hydrusnetwork/hydrus v627

Version 627

on GitHub

misc

• windows that remember position are better about saving that position on a window move. some window managers (particularly on Linux) were not recognising the previous 'window move has ended' tech, nor non-mouse moves, so I wrote something that will work more generally. let me know if any windows start positioning crazy this week during drags!
• fixed variable framerate render timing in the native viewer for animated webp files. I messed up some logic when I first rolled this out in v620; should be working correct now (issue #1749)
• fixed an issue where import sidecars were losing their texts' order (they were being pseudorandomly re-sorted) in the outer layer of string processing where multiple sidecar texts are combined. the sidecar pipeline now preserves original sort throughout the pipeline (issue #1746)
• added a TEST: Stop mpv before media transition in an attempt to early-cancel laggy mpv when transitioning media when the current (looping and pre-buffering) media is near the end. may also help some situations with laggy storage, but we'll see. I think this mode might have some load bugs

visual duplicate detection is ready

• the 'A and B are visual duplicates' system now compares spatial edge data and is better able to recognise mosaic/blur patches, minor alterations or repositions, and subtle alternates in busy areas like eye pupils. all the remaining difficult false positives I have are fixed, and the test easily catches several pairs that were previously tricky, allowing me to remove a hacky test on the colour testing side of things that was producing some false negatives
• this test is now available as a comparator in duplicates auto-resolution rules! you can choose it to be at least 'very probably', 'almost certainly', or 'near-perfect'. I default to and recommend 'almost certainly' for now. I am still interested in seeing any new false positives you encounter
• note that this tool is CPU heavy--expect about a second per pair tested! the UI seems to be holding up well with the delay, but I've added a new label to the preview panel to show when it is running slow and there are pairs still to be tested
• added a new and generally cautious 'visually similar pairs - eliminate smaller' suggested rule. it does an 'A and B are almost certainly visual duplicates' test on similar filetypes, preferring an A that is larger in size and larger or equal in resolution, and is semi-automatic. if you have been following all this, please give it a go and let me know what you think
• assuming this tech holds up, this was the last big difficult thing in my auto-resolution superjob. there is still a lot I want to do, but it'll mostly be cleaner UI, some more comparator types, a way to interact with the denied pairs log, smoother counting and pair load in more places, a way to load up a duplicate rule's pending decisions in the normal filter, richer duplicate-aware pair preview/review, that sort of thing

cleanup/docs/boring

• added some notes to the 'migrate' and 'source' help regarding needing to rebuild your venv every time you move the install folder
• added some 'better software for x' links to 'getting started with files' to ComicRackCE, Lanaragi, and Calibre
• added a little about 'A and B are visual duplicates' to the duplicates auto-resolution help
• the 'LabHistogram' stuff that does the new 'A and B are visual duplicates' is now renamed generally to 'VisualData', and the LabHistogram stuff is rejiggered into a subclass
• fixed a small warning bug when the thumbnail pair list in the auto-resolution preview panel is updated to have a new rule while it is empty
• fixed a duplicate file system bug where it wasn't able to negotiate the correct media id for a file when the duplicate group had an identity definition but somehow was desynced and did not list the king as a member
• added a new 'pixel-perlfect jpegs vs pngs - except when png is smaller' alternate suggested rule that does pixel perfect jpegs and pngs but only actions when the jpeg is smaller than the png
• wrote some unit tests for 'A and B are visual duplicates'
• I think I silenced OpenCV's warning logspam. if you ever saw [ WARN:44@43420.823] global grfmt_png.cpp:695 read_chunk chunk data is too large kind of thing in your log, you shouldn't any more

future build

• I am making another future build this week. This is a special build with new libraries that I would like advanced users to test out so I know they are safe to fold into the normal release.
• it is here: https://github.com/hydrusnetwork/hydrus/releases/tag/v627-future-01 . if you are experienced and would like to help me, please check it out
• the new numpy version that caused trouble for Arch users also broke our build right afterwards, so this is an attempt to fix that while rolling in some security updates. updating PyInstaller more than a year to the same numpy fix they rolled out two weeks ago did not, I was surprised to discover, appear to break anything, so the builds may be significantly more stable and OS-compatible. as usual, I will be interested to know if anyone on Win 10 or any other older OS has trouble running this. as far as I can tell, a clean install is not required
• the specific changes this week are--
• On All--
  • requests moved from 2.32.3 to 2.32.4 (security fix)
• On Linux and Windows--
  • PyInstaller moved from 6.7 to 6.14.1 (handles new numpy)
  • setuptools moved from 70.3.0 to 78.1.1 (security fix)
  • numpy moved from 2.2.6 to 2.3.1 (lots of improvements, py 3.11+)
• On Windows--
  • sqlite dll moved from 3.45.3 to 3.50.1
• Source venvs--
  • numpy moved from >=2.0.0 to <=2.3.1