This really is the wrong way to go, generated files should not live
in the source tree. But lightening the load of phk is more important.

to remove the dependency on graphviz when building from a tarball

As suggested by Przemek Sobstel on github.

Spotted by: hermunn

In most circumstances the counters vsm_overflow and vsm_overflowed
should be zero. This patch tells the user what he should probably do
when they are not.

(and test the commit email hook)

(But mostly for testing the commit email system.)

internal state.

Submitted by:	Pål Hermunn Johansen

Refs #1866

Based on the sergesyrota's work
(https://github.com/sergesyrota/Varnish-Cache/commit/d514686c643fac19da2aa9dfaa7690461510bffb)

Discussed with Federico.

Fixes: #1725

Early version of the patch OK'd by martin@.

Fixes #1855.

vbit_init now initializes a vbitmap from an arbitrary memory extent.
If more bits are used than the extent provided, auto-sizing falls
back to malloc, so vbit_destroy() should still be used.

adding a test program

Avoid some of the mess in src/ when building multiple vmods
in the same git tree.

When setting bit n (the n+1th bit) of an n-bit vbitmap, we'd miss to
expand the bitmap and thus overflow our buffer and overwrite the first
bit of the next byte in memory.

Without this we could start running ban_lurker_test_ban with an empty
list of bans to test, causing a lot of unnecessary ban and oh mutex
contention on every ban lurker invocation until the REQ ban triggering
it is gone.

Also don't call ban_lurker_test_ban when our destination ban is the
same as the one being tested. This would be wasted work.

Adjust the c00049.vtc test case for the bans now not being
unnecessarily tested.

Resetting the pointer would race against the ban lurker moving the OC
in the ban list and any object event listeners wanting to see the new
ban timestamp on OEV_BANCHG.

Fixes: #1863

Without this we could ban an object erranously if the ban lurker has
moved the OC in the ban list while we were evaluating.

Previously we would force the protocol and method fields to HTTP/1.1 and
GET, respectively, but hide this from VSL showing 2 different entries
for BereqProtocol and/or BereqMethod.

Discussed with arianna@.

Fixes #1662

This was forgotten in ecbdb32a and made the change moot.

Spotted by:	Martin

ban-list if the lurker washes this OC while we check it.

Mostly diagnosed by:	Martin

Fixes:	#1864