Deleting a source after a successful upload first renames it to
.<name>.mhu-delete-<uuid> and then unlinks it. On a slow disk the unlink
can take several seconds. With an empty extension filter the folder
watcher and the reconcile scan reported the staged file as a new file,
so the automation queued and uploaded it again until the unlink removed
it. The jobs for the vanished staged file were then restarted by every
automatic start and skipped with "Datei nicht gefunden".
- source-cleanup-policy: shared stagedSourceName/isStagedSourceFile.
- folder-monitor and file-discovery skip staged delete files.
- Automatic starts no longer restart queue jobs that point at a staged
delete file; manual starts are unchanged.
- An ENOENT while unlinking the staged file counts as deleted instead of
a failed unlink with a restore attempt. EBUSY/EPERM keep their retries
and stay visible as failures.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Electron 45 deprecates the synchronous safeStorage methods and Electron 46
removes them. Credential encryption now uses encryptStringAsync,
decryptStringAsync and isAsyncEncryptionAvailable.
- secret-store: async encryptField/decryptField/encryptCredentials/
decryptCredentials, decryptFieldWithStatus exposes shouldReEncrypt.
Unavailable or failing encryption still rejects with the existing
SECRET_STORE_* codes; legacy plaintext stays readable.
- config-store: the first load runs asynchronously through init();
writes encrypt before the atomic write. A synchronous load() of an
encrypted config before init() fails with SECRET_STORE_NOT_READY and
schedules a refresh. Values flagged with shouldReEncrypt are
re-encrypted once after loading.
- main: await configStore.init() after app ready, before the first load.
- MCP access store and online backup keyring await the async helpers;
MCP saves keep their call order.
Existing enc:v1 values written with encryptString stay readable: on
Windows both APIs use the same v10 format and Local State key, verified
in both directions with real Electron 41 and 45.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Logs were split into a new CloudDrop document every 700 KB, so a
session log of about 1 MB appeared as two files. CloudDrop accepts
text documents up to 5 MB, so the limit is now 4 MB and the response
limit of the log API leaves room for JSON-escaped paths.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With full access the installation grants every capability, destructive and
conditionally confirmed actions (queue retry/remove with source deletion,
history clear, app restart, ...) run without the local dialog, and the new
mhu_files_grant tool can add or widen directory grants. Narrowing or removing
grants and changing permissions stay local. Enabling full access in the
settings asks for local confirmation once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Byse rejects MKV files that carry an attachment (for example an embedded
.nfo text file) with "Not video file format". Verified with clips cut from
a rejected episode: the full clip is rejected, the same clip without the
attachment track is accepted.
For Byse uploads the Attachments element header is now rewritten into EBML
Void elements of the same length while streaming. File size, offsets and
all video, audio and subtitle tracks stay unchanged, the source file on disk
is not modified, and other hosters receive the original bytes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The total only counted the initial tasks, so batches that grew through
addJobs() logged results such as ok=200 fail=-197.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When the same file was registered for cleanup a second time (for example a
repeated upload to one hoster), both groups blocked each other through the
source lease and the file was never deleted, even after every hoster had
succeeded. A later group now takes over the lease once every job of the
earlier group has reached a terminal state; the earlier group is audited as
superseded. Groups with a still running job keep blocking as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Previously sources were only removed when the whole upload batch ended. In
automation mode a batch keeps growing for hours, so hundreds of finished
files stayed on disk. A file is now removed once every required hoster
succeeded, its links are flushed to the upload log, the automation ledger
is written and the renderer confirmed the queue save. Without that evidence
the file still waits for the batch end as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The shared upload limiter started with a full second of tokens and refilled
up to one second during short idle gaps, so new uploads could briefly send
almost twice the limit. In a parallel-upload test it released 3.87 MB in one
second at a 2 MB/s limit. Uploads now reserve send time on one shared
schedule with at most a 200 ms burst and exact waits.
Speeds and limits used 1024-based kB and MB while being shown as kB/s and
MB/s, so a limit of 100 meant 104.9 MB/s. Limits, measured speeds, ETA and
the speed graph now use 1 MB = 1,000,000 bytes. Saved limits are converted
once so the displayed value stays the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Setting UV_THREADPOOL_SIZE in main.js had no effect: Electron
initialises libuv before the main script runs, so the uploader kept the
default four threads for upload reads, logs and queue saves. Packaged
builds now relaunch once at startup with UV_THREADPOOL_SIZE=32 before
the single-instance lock. Portable and development runs are unchanged.
App info and the startup marker report the effective threadpool size.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>