Version 681

misc

• status bar 'busy' indicator is simpler, and the tooltip now says the number of threads working
• wrote a catch to stop QSS stylesheets from specifying a hydrus property colour with transparency. these were drawing crazy, with noclip garbage left on bitmaps used for transparenty images and animations in the media viewer. also fixed the e621_redux stylesheet, where this issue was discovered, to simply set a background colour for the media viewer

more human-readable embedded metadata filtering

• I further cleaned up what hydrus considers 'human-readable embedded metadata'. many more technical fields are now hidden from the little text panel, meaning when this guy does pop up, or you search 'has non-EXIF embedded metadata', you'll now tend to see something nice and human like 'Artist' or 'Title', or otherwise rich like an AI prompt or Character card
• I now invite advanced users to do some 'determine if the file has non-EXIF embedded metadata' maintenance regenning on some 'has non-EXIF embedded metadata' files and tell me what technical/inhuman text remains in their collections. I'm getting pretty happy here and I expect to (optionally) schedule a big regen for all users pretty soon
• the newly excluded fields are: 'software', 'transparency', 'bit_depth', 'duration', 'primary', 'default_image', 'Creation Time', 'create-date', 'modify-date', 'date:create', 'date:modify', 'chroma', 'loop', 'timestamp = 0', 'extension', 'photoshop', 'bbox', 'blend', 'mpoffset', 'disposal', 'interlace', 'aspect', 'sizes', a ton of 'Thumb::' garbage, 'iptc', 'Raw profile type exif', 'Raw profile type iptc' (ye olde XMP), 'xmp', 'XML:com.adobe.xmp', 'adobe', and 'adobe_transform'. a bunch of these are format specific DPI data and such and not useful to elevate to the user. the IPTC and XMP stuff will come back when I expand EXIF parsing
• excluded 'comment' or 'Comment' from human-readable metadata fields if it matches a handful of 'Created with X' style regexes; this is now parsed as a 'software' field
• the 'Creator' and 'Source' fields are now appended to the 'software' field
• excluded any human-readable metadata fields that have null value or empty list/dict
• I realised while doing this that maybe going for a whitelist of what we do want would be saner, but I think I prefer, for completeness and fun, to get everything and actually deal with the sore thumbs
• put 'software' metadata field in the 'extra info' panel below, as 'source' ('source' rather than 'software' because sometimes stuff like Kodak blah blah comes through here). there's an argument to include this in human-readable fields, if we are keeping artist/title, but I'm falling on the side of extracting it. I like this field a lot as human-valuable and it'll be high priority to be its own thing when we add richer metadata storage/search
• similarly, the 'create-date' and friends are cool and I'll figure out some way to pipe that to a 'file metadata' modified time 'domain'. you can't generally get any better than the file metadata for this
• if a file has a 'chroma' field talking about '420' subsampling, this is now converted to a subsampling field in the 'extra info' panel, like for jpeg
• optimised 'has human-readable metadata' test for a fast yes in many cases

new graphicsview thumbnail grid test

• the new 'thumbnail rendering tech' TEST under options->thumbnails is fixed up a bit and ready for advanced users to tentatively try out. I think I have it doing pretty much everything the existing code does without complaint, but I would like to confirm that with a broader test. it only applies to new pages, so I recommend first opening a new page and do some of your normal stuff before you test your whole session with a full client restart
• fixed 'collect-by' transitions, removing the ghost singles that hung around
• fixed a late waterfall thumb keyerror when transitioning through collect-by states
• fixed collections not redrawing for a new first media on sort-by changes
• fixed updating the status bar on any collect-by change, in either old or new thumbgrid code. it now flips between '200 files' and '200 files in 12 collections' when you change the collections
• fixed drawing of thumbs at higher than 100% UI scale (the bitmaps were cropped due to DPR fun)
• fixed a blurhash thumb fallback route in the new test
• shaved the test animation time down for snappier feel. I'm pretty sure the old animation is accidentally heavily curved so despite ostensibly being 0.5s it was visually done faster. new guy is now ~13 frames at 60Hz
• did some KISS refactoring of this code

