reflect their recent html changes). I understand they may be updating again soon, so let me know if anything breaks again (issue #1628)
• I cleaned the media viewer hover window show/hide and size/position code a bit more and think I reduced more layout flicker, particularly the 'do another adjustment layout right after showing' issue the notes hover often does when super tall. I might have also fixed the 'ok I guess I am going to sometimes initialise as a super wide guy in the middle of the viewer and when the user moves the mouse over my ghost I will flicker for one frame before moving where I belong' issue too
• the timestamps in the media viewer top hover window and all the timestamp lines in the media right-click menu's top-row submenu now have a tooltip that has times with the reverse of your 'always show ISO times' settings. if you see '2022-11-20 14:39:52', it'll say '2 years ago' on the tooltip, and vice versa!
• when you hit "open files in a new duplicates filter page", the file domain is now set explicitly to "all my files". if you know what you are doing and need to turn this off, there's a new checkbox under options->duplicates
• new duplicates filter pages now also start on "all my files" rather than your default fallback file domain. "all my files" is the ideal default duplicates search context bros
• if you have an OR predicate under construction in a file search tag autocomplete (use Shift+Enter when you enter stuff), hitting Escape on an empty text input now 'rewinds' it one predicate at a time before cancelling it entirely
• animations that report they have exactly 100fps are now distrusted and their frames are counted manually. this 'count the frames manually' routine will now trigger generally in more cases and on larger files
• all files with ~100fps will be metadata-rescanned on update

new vacuum tech

• the vacuum in the client and server now uses VACUUM INTO SQLite tech. rather than writing a copy of the db to your temp dir and then to the WAL and then writing that 'commit' into the original file (which it seems is actually what happens--there's no atomic filename swap at the end), we now vacuum to a fresh file beside the original, no temp dir or WAL gubbins needed, and do a filename swap afterwards. this appears to work significantly faster than the old method, with the only caveat that there is a very brief dangerzone where neither file is named what we need, so if the hard drive disconnects or similar in that 2ms window, hydrus cannot automatically recover itself
• all operations err on the side of failsafe, and I have added copious error handling code to navigate all the possible problems. if anything does go wrong, the user will be presented with a record of what happened, and in the case where hydrus could not fix itself, how they can fix it
• the various 'do we have enough space to vacuum?' and 'how long will it take?' tests are adapted to the new rules. it doesn't use the temp dir any more, so if you have been struggling to find system drive space for client.mappings.db, this is no longer a problem
• when the job is complete, the log message, which previously just said the time taken, now records the size change and the bytes per second, and this message is also now thrown into a popup. on my dev machine SSD, I have seen 170MB/s on database files that are in memory and ~30MB/s on a 4GB file not in memory. I guess we now asymptote to about 10MB/s on a superhuge file. I am interested in what users see in their different situations
• If you are on an HDD and have a big db, you still have no hope, its over
• thanks to the user who suggested this option

duplicates auto-resolution

• the duplicates auto-resolution system is moving forward. still lots to do, but I'm still feeling good about it all
• fleshed out the objects more--most stuff is now serialisable and has typedefs, getter/setters, and summary generation methods
• connected most of the decision pipeline together, and we are basically ready to process our initial jpeg/png pixel pairs
• did a hair more UI
• my taglists now handle setEnabled calls properly, so the stub UI in the duplicates auto-resolution panel is now properly un-editable

boring cleanup

• overhauled the way medias produce the various nice info strings on the right-click menu top-row submenu (and some other places). rather than a tangle of tuples, there's a couple simple classes being passed around that can do tooltip overrides and stuff. I also cleaned up the code around here generally
• the main gui window status bar info when you only have one file selected is now much simpler. I previously piped the 'interesting' file info lines to it, but it too often ended up a spammy huge long line--not a good summary!
• rewrote the 'always show iso time' solution from the old BaseMethod-to-Method replace trick to a simpler and saner global bool, as I recently did with some PIL/ICC settings
• import folders now 'action' their original files immediately after import is done. previously they would do it in batches of ten, and if the import folder were interrupted by something like program shutdown, they'd have to wait for the next run to be cleared. I'm not totally sure, but I also think import folders set to 'ignore, not try again' on a large number of files may run a bit faster now
• clarified the additional ways to import downloaders in the Lain import dialog. Lain's paste button now also accepts URI-aware file paths (i.e. if you select some files in your file explorer and tell your OS to 'copy') as the clipboard source
• converted some duplicates processing code, and, relatedly, some delete-lock reporting stuff, from the being-overhauled MediaSingleton to MediaResult. this allowed me to clean up some wew code in duplicates auto-resolution, the duplicates filter, test code, and the Client API
• updated the running from source help to talk about mpv/libmpv on Linux
• fixed some bad panel/dialog calls into nicer Qt signals
• added some typedefs to clear out about a hundred more PyUnresolvedReferences that PyCharm found. mostly custom widget calls