Repository navigation
Links without byte ranges, save a link to a PS5 folder (#872), finalizing count (#873), full review fixes - #875
Merged
Merged
Conversation
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
The limit was filled in from settings.json whenever the two differed, so a file written before the last change (a quit inside the mirror's debounce, a change made through the settings API) undid it at the next launch. It is now filled in only when this app has never stored one, and "no limit" is stored as 0 instead of removing the key, so clearing the limit is kept too. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Both started the console's whole queue, so rows the user had queued but not
started ("start when I add" off) began uploading. An upload row's Retry was
already limited to its row.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ck repair refuses a mounted device sceFsMountSaveData / sceFsUmountSaveData (save.mount, save.unmount, the orphan release) and sceFsMountGamePkg / sceFsUfsFsck / sceFsMountLwfs ran outside sony_api_lock, so they could overlap any other Sony call from another management thread. They now take it (after g_save_mtx; nothing takes the two the other way round). The lazy dlopen/dlsym of libSceFsInternalForVsh is under the same lock, so two first calls no longer race on its statics. fs.fsck with repair:true wrote to a device whatever was mounted from it. A repair is now refused (device_mounted) while the device or one of its partitions is a mount source; the read-only check is unchanged. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… manifest lines restore whole Two paths that flatten to one name (a/b_c and a_b/c, or two long paths cut to one tail) were copied over each other, and a restore wrote the second file's bytes to both paths. A name is now made unique inside the snapshot. A flattened name over 255 bytes could not be created, so a deep file was left out without a word: names are cut to their last 200 bytes. A path with a line break cannot be written to the manifest and is left out. Restore read manifest lines into 1100 bytes while a line can be 1536: the first piece (name + a path cut short) was restored to the wrong path. Lines are read whole (4096, heap) and one that does not fit is skipped whole. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…s turning others off collect_enabled_ranges failed when any enabled cheat of the title could not be resolved (bad offset, missing bytes, module not loaded) or one of its files did not load, and every turn-off for the title was then refused. Such a cheat was never written (apply resolves the same way before writing), so it has no ranges to keep; only one the watcher reports as applied to this process still refuses. A file that does not load is skipped as slots_load and cheats_toggle skip it, which also keeps the flat numbering the same in all three. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
When the extractor cannot write its status (the drive holding /data/ps5upload/extract is full), the helper sees nothing, gives up after 20 s as not started, and removes the destination folder while the extractor is still writing into it; nothing follows that run, Stop cannot reach it, and a second extraction can start beside it. The extractor now ends before writing anything when its first status cannot be written. For #869 (cause of the crash itself not proven). Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
fetch_piece slept through every Retry-After (up to 512 waits of 30 s a piece) with no look at the cancel, so a cancelled download kept its worker, its partial file and its name for as long as the host kept answering 429/503. RemoteSource now takes a stop check, asked before each request and during each wait; the link download passes its cancel flag and reports the stopped fetch as a cancel, not a failure. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
The console keeps one save open, and opening another closes the first. Share, use-shared, copy, restore and backup each check 'is a save open' and then open, fill and close over many requests, so two overlapping on one console passed the check together and closed each other's save half filled. They now hold the console (both consoles for a copy between two) for the whole operation; the second waits for the first. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
When the helper died during an extraction, the extractor (its own process) went on and could finish. The next sweep (at the next helper start, or at the next extraction, possibly days later) still removed the folder as half-made: a complete extraction deleted. The sweep now reads the extractor's last status and keeps the folder when that same extractor reported done. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…lost The progress write after starting a part or reading its status reset the stop flag, so only a Stop pressed during the one-second pause between reads worked. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
A failure that did not come from the console (source gone, engine unreachable) has no reason code, and its text was dropped: the task said only 'upload failed' and the Activity row had no reason at all. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
