hydrusnetwork/hydrus v589

Version 589

on GitHub

misc

• the similar-files search maintenance code has an important update that corrects tree rebalancing for a variety of clients that initialised with an unlucky first import file. in the database update, I will check if you were affected here and immediately optimise your tree if so. it might take a couple minutes if you have millions of files
• tag parent and sibling changes now calculate faster at the database level. a cache that maintains the structure of which pairs need to be synced is now adjusted with every parent/sibling content change, rather than regenerated. for the PTR, I believe this will save about a second of deferred CPU time on an arbitrary parent/sibling change for the price of about 2MB of memory, hooray. fingers crossed, looking at the tags->sibling/parent sync->review panel while repository processing is going on will now be a smooth-updating affair, rather than repeated 'refreshing...' wait-flicker
• the 'the pairs you mean to add seem to connect to some loops' auto-loop-resolution popup in the manage siblings/parents dialogs will now only show when it is relevent to pairs to be added. previously, this thing was spamming during the pre-check of the process of the user actually breaking up loops by removing pairs
• added an item, 'sync now', to the tags->sibling/parent sync menu. this is a nice easy way to force 'work hard' on all services that need work. it tells you if there was no work to do
• reworked the 'new page chooser' mini-dialog and better fixed-in-place the intended static 3x3 button layout
• showing 'all my files' and 'local files' in the 'new page chooser' mini-dialog is now configurable in options->pages. previously 'local files' was hidden behind advanced mode. 'all my files' will only ever show if you have more than one local files domain
• when a login script fails with 401 or 403, or indeed any other network error, it now presents a simpler error in UI (previously it could spam the raw html of the response up to UI)
• generally speaking, the network job status widget will now only show the first line of any status text it is given. if some crazy html document or other long error ends up spammed to this thing, it should now show a better summary
• the 'filename' and 'first/second/etc.. directory' checkbox-and-text-input controls in the filename tagging panel now auto-check when you type something in
• the 'review sibling/parent sync' and 'manage where tag siblings and parents apply' dialogs are now plugged into the 'default tag service' system. they open to this tab, and if you are set to update it to the last seen, they save over the value on changes

default downloaders

• fixed the default safebooru file page parser to stop reading undesired '?' tags for every namespace (they changed their html recently I think)
• catbox 'collection' pages are now parseable by default

boring list stuff

• fixed an issue with showing the 'manage export folders' dialog. sorry for the trouble--in my list rewrite, I didn't account for one thing that is special for this list and it somehow slipped through testing. as a side benefit, we are better prepped for a future update that will support column hiding and rearranging
• optimised about half of the new multi-column lists, as discussed last week. particularly included are file log, gallery log, watcher page, gallery page, and filename tagging panel, which all see a bunch of regular display/sort updates. the calls to get display data or sort data for a row are now separate, so if the display code is CPU expensive, it won't slow a sort
• in a couple places, url type column is now sorted by actual text, i.e. file url-gallery url-post url-watchable url, rather than the previous conveniently ordered enum. not sure if this is going to be annoying, so we'll see
• the filename tagging list no longer sorts the tag column by tag contents, instead it just does '#''. this makes this list sort superfast, so let's see if it is super annoying, but since this guy often has 10,000+ items, we may prefer the fast sort/updates for now

client api

• the /add_files/add_file command now has a delete_after_successful_import parameter, default false, that does the same as the manual file import's similar checkbox. it only works on commands with a path parameter, obviously
• updated client api help and unit tests to test this
• client api version is now 70

more boring cleanup

• I cleaned up a mash of ancient shortcut-processing jank in the tag autocomplete input and fixed some logic. everything is now processed through one event filter, the result flags are no longer topsy-turvy, and the question of which key events are passed from the text input to the results list is now a simple strict whitelist--basically now only up/down/page up/page down/home/end/enter (sometimes)/escape (sometimes) and ctrl+p/n (for legacy reasons) are passed to the results list. this fixes some unhelpful behaviour where you couldn't select text and ctrl+c it unless the results list was empty (since the list was jumping in, after recent updates, and saying 'hey, I can do ctrl+c, bro' and copying the currently selected results)
• the key event processing in multi-column lists is also cleaned up from the old wx bridge to native Qt handling
• and some crazy delete handling in the manage urls dialog is cleaned up too
• the old EVT_KEY_DOWN wx bridge is finally cleared out of the program. I also cleared out some other old wx event definitions that have long been replaced. mostly we just have some mouse handling and window state changes to deal with now
• replaced many of my ancient static inheritance references with python's super() gubbins. I disentangled all the program's multiple inheritance into super() and did I think about half of the rest. still like 360 __init__ lines to do in future
• a couple of the 'noneable text' widgets that I recently set to have a default text, in the subscription dialogs, now use that text as placeholder rather than actual default. having 'my subscription' or whatever is ok as a guide, but when the user actually wants to edit, having it be real text is IRL a pain
• refactored the repair file locations dialog and manage options dialog and new page picker mini-dialog to their own python files