Attempt to recover from "loose object" error in git status and git fetch commands. It is no longer necessary to run git fsck on every refresh attempt, only run it after a recovery attempt.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
When no callback is set for stderr, the outputfrom stderr is piped to stdout. This makes setting the callback redundant.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
USER CHANGE: The deprecated "client_repo" and "client_path" options have been removed in this commit. Use [update_manger client client_name] to configure clients.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
This should allow git's internal timeout mechanisms to trigger at the expense of potentially delaying response time if a fetch or pull hangs.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
Always run git fsck when initializing a repo. If a loose object error is detected when running git pull or git fetch, attempt to remove loose objects and retry.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
The Klipper repo always initializes last. Send a notification at this time, so if any clients are connected they can receive the full update status after init.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
During normal updates dependencies will always been updated when necessary. It is possible to force a dependency updating using the new "/machine/update/recover" endpoint.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
Rather than attempt to automate recovery, provide an endpoint for clients
to initiate. Clients can choose between a "hard" recovery, which removes
a corrupt repo and restores from a backup, or a "soft" recover which attempts
a git clean followed by a git reset.
Backups are now only taken when a valid repo is detected.
Signed-off-by: Eric Callahan <arksine.code@gmail.com>
