hydrusnetwork/hydrus v600

🎉 Version 600 🎉

on GitHub

misc

• fixed a typoed type def that caused a boot error in python 3.10. this also sometimes/always broke the v599 macOS App, which I yanked. I have improved my testing situation to catch anything like this in future (ironically, I accidentally stopped booting the program in py 3.10 every Wednesday just recently when I stopped testing for Qt5) (issue #1630)
• brought back the 'show more text about one file on the status bar' functionality with the new 'show additional status bar text when a single thumbnail is selected' checkbox under options->thumbnails. this tech works with single-file collections now, too. this checkbox is default off, so go hit it if you miss this!
• the 'manage times' dialog now also shows '(3 days ago)' accompanying string variants with all the times displayed. time objects can get complicated, so let me know how this works out if you have weird times from the future or whatever
• if your similar files search progress is greater than 99%, the tab no longer says (needs work). don't worry about that tiny little bit on every file import lads, the ride never ends
• the 'open similar files in a new page' file menu now has a 'custom' entry that spawns a simple spinner to choose a different hamming distance. it starts at 10 and has step 2 (i.e. 10, 12, 14...). related: I understand there are sometimes small file count differences between a hamming distance of (even number) n and (odd) n+1, but it typically isn't significant. I force a hamming distance step of 2 in some places and allow a step of 1 in others. if you care about this stuff, would you be ok with me enforcing step 2 everywhere, or do you have use for odd-number hamming distance searches?
• the various core file copy/move operations used in hydrus are now wrapped in retry modes for BlockingIOError, which can be raised by NASes and the like that are under deep stress. these will be retried five times, with ~1-3 second delays between re-attempts, before raising the Exception as before. thanks to the user who pointed out this could even happen. let me know if anything more complicated, let's say folder mirrors/merges, can still trip the problem
• the 'analyze' database maintenance job's yes/no dialog is now a yes/yes/no dialog. the 'soft' and 'full' options are green, the cancel is red
• cleaned up how hidden or non-functional galleries are chosen in the gallery-selector. the respective entries now have ellipses, and if there is only one item to select, it still shows the 'select from list' thing rather than confusingly insta-selecting that item
• collections now sort by anything involving width and height by using its largest-num-pixels image as proxy. this isn't perfect, and it is invisible, and I suspect in particularly crazy situations the sort may change depending on the previous sort, but collections should at least do something here now rather than always counting as 'none' for their dimensions. a thought was floated to optionally sort collections by 'average' file property in places, which may be another avenue to explore here--for instance, although sum num_pixels might make some sense, sum of height doesn't
• the 'sort by dimensions' submenu now lists width first, then height. a three-part hack was needed to make this work lol

more vacuum and db stuff

• last week the vacuum stuff went well, but I forgot to update the dialog to use the new 'do we have enough free space to do a vacuum?' check! the dialog now properly uses the new check and won't moan about you not having enough temporary space
• if you try to do a vacuum with an external program connected to the database file we want to vacuum (it tests this by looking for '-wal' or '-journal' journal sidecars), the program now abandons the vacuum attempt and tells you to disconnect and try again (or switch to WAL, if you are in TRUNCATE). one user did this by accident this week, with a pending write commit from the external program, and it caused malformation!! hydrus will now not let this happen
• if the client or server have a failed vacuum due to a file rename failing and do not fix it themselves, and the program thus next boots with a db file with the db_path.prevacuum filename, the program now recognises this, tells you about it, automatically recovers from the situation, and, if one exists, tells you what to do with the db_path.vacuum file
• the 'do we have a missing database file?' initialisation check is improved: it will now report on a mix of missing main and external files; it will now differentiate between the first connection attempt and subsequent normal reconnections (and only run 'create db' checks and so on on the first attempt); and it will recognise if a database file suddenly goes missing during program operation and trigger an immediate program halt after its popup
• the various places the database can trigger an immediate program halt are now formalised into one careful method in a special place that won't be accidentally shuffled around or called via typo
• I rejiggered the 'let's map our incoming db command to the bound method call' process and finally replaced the comp-sci-tier 80-line-tall if/else towers in the main ClientDB read/write methods. it is all now a much nicer dynamically initialised name->method dict. I don't suppose it really saves all that much overhead per call, but it is finally done
• the database now reports 'db committing' in the main gui status bar when it does this in its 'idle' time. previously it only reported after a big job overran the due time

more hover windows

• I may have fixed another hover window position bug (most probably certain window managers in Linux only, where sometimes windows will defer geometry updates until they are shown), where, before being shown any other time in that media viewer, it could flash into place in the correct position for one frame despite the mouse not being over it
• simplified the hover window show/hide logic a bit more and removed an unusual hack that handled Window Managers that weren't happy about taskbarless top level hover windows appearing (because hover windows are no longer top level but just internal widgets to the normal media viewer)
• re-integrated some spammy hover window reporting text into the 'hover window report mode', and rejiggered the logic to ensure the critical show/hide calls here are not so spammy

client api

• added a new command, /manage_database/force_commit, which immediately and synchronously commits the database, flushing all pending (savepointed) changes to disk
• added help for this: https://hydrusnetwork.github.io/hydrus/developer_api.html#manage_database_force_commit
• the client api version is now 76

debugging

• added a new help->debug->report modes->file sort report mode, which spams the thumbnail sorts going on with all the sort_key results it saw. we're trying to figure out some weird namespace sort, so let's see what this generates
• file and search logs have a new advanced menu item, tucked under the 'whole log' submenu, to export the current selection to the clipboard in JSON serialised format

boring cleanup

• fixed some more unresolved references caused by Qt enums

new mpv and sqlite on Windows

• I am rolling in two new dlls for Windows today, for SQLite (database) and libmpv (video/audio player)
• SQLite is updated from 3.45.3 to 3.47.0
• mpv is updated from 202-08-20 to 2024-10-20
• both dlls are mostly just bugfixes and performance improvements, but the mpv release is slightly special--in the 'future test' we ran a few weeks ago, users with unusual Windows, be that Windows Server, under-updated Windows 10, or Windows 10 on a VM, might see a grid of black bars over some webms. no one on Windows 11 or normal updated Windows 10 reported any problems. the new mpv does perform much better than the older, and I am told it fixes some gif bugs, so I do want to update, but I do so hesitantly. if many users on updated/normal Windows do run into trouble with this release, I expect to roll back again. in the meantime, I have updated the 'running from source' help to talk more about stable versions of mpv on older Windows. users who need to keep their OS under-updated are now recommended to run from source: https://hydrusnetwork.github.io/hydrus/running_from_source.html