Hello,
The v5 I am about to send is not a re-send, and I would rather explain
the two things you both raised before it lands, because both are fair.
On the flooding: the v1..v4 postings on 2026-10-09 went out within ten
minutes with no version number in the subject. That was not four
revisions, it was one revision sent four times by a bug in my sending
script. It had no record of what it had already sent, so each invocation
sent the whole queue again. It now hashes each group and refuses to send
the same content twice. Sorry for the noise.
Rob, on the schema points, both are fixed in v5 and both were my
mistakes:
- the gpios property is back to "maxItems: 3" alone. I had added
"minItems: 3" and a required entry because sashiko flagged it, but
as you said, dt-schema makes the two bounds equal when only one is
given, so maxItems: 3 already means exactly three. Adding minItems
was wrong and I have reverted it.
- the two ports now reference "/schemas/graph.yaml#/properties/port"
with the unevaluatedProperties lines dropped.
Laurent, on the wider point, that one-at-a-time LLM conversions cost
more review time than they return: I agree, and this series is the last
one I send on its own. The remaining text-binding conversions I have
queued are all device tree, so I will send them as one group, per
subsystem as Conor suggested, so that the same class of mistake is
reviewed once instead of fifteen times.
If the view is that this conversion should not be carried at all -
Eduard Bostina posted the same conversion on 2026-07-30,
<[email protected]> - say so and I will drop
mine rather than add another one.
--
Chaosheng Qu
