tests: Improve hitch_hosts(), add configure checks

[ghost]

This fixes hitch_hosts() on OpenBSD where fstat(1) or netstat(1) from
base are used in such cases.

Follow up to #209.

[dridi]

With this pull request, we now have 3 inconsistent uses of awk and sed (and I'm responsible for 2 of them). Since you already agreed to add the configure checks ensuring that we have at least one of each, I won't bother you with code style for which I'm guiltier.

The code looks good to me, in the mean time I'll try to test it on OpenBSD to get familiar-enough with fstat in case of future changes in this area.

Thanks!

[ghost]

I omitted unifying hunks in my diff since I couldn't test them on other platforms. Maybe I can do so later on and amend here.

[ghost]

Just added the configure bits and made the filters consistent across different tools.

Please run the test suite with this on your machines if you can, I will do so tomorrow on CentOS and OpenBSD.

[dridi]

Doesn't work for me, tests fail.

$ uname -a
OpenBSD localhost.localdomain 6.2 GENERIC.MP#134 amd64

OpenBSD doesn't seem to have pathchk so when I comment those out I'm down to two test failures:

• FAIL: tests/test03-multiple-listen (very much related to this patch)
• FAIL: tests/test14-ocsp-vfy (looks as if SSL_CERT_FILE is ignored)

I started from this pull request's branch, rebased against master, produced a clean make dist archive, and tried to build that archive on OpenBSD.

[ghost]

pathchk is another issue on my todo list, yes; it's not available on OpenBSD.

Well, what does test-suite.log say? You're most likely running into unrelated issues here.

My test results look good so far, nothing changes/breaks.

[dridi]

207a382 and c8cbd67 are the results of discussions on IRC instead of here.

[dridi]

Today I tried to support sockstat on Linux too and I'm offensively putting it in the "garbage until proven otherwise" category. It's available on Debian 9, and to put it simply doesn't work:

vagrant@stretch:~$ nc -p 8080 -l 127.0.0.1 &
[1] 11901
vagrant@stretch:~$ lsof -a -i TCP -P -p $(jobs -p)
COMMAND   PID    USER   FD   TYPE DEVICE SIZE/OFF NODE NAME
nc      11901 vagrant    3u  IPv4 217461      0t0  TCP *:8080 (LISTEN)
vagrant@stretch:~$ sockstat -P $(jobs -p)
USER     PROCESS    PID      PROTO  SOURCE ADDRESS  FOREIGN ADDRESS  STATE
vagrant  nc         11901    tcp4   *:7936          *:*              LISTEN

This off-by-144 is a mystery, so that's now settled: lsof if available, sockstat on FreeBSD and fstat on OpenBSD. We can revisit that if someone like @klenan knocks at our bugtracker for another OS.

The project looks dead by the way:

$ dpkg-query -s sockstat | grep Homepage
Homepage: http://nenolod.net/sockstat

Unless it's only me that can't reach http://nenolod.net/ at all.

Now the reason why this ticket is still open is because one test is failing on OpenBSD, the one test making use of the SSL_CERT_FILE environment variable. I'm clueless and currently suspect a difference in LibReSSL.

[dridi]

The SSL_CERT_* environment variables seem to be supported by openssl 1.1.0+ which would explain why OpenBSD doesn't support it currently. I will try again later with Debian 8/Ubuntu or CentOS 6/7 to confirm this but I find it surprising that we'd never seen it fail there before.