On Android Tauri's app_config_dir is the app's whole data folder, so the reset deleted the running WebView's profile with everything else and then failed with Permission denied on the folder itself, before the engine's reset was marked. Only settings.json and the bug-report folders go now. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Open folder (xdg-open), the keep-awake hold (systemd-inhibit sh -c) and the port query (sh -c ss) inherited the AppImage's LD_LIBRARY_PATH. Where /bin/sh is bash with a newer readline than the bundle's (Arch, #603) the shell died at once: nothing opened, the PC was not kept awake, and an engine left on the port was not found. They now drop those variables, as the update script already did. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ued again A Reinstall waiting behind an upload still reads finished, so Clear finished deleted its file and the queued install then failed. Clear all already had this guard. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… Files) Both showed 'installed' for a result the app could not confirm. They now show the installer's own note as a warning, like the other install paths. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… EiB free (#869) statfs counts the blocks a writer may take as a signed number, which goes below zero once root writes into the drive's reserve; statvfs hands it over unsigned. Every room check that multiplied it read a drive past full as about 16 EiB free: the extractor's per-file check, fs.freespace, the unpack, dump, pkg-stage and web upload checks. They now go through statvfs_avail_bytes, where zero or below is no room. The extractor also looked at the room only when a file began. A file with no size in its header (.gz, a stream), or the internal drive losing free space faster than the bytes written, got past that and, as root, wrote on through the reserve to the drive's last block. It now looks again every 4 MiB written and ends as fs_extract_no_space when nothing is left for an ordinary writer. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
lstat of a mount point answers with the mounted drive's device, so /mnt/usb0 -> /mnt/usb0/dir/x compared as same-device and reached rename(2) with the source's own directory on another filesystem. The source must now be on its parent's device too (fs.rename and the shell's mv share the guard). Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…es '..'
A relative mount_point ("data") shared the tracker key of an image mounted
as /mnt/ps5upload/data, skipped the image-unit check (no mount has that
name) and was handed to unmount(2), which resolves it against the working
directory: /data, then MNT_FORCE. The unpack sweep trusted a record's mount
point by prefix only, so <base>/../../ext1 was unmounted by force.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…stall by download first (#872) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…rd's list When the engine never became ready in 30 s (or start() found a dead child in the slot), start() killed and reaped it but its PID stayed registered. An exit clean-up that later overran would kill whatever process had that PID by then (with its tree on Windows). Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
The app stayed pointed at the removed address: it kept polling it, and the next launch added it back as a console (an empty roster is seeded from the stored address). Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
The failed upload paused the queue; the resume on reconnect finished that one upload and left the rest waiting for a click. The Resume button already un-paused the queue. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…e switch Server backup: picking another server was undone at once by the effect that shows the saved folder's server, so the folder could not be moved to another server. Save to USB: the backup failed with 'Host changed' when another console was selected while it ran, although it uploads to its own console's address. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Their 'console changed' effects run again whenever a hidden tree is shown, so a look at another console or screen cleared the shell session, the log, the search results, and cut a multi-package pick short. They now reset only on a real change of console, and a result that lands while hidden is kept. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…y_api_lock; fan settings saved through a temp; cheat engine switched on only after the checks - drive_sensors walked getmntinfo()'s shared array, which any other thread's call rewrites or frees: it now uses mntinfo_snapshot like every other reader. - focus.probe called sceSystemServiceGetAppIdOfBigApp/MiniApp (and resolved them) outside sony_api_lock. - fan_curve.json, fan_threshold.conf and fan_reapply.conf were truncated in place before being written, so a full drive left them empty; they go through atomic_file_write. - cheats_toggle switched the engine on (and saved it) before checking that the game runs and has a cheat file, so a refused request left it on. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
try_wait reaps an exited engine, after which Child::id() is None, so the stderr watcher and stop() could no longer unregister its PID. Both callers (the readiness loop and engine_diagnose) now unregister it at that moment. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
