When running really massive runs, "-j180 -n10000" kind of things,
the "rm -rf" of the tmpdir becomes the limiting factor.

The new -C option sends that int nice(1)'ed child process.

Two of them have no init step, and the third one is fail-safe.

And get a chance to catch vdp->fini() failures with the offending VDP
still present in the list of filters.

also when nothing has been received.

but also make it fail faster when appropriate

Apologies for the spam, I didn't think this would be a recurring
pattern.

Fixes: #2844

I caught one during a review and figured I might as well sweep the whole
tree, so this patch was created using Coccinelle and the following steps:

    $ cat >check_obj.cocci <<EOF
    @@
    expression obj, magic;
    @@

    if (obj != NULL) {
    - CHECK_OBJ_NOTNULL(obj, magic);
    + CHECK_OBJ(obj, magic);
    ...
    }
    EOF
    $ spatch --dir . --in-place --sp-file check_obj.cocci

This is my fourth semantic patch, I wouldn't mind checking them in to
have them handy in the repo to semi-automate code polish every now and
then. For example in a /tools/cocci directory. This way people could
study them or use them in their out-of-tree projects like VMODs or VUTs.

Fixes: #2297

Spotted by:	Nils

Tests #2887

Tests #2852

Sorry, not the big overhaul of u6 yet...

Neither the vmod_ default $Prefix not an author-supplied $Prefix has
got anything to do with the actual name of the vmod.

Allow VMOD writers to permit with NULL_OK flag.

Only call object destructor on initialized objects.

Fixes #2839

Yes, this is fugly, yet formatting of the backend.list output
is spread all over the place (including vmods via the vdi_list_f
callback), so until we decide it is worth spending time on separating
formatting from content, I see no other option than making the
field widths known to all vrt.

The route ahead will likely be to only generate json and make varnishadm
reponsible for formatting plain text output.

$ /tmp/bin/varnishadm backend.list -p
Backend name                   Admin   Probe      Last change
boot.a1                        probe   healthy    Tue, 05 Feb 2019 09:58:23 GMT

$ /tmp/bin/varnishadm backend.list -pj
[ 2, ["backend.list", "-pj"], 1549360935.584,
  {
    "boot.a1": {
      "type": "backend",
      "admin_health": "probe",
      "probe_message": "healthy",
      "probe_details": {},
      "last_change": 1549360703.515
    },

Fixes #2894

one <= 0 test is an oversight.

Use boilerplate for vmod_debug

We have code which relies on it being able to contain negative
values as error indications, but 'ssize_t' from unistd.h is a bridge
too far for vrt.h which is meant to be _incredibly_ portable to
avoid problems at compile time.

... to bring "more nuance" (quote phk) to the director health
state output, the list callback needs to be able to query
VRT_Health() of the director's backends, so, in turn, we also
need a CTX here.

Ref https://github.com/varnishcache/varnish-cache/pull/2888#issuecomment-460258147

Of the following out-of-tree vmods, none uses the .list callback,
so I expect minimal vmod author uprise:

	- saintmode (varnish-modules)
	- all_healthy
	- dynamic
	- cluster
	- weightadjust (someome(tm) needs to update to vdi_methods)