Hi Nick, Jona,
First, the split thread is my fault. My last mail went out as a new message instead of a reply, so that's why there are two. Sorry about that.
On history: there's no lessons-learned reason UDS was left out. In my fork it worked fine, it just never went upstream because the broadcast frames covered what we needed. Nothing broke.
On when to poll: in my fork the poll only ran while charging (gated on 0x594) and with ignition on. Off and awake-only didn't poll at all. For SoH I agree with Jona. Read it once near the end of a charge session, when the pack has settled. It barely moves, so more often doesn't buy you anything. I'd also add an xvg command to read it on demand. Then nobody has to wait for the next charge to get a value. Cell voltages could use the same schedule.
One thing worth knowing: metrics are persistent, so if you only set SoH on a real read (no default in the constructor, which dexterbg will also ask for) it just holds the last value between reads. That works well for something this slow.
On FCAN: on my car 0x191 is FCAN only. I found it zero times in about 2.5M KCAN frames over 40 captures, so if you want it you do need FCAN. Unfiltered FCAN is heavy, though. In one short unfiltered capture I saw around 1700 frames/s, the poller queue overflowing and the logger dropping frames. It didn't crash, but I wouldn't run that every day. I used the mcp2515 hardware acceptance filter to cut it down to the IDs we used, which is the other way to keep under Jona's 75% CPU target. That PR (#1404) is closed now since #1524 removed FCAN, but the filter code is in tag archive/pr-1404 on my fork if you want to look.
On the BAP channels you're pulling out of the JXE: check https://github.com/thomasakarlsen/e-golf-comfort-can before you test them. It's the best KCAN/BAP reference I've found for the e-Golf and was my main source for the clima work, so some of what you're finding may already be confirmed there. If you find channels it doesn't have, I'd be interested.
Andreas
On 2026-10-26T08:24:16+01:00, Andreas Jansson wrote:
Hi Nick,
I had UDS polling running on my own e-Golf (2020) in my fork for a couple of months, so some of the groundwork is done. It never went upstream, mostly because the broadcast frames covered what we needed at the time.
What I found, all on my car:
- BMS is reachable at 0x7E5 / 0x7ED on can1 (diag CAN, from the J533 harness, so all three buses need to be wired)
- DID 0x1E3B = pack voltage, u16 / 4
- DID 0x1E3D = pack current, (u16 - 2044) / 4, negate for discharge-positive
- That's the e-Up scaling, and it holds on the e-Golf. Against the FCAN broadcast (0x191) it agrees within ~0.5 A typically, 2 A worst, voltage exact. Small sample though, and only -80..+4 A, so not tested at high DC charge current.
- DID 0x74CB SoH reads about 10x the e-Up scale here. raw / 500 gave me 86.35%, which looks plausible but I never checked it against VCDS/ODIS. Treat it as a guess.
- CAC: I couldn't work out the scaling, the e-Up formula gives nonsense (430 Ah).
Two practical things:
- The e-Golf entry in main/Kconfig doesn't have "depends on OVMS_COMP_POLLER" like the e-Up does. You'll want that before using the poller.
- Since FCAN was dropped (#1524), current master doesn't set v.b.current / v.b.power during charging at all, so pack current over UDS while charging would fill a real gap, not just add cell detail.
On Jona's alarm point: I only polled while charging and with ignition on, never with the car off or just awake. So I can't confirm or rule it out. I'd gate the poll on those states anyway.
Code is in my fork, tag archive/vwegolf-20260815 on github.com/slothish/ovms3-VWeGolf (commits 32e29846a and 2c99b2724). It's on an old dev line so don't cherry-pick it, but the DID handling and the native tests (test_uds_poll.cpp) should be easy to lift.
Andreas