OLED: IndexError when oled_network_interface="all" and interface count changes between updates

[jesus33lcc]

Versions

• pm_auto: 1.2.12
• pironman5: 1.2.22 (parent service)
• Raspberry Pi 5 (16 GB), Debian Trixie, Python 3.13

Symptom

With default oled_network_interface: "all", on a host where the set of
network interfaces changes frequently (e.g. Docker bridges coming up and
down when stacks are recreated), the OLED render loop raises
IndexError: list index out of range once per second:

File "/.../pm_auto/oled.py", line 172, in draw_oled
    ip = ips[self.ip_index]
         ~~~^^^^^^^^^^^^^^^
IndexError: list index out of range

The service stays alive (the @log_error decorator swallows the
exception), but the journal fills with tracebacks and the OLED frame
that renders the IP stops updating.

Root cause

In draw_oled, after rendering one frame, the index is bumped:

self.ip_index = (self.ip_index + 1) % len(ips)

…using len(ips) as it was during this call. On the next call,
get_data may return a shorter data['ips'] (an interface disappeared
between calls), so self.ip_index can be >= len(ips). The subsequent
ips[self.ip_index] then raises.

This is easy to trigger on a host with many short-lived bridges
(typical Docker setup with several compose stacks).

Reproduction

1. oled_network_interface: "all" (default).
2. On the host, repeatedly recreate Docker stacks so the bridge count
   fluctuates (e.g. docker compose down && up -d on a few stacks).
3. journalctl -u pironman5 -f shows the traceback every 1 s.

Suggested fix

Clamp self.ip_index against the current list size before indexing:

ips = data['ips']
ip = 'DISCONNECTED'

if len(ips) > 0:
    if self.ip_index >= len(ips):
        self.ip_index = 0
    ip = ips[self.ip_index]
    if time.time() - self.ip_show_next_timestamp > self.ip_show_next_interval:
        self.ip_show_next_timestamp = time.time()
        self.ip_index = (self.ip_index + 1) % len(ips)

Workaround

Setting oled_network_interface to a specific, stable interface
("eth0" in my case) sidesteps the issue.

Related

#2 (filtering out virtual interface prefixes) would reduce how often
the list size changes, but doesn't fully prevent the crash: any
transient interface that isn't in the prefix list (or any future
interface change) would still trigger the same IndexError. The two
fixes are complementary.