Root cause (post libp2p f54c715 update + rv/kad/service-disco changes):
- chat2disco (and waku factory for circuitRelayClient) do direct
`node.switch.services = @[Service(hp/autonat)]` (bypassing the
deprecated switch.add that called .setup).
- AutonatService (enableAddressMapper=true by default) + HP require
their .setup(switch) to populate .addressMapper (and handlers)
before start().
- In switch.start: services.start → autonat.start does
addressMappers.add(nil); await peerInfo.update() → for m in ...:
await m(...) → nil proc deref (Defect/SEGV in release).
- Surfaces at the `await node.switch.start()` (waku_node:589).
- Secondary: wakuKademlia.start() + waku mapper (capturing node,
returning announced which hp mutates) scheduled before switch.start
(which activates the mounted kad via ms + runs the hp mappers/updates).
Fix:
- After services= in apps/chat2disco/chat2disco.nim and in
waku/factory/waku.nim (both hp and bare autonat branches), explicitly
call the .setup(node.switch) (or hp.setup) and handle error.
- Move `if not node.wakuKademlia.isNil(): ...start()` to after
switch.start() + reconnectRelayPeers (correct ordering for mounted
protocol user loops).
- Harden waku addressMapper (nil/empty guard, return listenAddrs) and
set peerInfo.announcedAddrs (short-circuit) at the add site, in
updateAnnounced..., and in the onReservation callbacks (chat2disco +
factory) so expandAddrs prefers it.
- Minor: lookup/periodic guards in waku_kademlia; doc in
autonat_service.
Also nph reformats on touched files.
Reuses: the .setup methods, existing post-switch init patterns,
isNil guards, CatchableError handling, make chat2disco + nph.
Verified: make chat2disco (twice, pre/post nph) SuccessX; no SEGV in
multiple start-path runs (only expected thread EOF on pipe close);
diff only our 5 files.
Builds on 82d87cfa (libp2p update) without touching pins or vendored.
Caveat: clean `make update` still requires the mix temp patches in
pkgs2 (as documented in the update).
Fixes the reported chat2disco startup segfault.
