_Europe · 2026-09-24_
On a machine with the receipt dialog switched on, the dialog did not come up and nothing was printed, although the money was already on the card. The printer status was asked for from the UI thread, which the printer driver refuses on purpose, and the refusal reads like "printer not ready". The status is now taken from the reading the machine polls every five seconds, so the decision no longer depends on which thread asks. The same refusal also silently swallowed remotely triggered printouts.
Deleting expired log files walked the whole log tree on the main thread, once a minute — and on a machine with a TF card that tree is the card. PL0028 stood still for 11 minutes because of it, 0006 lost 68 minutes in a single day in pieces of one to nine minutes. The walk now runs on its own background thread, at most once an hour (plus once after a start), and it reports what it did: duration, space freed and the configured retention. It used to write nothing at all, so a walk that took minutes looked exactly like one that did nothing.
The pass that moves aged files to the card counted only successful moves against its limit of 500 files. On a full card nothing succeeds, so the limit had no effect: PL0028 ran one pass from 07:38 to 17:06 — 18 files moved, 85,538 attempts, each one a write attempt on a card with no room. A failed move now counts as well, and the limit is read per file instead of per folder, so a single folder cannot outrun it either. A full card is recognised as such (either from the error or from the free space) and ends the pass at the first file, with an hour before the next attempt.
Nothing ever deleted the AI test-case JSON: the seven-day retention only covers the recording path, which on a machine with a card is the card, so the copies in internal storage were kept forever and the ones on the card only until it filled up. PL0028 still held files from the 3rd of August on the 23rd of September. Both directories are now cleaned by age in both places, at most one week, adjustable downwards per machine. Every run reports what it did — deleted, split into internal and card, and how many files are still within the week — and a run that actually deleted something also appears in the backend's audit log.