it more than once.

which means not in vcl_backend_error{}

Fixes: #1865

with their length.

Explains some of #2117 and friends

I'd really like to come back to https://lists.gt.net/varnish/dev/39240 somewhen.

We should not leave the workspace sizing calculations to the admin based on
incomplete information. The admin should only need to set the "space available
to VCL".

Varnish doesn't like empty hosts in backend definitions

It actually fails to build with recent enough GCC or Clang.

- Try to distinguish between DURATION, REAL and INT automatically.
- Explicitly mention which operands are possible when adding.

Related to #2101.

This reverts commit a4e07fc5.

It still fails on FreeBSD and SunOS.

If we receive a cancellation during pthread_cond_(timed)wait() we will
own the mutex on the way out. Avoid locking it twice, otherwise we
will deadlock.

(Ab)use the bound address used for the bad_backend macro to expose a
random port and avoid the dreaded collisions.

The port may not be hardcoded, it is still available for other processes
to steal during the race between `varnish v1 -stop` and the next `varnishd`
execution.

Fixes: #2200

Because it uses hard-coded ports, it often appears as failing in vtest
so now it should succeed if the address wasn't available.

Spotted by Martin, when a VMOD relying on varnish.m4 is built with
--prefix in the configure command-line, a pseudo hard-coded vmoddir
is assumed:

    $libdir/varnish/vmods

Instead, it now asks pkg-config for the vmoddir relative to the
libdir. It worked in all cases with upstream Varnish Cache, but
downstream modifications of the vmoddir would then be ignored.