- build with c++26 and c23, fixes warnings due to this change.
- use #embed over romfs where applicable.
- load all configs upfront in the app menu, massively improves boot time
- enable boost mode during load/scan time in all (slow loading) menus, huge load time improvement.
- enable boost mode when exiting the app, to speed up closing all the menus and saving the config.
- reduce the size of the nro nacp when loading to just the title + author + display version.
- add option to enable boost mode for all progress bar menus, huge perf improvement again.
- remove unused launch_count var from the playlog file.
- display full path when dumping.
- optimise appstore unzip code by iterating through the zip rather than finding a specific file, reduces retroarch extract time from 52s to 26s.
overall, this commit has reduced boot time from 0.4s to 0.3s and massively increased load times of other menus.
(it also reduced the binary size by 4kb, so yay)
due to the previous commit, i broke dumping multiple games via usb as the stream offset wasn't reset.
because of this, the first transfer would complete, but the 2nd one would fail.
Some notes i made when adding stream support:
The tinfoil API makes it hard / impossible to multi thread the file upload because each data transfer passes the the file name and offset, meaning that it can and will change files and offset in the middle of a transfer
So that prevents the read thread from running freely in the background and the pull thread pulling data when requested.
The extension adds a flag to the usb header which if set, enables stream mode (same as ftp installs). This removes random access, but allows for multi threading as the data will be requested in order.
previous uploads were all single threaded, which meant that it only uploaded as fast as the slowest source.
usb transfer is still single threaded due it being random access for both files and data, making it
hard for the read thread to run freely.
turns out that the event is auto cleared when waited.
this meant that if sphaira handled the event before qlaunch got chance to handle it,
then qlaunch won't update when app records changed.
this can result in a gamecard not being mounted, deleted games still apearing, installed games
not being displayed etc...
- added support for custom upload locations, set in /config/sphaira/locations.ini
- add support for various auth options for download/upload (port, pub/priv key, user/pass, bearer).
- added more es commands.
- fixed usb install potential hang if the exit command is sent, but the client stops responding (timeout is now 3s).
- added multi select to the games menu.
- added game dumping.
- added switch2switch support by having a switch act as a usb client to transfer games.
- replace std::find with std::ranges (in a few places).
- fix rounding of icon in progress box being too round.
- fix file copy helper in progress box not updating the progress bar.
on fw 19 and below, loading from cache takes ~5ms, whereas manually loading takes ~20ms.
manually loading is still faster than relying on ns to load control from storage (~50ms).
- loading the control data is ran on its own thread, it does not block the main thread. allows for smooth scrolling like nintendos home menu.
- on fw20+, sphaira manually parses the control data, rather than using ns. manually parsing takes 20-40ms, which is faster than ms which can take 50-500ms.
- on fw19 and below, if the control data is not in ns cache, sphaira will manually parse the data as its twice as fast as ns. You can see how fast this is by loading the gamecard menu as that manually parses everything, and it loads the gamecard faster than the home menu
getting the list of title_ids is very fast (less than 1ms), however parsing the control info, such as title names
is very slow.
depending on how many games the user has, blocking until we read all control info can take several seconds...
we would only need to block if the user wants to sort by name.
normally, we lazy load the control data, so we don't suffer from slow load times at all.
i decided that its not worth slowing the whole system down just to give the option to sort by name.
some other changes:
- shorten the next page and prev page to just next/prev in themezer.
- remove misc shortcut name. the function itself still exists.
this crash was found by @WE1ZARD.
to trigger it, press L2 in the filebrowser whilst a hidden file exists and the hide hidden is enabled.
the was due to GetEntry(i) internally using m_entries_current, and the select all was using the index from m_entries_current.
this would result in an index that goes oob, and as its a write, it crashes.
