main is red: synced specs contradict their own datasheets (PoE-in declared as PSE, 10G in 1G cages) #953

Closed
opened 2026-11-08 06:54:38 +00:00 by jaylouisw · 0 comments
Owner

main is red, and has been since the sync commit 9e1bf8a brought the GitHub history in.
Two suites fail, for two unrelated reasons, both of them specs contradicting their own text.

✗ hardware-invariants   21 passed, 1 failed
    no port claims a line rate its cage cannot carry
    (ER-8-XG/sfp1 claims 10000 Mbps on a 1000 Mbps sfp cage, ...)
✗ poe-inference         1 failed
    off a switch, only ports that name themselves PoE source power
    (50 inferred PSE port(s) that are not named PoE on a non-switch)

1. Devices that accept power, declared as power sources. EP-R6, ER-10X and ACB-AC say
"PoE-in" in their own descriptions and declare every port a PSE. A sweep of the catalogue for the
same contradiction found three more that were fine - EP-R8 passes power through, EP-S16 has
4 PoE + 12 PoE+, N-SW has 3 PoE out - so this is not a category error, it is three bad files.

2. poe_in treated as a power source. isPoePse() returns true for any port whose name
contains "poe", so the inline splitters planet/POE-E201 and poe-texas/GAF-12V12W became PSEs on
the strength of their own input jack's name.

3. 10G ports in 1G cages. ER-8-XG (8 ports) and ENVR (2) declare connector sfp at 10G.
Ubiquiti documents both as SFP+.

The invariant itself is also part of the problem: it tests the port name, which is a proxy for
intent, and so fires on legitimate PoE outputs whose datasheet names them "PoE 1..16" while never
asking whether the engine invented the watts. That is the same class as #1069.

Worth noting for whoever owns the sync: this landed on main without a CI run, because
pull_request events are not creating tasks (#930). Two red suites sat there until someone ran the
suite by hand.

`main` is red, and has been since the sync commit `9e1bf8a` brought the GitHub history in. Two suites fail, for two unrelated reasons, both of them specs contradicting their own text. ``` ✗ hardware-invariants 21 passed, 1 failed no port claims a line rate its cage cannot carry (ER-8-XG/sfp1 claims 10000 Mbps on a 1000 Mbps sfp cage, ...) ✗ poe-inference 1 failed off a switch, only ports that name themselves PoE source power (50 inferred PSE port(s) that are not named PoE on a non-switch) ``` **1. Devices that accept power, declared as power sources.** `EP-R6`, `ER-10X` and `ACB-AC` say "PoE-in" in their own descriptions and declare every port a PSE. A sweep of the catalogue for the same contradiction found three more that were *fine* - `EP-R8` passes power through, `EP-S16` has 4 PoE + 12 PoE+, `N-SW` has 3 PoE out - so this is not a category error, it is three bad files. **2. `poe_in` treated as a power source.** `isPoePse()` returns true for any port whose name contains "poe", so the inline splitters `planet/POE-E201` and `poe-texas/GAF-12V12W` became PSEs on the strength of their own input jack's name. **3. 10G ports in 1G cages.** `ER-8-XG` (8 ports) and `ENVR` (2) declare connector `sfp` at 10G. Ubiquiti documents both as SFP+. The invariant itself is also part of the problem: it tests the port *name*, which is a proxy for intent, and so fires on legitimate PoE outputs whose datasheet names them "PoE 1..16" while never asking whether the engine invented the watts. That is the same class as #1069. Worth noting for whoever owns the sync: this landed on `main` without a CI run, because `pull_request` events are not creating tasks (#930). Two red suites sat there until someone ran the suite by hand.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jaylouisw/packetforge#953
No description provided.