more de-laggification work

• I hacked in a test last week to try a different way of scheduling jobs in the background. it worked very well and I seem to have found the cause of the 'I have a fat client and when it does heavy work like archive/delete filter commit, everything slows down a lot' issue. basically, when certain sorts of heavy work was going on, threads waiting on new work to do or database jobs to come back were getting into a kind of traffic jam and wasting lots of time figuring out who had to go first rather than letting the guy who was actually doing the work go ahead and do it. I have now replicated and cleaned my work to several places and big clients that do a lot should be a good bit less laggy when big things are going on
• cleaned up the successful CallToThread anti-CPU-thrashing test, taking out older code and cleaning the loop of it
• added anti-CPU-thrash tech to db mainloop
• added anti-CPU-thrash tech to threads waiting on the db to finish current work
• added anti-CPU-thrash tech to database jobs; when many threads are competing for the db, they'll wake less
• added anti-CPU-thrash tech to network job waits; when many threads are competing for network resources, they'll wake less
• added anti-CPU-thrash tech to pubsub, both in the managing daemon and any threads waiting on 'background work to free up'
• added anti-CPU thrash tech to 'are we currently rendering any thumbnails?' wait, which a bunch of threads wait on before going ahead with their own work
• added anti-CPU-thrash tech to the core read-write lock used in the filesystem. I was very careful here as this is not my forté, and tests are good. this should improve latency, particularly thumb and file loading latency, when many file imports are going on and the locks are churning, which is another place we've seen inexplicable lag during certain busy periods
• retired the debug 'db ui-hang relief mode'. this was an interesting experiment but the incorrect solution to the problem of GUI threads waiting on the db

boring de-laggification stuff

• wrote a new thread worker pool object to look after calltothreads
• primary controller and test controller now use this same object to manage threads (test guy had its own hacky code before)
• tightened up db mainloop shutdown--to assure that db jobs can wait forever safely, we now guarantee that every job posted will get explicitly woken with a result through db shutdown
• network engine now guarantees every network job it gets will be done'd so waiting threads shall all be woken up on program shutdown
• rejiggered some 'is it time to commit?' in db idle time stuff given the changes
• cleaned up some thread status-checking locking, mostly in debug code
• KISSed some network job status stuff, and removed some confusing shutdown detection from the core 'isdone' calls
• removed defunct db I/O-locking error-handling code
• pubsub takes work breaks more often
• pubsub system now has a shutdown signal to tidy up waiters properly on shutdown
• thumbnail caches now have an explicit shutdown signal and clean up after themselves better on shutdown
• cleaned up some misc thumb cache code
• misc shutdown code cleanup
• reordered some controller shutdown code
• misc cleanup of controller wait code
• removal of spurious event flags in db and pubsub

future build committed

• This release commits the changes tested with the recent future test build, which went well
• the summary for this update is: loads of stuff
• there are no special instructions for the update. update as normal
• since library versions have been bumped, users who run from source will be encouraged to rebuild their venv on update this week.
• the library changes are--
  • PySide6 (Qt) from 6.9.3 to 6.10.3
  • OpenCV (opencv-python-headless) from 4.12.0.88 to 4.13.0.92
  • beatifulsoup4 4.14.3 to 4.15.0
  • cbor2 6.1.1 to 6.1.3
  • cryptography 48.0.0 to 49.0.0
  • Pillow 12.2.0 to 12.3.0
  • pillow-heif 1.3.0 to 1.4.0
  • pillow-jxl-plugin 1.3.7 to 1.3.8
  • pyopenssl 26.2.0 to 26.3.0
  • service-identity 24.2.0 to 26.1.0
  • numpy 2.3.1 to 2.4.6 (this jump was a bit of a juggle, and now possible with OpenCV 4.13)
  • requests 2.33.1 to 2.34.2
  • SQLite dll and sqlite3 terminal updated on Windows from 3.51.2 to 3.53.3
  • SQLite terminal executable 3.53.3 now added to Linux and added to new db folders just like the Windows extract
• the 'test' library changes are--
  • PySide6 (Qt) from 6.10.3 to 6.11.1
  • OpenCV (opencv-python-headless) from 4.13.0.92 to 5.0.0.93. I did some research, and this upcoming jump doesn't seem to be a super significant change for our purposes. mostly C++ cleanup and finally dropping python 2 support.
• other stuff--
  • db folder is now completely masked in gitignore
• build environment--
  • updated the workflow actions versions across the board
  • Windows and Linux builds are moved from python 3.12 to 3.13
  • pyinstaller 6.16.0 to 6.18.0
  • the builds now share the core pyproject.toml rather than using their own requirements.txt files
  • pywin32 in Windows build env from 311 to 312
  • Windows ffmpeg (GyanD) 8.1.1 to 8.1.2
  • Linux build now comes with ffmpeg (BtbN) 2026-06-30, which I think covers 8.1.2