hydrusnetwork/hydrus v677a

Version 677

on GitHub

misc

• fixed an issue with last week's downloader metadata overhaul that broke some downloaders. specifically, when downloaders had a post with one file url and that file url was changed through normalisation (typically through a URL Class), metadata application was not working. if you had a twitter-type downloader that seemed not to add tags in v676 when there was only one file in the post, please queue up those downloads again in a new urls downloader page and they will get their tags and so on
• the 'min time to view a file in x' settings in options->file viewing statistics are now minimum 50ms (previously 1s)
• when doing 'special duplicate' on a shortcut with F12 as the key, it now jumps to F13 (issue #2042)
• reduced the overhead on two important 'wait a moment' checks that many of the thread workers consult. previously, when checking if the pubsub or db were busy, threads would wake too frequently in busy periods and thrash as they competed for 'is busy' locks. now the pubsub and db themselves maintain a single 'I am idle' signal the waiters can wait nicely on without needing extra checks
• the code that terminates and then kills a timed out subprocess (ffmpeg, typically) now catches permission errors better (issue #2046)
• fixed a stupid typo from the non-interactive setup_venv.py mode that broke the 'advanced' manual, interactive install. I was so focused on the new thing, I didn't test the old thing to see it still worked

client api

• fixed the new /manage_pages/new_page command for a 'page of pages' type and updated the unit test to catch this (issue #2044)
• client api version is now 94

human-readable metadata fix and improvements

• a user noticed my recent 'chara' file metadata parsing was causing issues for files with metadata that included non-text datatypes and spotted where it was happening. just a stupid logical typo that was causing a bunch of filres with human-readable metadata to parse as not having any at all. I have fixed this issue so these files will show up with metadata again, and it will render correctly
• any image imported after june 2nd (v674) will get a 'has human-readable metadata?' rescan
• I have hidden the 'progression/progressive' keys from human-readable metadata presentation; we scan this elsewhere and show it as 'progressive? yes/no' on the same panel. I think I'd like to do the same for some other stuff like the 'jfif' and 'dpi' gubbins you often see

modern animations and ffmpeg

• tl;dr: fixed some video parsing bugs with the new ffmpeg 8.x.x. avifs and heifs with num_frames=1 should be fixed, you do not have to do anything
• many users, including all windows built users, have been on ffmpeg 8.x.x for a while now, and my video metadata parser recently broke for the animated 'sequence' variants of AVIF, HEIC, and HEIF. these files were being parsed with a num_frames of 1 and rendering as just a still image or throwing an error depending on the renderer
• hydrus now recognises when any video file has multiple tracks, and if one track seems to be just a still image file, it selects the true animation stream for fps calculations and so on
• also, the native renderer will now recognise this situation; if there seems to be no second frame during a render run, it will inspect the file closer to see if there is a different video track to select
• also, I updated the deprecated vsync ffmpeg call to be fps_mode. this was another source of errors for various native rendering with modern ffmpeg
• also updatedk the old -s to a combined -vf line with optional crop
• also updated the overcomplicated -f image2pipe to -f rawvideo. works the same, but it is semantically better and may fix some frame timings
• also updated the core metadata parse routine to no longer render the first second of the vid at small scale since this is an old hack that burns CPU but isn't used for anything any more and crops resolution parsing to 120 height for certain formats in ffmpeg 8.x.x (old mpegs at least, by my test)
• all animated AVIF, HEIF, and HEIC files are scheduled for a metadata reparse (issue #2041, #1891)

new thumbgrid drawing tech

• I found some time to work on the new thumbgrid test. it is better integrated and I fixed some bugs, but there's still some stuff not working so I'll hold off on the wider test
• so, just as a record, I did--
  • fixing up some type hints
  • misc refactoring
  • undoing thumbnail movable/selectable flags to stop some inherent Qt behaviour stepping in on mouse events (this fixes ctrl+click selection, which was being pseudo-randomly undone I think in the QGraphicsScene event handling)
  • thumbs now have their own selection bool
  • moving click-event/selection responsibility to the GraphicsView since thumbs don't do anything but call the parent atm anyway
  • fixed thumb resolution stuff for non-resolution-having media when cache entry is invalidated
  • made thumbnail 'media' (and the new 'is_selected') bools public
  • tiny bit of thumb gen optimisation