The plain text output is a suitable input for tsort(1), initially
meant to find the correct discard order, I realized that it would
be more reliable to let vcl.discard figure that one out.

It wouldn't really work in my initial scenario:

    varnishadm vcl.deps |
    <filter some VCLs out> |
    tsort |
    xargs -n 1 varnishadm vcl.discard

The filtering part only worked for direct dependencies but we can
have two levels with return(vcl).

I'm keeping this command only for setup analysis, the output can
easily be turned into a graphviz dot file for example.

To put it simply, let's take a simple CLI script:

    vcl.discard vcl1
    vcl.discard vcl2
    [...]
    vcl.discard vclX

We can now achieve the same with a single command:

    vcl.discard vcl1 vcl2 ... vclX

But there is slighty more to it, because vcl.discard operates on both
VCLs and VCL labels, and dependencies can exist between both. So in
addition to operate on multiple VCLs it also does so in the correct
order.

This is going to be useful once vcl.discard can take multiple VCL names
at once.

The first operation consists in checking the ability to discard the
requested VCL, the second part effectively performs the VCL deletion.

closes #3379

avoiding local variables and VPI_BundleStrands()

Fixes #3392

I do not understand the global -esym does not work, but the warning is
definitely bogus, the (unisgned)-1 argument is intentional and a
documented feature.

we see this issue in two places (for libvcc and varnishd), hope this
will silence both

it does not register use via tbl/sess_attr.h

leaving these in the "to be reviewed" section because I am not entirely
sure.

I think these cases are:

- ses_set_attr / beresp_filter_fixed: flexelint bug? Looks like it does not
  register use via macros

- ObjGetU32/Lck_DestroyClass API

- HTTP_IterHdrPack / SYMTAB_NOERR cases we do not want to make static
  by design

Solve comparison between unsigned and -1 by moving a cast to where we need it.

the existing code already implied that a VCL_INT be large enough to hold
uid_t / gid_t.

Ref 31164908

Fixes #3243

we do want to use FLAGX | FLAGY even if zero for clarity

error:

Info 835: A zero has been given as right argument to operator '|'

Another case of struct instance names which we want to have, even though
we do not currently use them.

Yes, we do not actually use the mode name, but we still want to have it
for debugging and clarity.

Info 754: local struct member 'vsc_sf_mode::name' (line 57, file
    ../../lib/libvarnishapi/vsc.c) not referenced

Fixes #3243

We want to avoid unsigned integers for reference counting since they
would hide underflows.

Seeing a test failure on vtest I think there might be a race here that I
hope to make more explicit next time it occurs.

A nice and tidy single token to integrate with the parameters table.

Refs #3250

This reverts commit 5367be3f.

polish differently and more sensibly

constify and avoid (char *) = strchr((const char *), x)

error:

vtc_varnish.c  559  Error 158: Assignment to variable 'la' (line 548)
increases capability

as copies of the respective client side flags

When a parameter set is associated at the director level, it should be
in effect with resolve=NOW as with resolve=LAZY.

This adds VTCP_Assert() on the result of read and write calls that deals
with TCP sockets.

When used to check the result of read() and write() calls, it is useful
that a positive return value is accepted in VTCP_Check().

Now that VTCP_Assert() accepts EAGAIN as a legal errno value for read()
errors, uncomment this check.

Since the input value is sometimes the result of a read()/write() call,
avoid truncating the ssize_t value on calling it.

When a socket timeout is set on a socket and the timeout expires, read()
and write() calls on that socket will return (-1) with errno set to
EAGAIN/EWOULDBLOCK.