os-s3-backup-greelan changelog

1.0_5 - 2026-10-03

Backups upload to a bucket with Object Lock.
Old backups are removed one by one only when the service has no bulk delete, and a failure says why.
A configuration saved in a language with a decimal comma is sent.
A user with read-only access cannot change the settings.
Unusual object names in the folder no longer break the listing or count as backups.
Names from a clock that ran ahead are removed first, and one save is stored once.
With 0 to keep, only the new backup is looked up, however large the folder.
The page shows settings as stored, an invalid setting is named, a bucket in another region names its endpoint, and a port above 65535 is refused.

The running configuration is what is sent, found by its contents rather than by the newest file name, and read only once a save in progress has finished.

The newest local backup is found by its time, so a backup with a shorter name no longer stops later ones being sent.

Old backups are removed in one request where the service supports it, and requests share one connection, so lowering the backup count a lot no longer risks a timeout.

A bucket listing that S3 cuts short without a continuation token is reported as an error instead of treated as complete.
Unexpected errors during Setup/Test S3 show on the page instead of breaking it.