ccan: Update vendored ccan to commit 7c38521d, which includes rustyrussell/ccan#128 and rustyrussell/ccan#130

[nGoline]

On macOS, failed connect() calls return POLLHUP|POLLERR or POLLOUT|POLLHUPfrom poll, and the socket error is set to ECONNREFUSED. Previously ccan's io_poll hardcoded EBADF in the POLLHUP case, causing misleading Bad file descriptor errors. The fix reads SO_ERROR from the socket to surface the real errno.

Also fixes update-ccan Makefile target:

• Use git rev-parse --short HEAD instead of git describe (ccan has no tags)
• Remove reference to ccodearchive.net following rustyrussell/ccan@efd4838

Closes #9236.

The test-level fix for test_networkevents (PR #9140) is still needed but will require a different approach now that the underlying error is correct.

Changelog-None

[ddustin]

Hm something weird happened here. This was changed with #9098 in commit 6664d34, but ccan wasn't updated.

Looks like it was a useful check but not critical.

@sangbida did this solve a particular issue you saw? If so let's get this fixed upstream in ccan so we stay in sync with it.

[sangbida]

I remember it being related to this previous commit here 400f97d. Rusty used json_out_addstrn instead of json_add_str_fmt(result, fieldname, "%.*s", (int)value_len, value);
for better efficiency but injson_out_addstrnthe escape path wasn't using the length that was passed in. The old path did json_add_str_fmt("%.*s", len, ...) + json_escape_len, so escaping respected the explicit length. json_out_addstrn's escape branch instead called json_escape(NULL, str), which uses strlen.

That broke getlog becuase log messages live in a ring buffer and are not NUL-terminated at loglen, so the json_add_stringn(info->response, "log", log, loglen) call in lightningd/log.c hands json_out_addstrn a non-terminated string. When a log line contains an escapable character, strlen reads past loglen into adjacent ring-buffer memory and emits invalid UTF-8.

[sangbida]

I've created a PR to add this into the ccan repo: rustyrussell/ccan#130

Thanks for catching that @ddustin @nGoline!

[nGoline]

Done! Updated ccan to the latest version containing your commit @sangbida.

Are we good to merge @ddustin?

[ddustin]

Looks good thanks for making the PR.

Would be good to get CI passing before merging in but this one should actually help CI flakes itself and seems low risk, so we can probably just merge it.

ACK fb93364

[sangbida]

Merging it because the only CI failure seems to be a timeout and unrelated to this PR.