#872) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…g out 429/503 Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…never run as two moves, a second check keeps the download - Send to another console: Send stays off until the destination folder's listing is in, so what is already there is always named first. - Files paste, drag-move and bulk delete ask again whether a move is running after their awaits. - Check for updates with the update already downloaded keeps the download when the answer is the same version. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…wn again, one action per click - Screenshots and Video clips: the selection and the list are cleared only when the console really changes, not each time the tab is shown again. - Cheats: the newest repo search wins; one Download at a time; each switch in flight is tracked on its own; the game list is no longer rebuilt on every 5 s status poll. - Clean up: a Remove the console accepted is followed even when the first status read fails, and cannot be sent twice. - Make room: a run on another console no longer freezes this console's planner (only Do it waits, and says why), and does not lose its result. - Home: the Updated card dismissed on one console is gone on the others too. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…st captures, notices that fit a phone - A system with no WebCodecs whose stream begins while the picture is on no page: nothing throws on each frame, and the media player starts when the picture is shown (it stayed black for the session). - With the media player showing the stream, a screenshot comes from its video (it was a blank image saved as a success) and a clip of the empty canvas is not recorded. - An HDR presenter made while Stop lands is disposed; a pad or join link whose first socket cannot be made leaves no timers running. - Set up Play and the Game panel keep what they read when the screen is shown again. - A stream's error and its quality notice show only on that console's Play page. - A macro's key is caught on the window, so it works where a click does not focus a button. - The phone link panel takes its no-address line down once a later open works. - Centred notices over the picture may use the stage's full width, not half of it. - Join and pad pages regenerated. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…t a row is doing, no second send - The game page's installed list, its Saves and Cheats tabs, On your PS5 and the ShadowMount+ panel keep what they loaded (and what was open) when the screen is shown again. - A row's progress strip names Play, Register, Unregister, Heal and Edit files instead of calling them all a download. - The next-step Send button queues the install when the drives hold the game only as a package (it did nothing). - A failed hand-off to ShadowMount+ is recorded as failed, not done. - A play session is not collected while the game still runs after looking at another console. - A payload card's old error goes when it is tried again; a file check cannot put Send back during a send. - A long image name no longer pushes a drive row off a phone's edge. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…through a temp; waits off the wall clock - fs.list_volumes and hw.storage fell back to bfree when bavail was zero or below, so a drive written into its reserve showed the reserve as free space (and a reserve of 0) and the apps admitted uploads onto it. Zero now, as fs.freespace answers (statvfs_avail.h). - register.c copy_file_bytes truncated its target in place: registering a title again on a full drive emptied the param.json and icons of a registration that worked. It writes a sibling temp and renames it. - The power, notification and kernel-extras resolvers raised their "tried" flag before the pointers were stored; a second thread's first call in between answered "not supported". The flag is set last. - The tile watch only ended for a newer request, not for a sweep request: the sweep ran when the watch was over (up to ~2 minutes) with tile_busy true all that time. The tile waits and the vinput relay's tick were on wall-clock deadlines, which a clock set back (time sync, a wake) stretches by the jump; they nap in short steps and count on the monotonic clock. - A killed extractor's folder was removed after a fixed 500 ms; the helper now waits (up to 10 s) for the pid to be gone first. - A shell command that filled the 2047-byte buffer was cut and still run. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… path too long and cp onto itself are refused; cp -r reports a failed write - The protected-folder check skipped every argument starting with '-', which cat, xxd, stat, file, tail, sfoinfo, mv and chmod take as a path: with a link named "-L" into the store, 'cat -L/identity' printed the key. - A path too long for the policy to hold was answered "not protected". - More than 31 arguments were dropped without a word and the rest ran (mv f1 .. f31 dir moved into f31). Refused for commands that take paths. - 'cp a a' / 'cp a .' truncated the file and reported success. - cp -r went on reading after a failed write and lost the failure: a full drive left short files with exit 0. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ge is not held to the whole archive - The check runs before the destination is made, and statvfs of a folder that is not there fails: the room was "not known" and the check skipped, so on the console (which refuses an existing folder unless merging) an archive too large for the drive was always started (#869). The nearest existing folder above is asked. The test's fake free space no longer bypasses that lookup. - 7z/zip/tar: a merge was refused for the whole archive's size though it skips what is already there (RAR already kept the per-file check only). - RAR5: a checksum error in the last volume was reported as a missing next volume. