[hugojosealmeida]

Summary

When -i@host resolves to several addresses in one hostent, enter_nwad() created one nwad entry per address but stored the same arg allocation (and the same proto pointer) in all of them. lsof_destroy() frees np->arg and np->proto for every entry, so multi-address host names caused a double free. On macOS that aborted lsof before its stdio buffers were flushed, which is why LTsock reported “no client socket by host and port” in CI.

Each nwad now owns independent copies of its argument and protocol strings, and the caller-owned protocol string is released once after all address entries have been created.

Changes

• src/arg.c: moved argument allocation into the per-address nwad construction, copy proto per entry, free partial allocations on failure, and free the caller’s n.proto once on the success path.
• tests/case-23-multi-address-hostname.bash: regression test that runs lsof -i@<local hostname> when the host name resolves to multiple addresses and fails if lsof is terminated by a signal.
• Makefile.am: registered the new dialect-neutral test with the autotools test suite.

Verification

• make -j4 -> success
• bash ./check.bash darwin -> 15/15 test cases passed, 0 failed, 0 skipped
• bash tests/case-23-multi-address-hostname.bash against the pre-fix binary -> fails with status 133 (SIGTRAP); passes against the fixed binary
• bash -n tests/case-23-multi-address-hostname.bash -> syntax check passed
• git diff --check -> no whitespace errors

Risks / Follow-ups

The new test skips when python3 is unavailable or when the machine host name has fewer than two addresses in one family. This host and the macOS CI runner both resolve the host name to multiple addresses, so the regression path is exercised there.

---

This PR is AI Assisted.
Closes #370

[hugojosealmeida]

@jiegec Thanks. There is still a CI failure for freeBSD. see #373