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… finished .tar A folder download has no Content-Length, so when a read error on the drive ended the stream (c2c714d) the connection simply closed and the browser saved the truncated archive as complete. The archive is sent chunked to HTTP/1.1 clients, and the closing chunk only after all of it. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ast byte is timed (#873) A queue row reads "Finalizing on PS5" when the job's bytes_sent reaches its size. That count was the bytes the console had acknowledged plus what it held when the job first opened, and never below the sum of its Durable reports. Neither is what the console has: - a resume right after a Stop finds frames of the stopped run still arriving. The console writes them, the resumed run sends the same ranges again (its map did not list them), and the console reports both as durable. Counted twice, the job read 100% while tens of MiB were still on their way, and its result carried rechecked_bytes > 0 with the engine's "the console found N bytes that did not match" warning although nothing had failed a check; - bytes acknowledged and then lost (the helper stopped, the job was reopened from its journal) were counted again when sent again. The sender's space check runs at every open with what the console's map says it holds. The count now starts again from that figure each time (progress::Opened): the console's figure at the latest open plus what it has acknowledged since. It reaches the job's size when the console has acknowledged the last byte it needed. rechecked_bytes is counted the same way, so only bytes sent again after a failed check add to it. Diagnostics for the stretch after the last byte, which is where #873 sits ("stuck on finalizing for up to 20 minutes" after a resume; not reproduced, see the tests): - the snapshot of a running job carries all_sent_ms and console_quiet_ms: how long ago the console acknowledged every byte, and how long since it last confirmed anything new (a durable byte, a finished file, a file saved in place); - at 60 s of nothing confirmed the engine log says so once, with durable/total bytes, files done, what the console is doing (saving files in place, reading large files back, still writing), bytes acknowledged and sent twice, what it held before, attempt, lanes and whether the session is open; once more when it moves again; and a job that ended more than 60 s after its last byte logs the console's own end-of-job line (it counts the read-back); - the row shows "The PS5 has reported no progress for {time}." from 60 s on. Tests (ps5upload-ctest, C receiver, fd_budget 120, read-back on): - a_resume_right_after_a_stop_counts_no_byte_twice failed before ("the count read 314572800 of 268435456: 46137344 bytes were counted twice") and passes now; - a game interrupted by a Stop or a lost link, resumed, then two fresh ones: each ends soon after its last byte. These pass before and after: no wedge of later jobs was reproduced locally. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…start fails instead of hanging - A game handed to the Workshop is chosen or queued once, not again each time the screen is shown again. - Convert and replace: when the console does not answer as the swap starts, the run fails and says so (it stayed running for good, with no Cancel and the queue waiting). - The PS4 game and update pair, a copy to the PS5 for mounting, and the emulator tools' results survive the screen being shown again; a rebuild of the AMPR index is followed to its end. - A folder made from an archive shows its copy, check and unpack stages. - Delete the old dump / Keep it parked runs once per swap. - Play Together settings keep what is being typed; removing a backport library set says when it fails. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ut kept if it fails Also: build-list locks let go before leftovers are removed, exFAT reads over 4 GiB on 32-bit. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ify poll, a stopped link lets go of its server Also: the part file stays open across served chunks, a saved-server header read once less, delivered bytes never over the total. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…of-file and sign-in waits are bounded Also: FTP lines with no group column, a case-twin file is refused instead of replaced, NFS reads capped at what was asked. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…es to an ended session is told Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…ted folders seen on Windows Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
…hours, a release survives an unwritable cache Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
… Windows Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Links from servers without byte ranges and saving a link to a PS5 folder (#872), an early "Finalizing" count plus diagnostics for the long wait (#873), the extraction room check that did not run for a new folder (#869), and the fixes from a two-pass review of the app, the engine and the PS5 helper.
For #872 #873 #869.
probe_any/open_whole/WholeLinkSource(one connection, in order), download-first for packages from such servers; cancel under 429/503; one save operation per console; Windows rename retry for finished builds; FTP hangs and listing; panics on short Remote Play packets and non-ASCII content ids.🤖 Generated with Claude Code
https://claude.ai/code/session_01GtiJiJdHq1hekoZ6vizvVT