Let the functions returning counters take an optional fantom pointer,
that can subsequently be used with VSM_StillValid to check for
changes.

Improve the VSC api documentation.

Add missing VSM_StillValid funciton to the .map file.

This allows Dispatch to run completely in SHM with no copying for most
cases, also when transactions come in multiple batches.

Introduce a VSLC_ptr type to represent a record pointer.

Move vtree.h from varnishtop to include directory so it's available for anyone.

Simple proof-of-concept single regex against any log line query
implementation.

Remove automagic reopening logic from vsl

Make vsl use the new VSL_head structure

Remove most of the old functionality

into the log.

Show UNSEEN in bottom bar when not hiding unseened counters

Add a VSC_Mgt() function to get direct access to the management counters

Add VSC_MgtValid(), VSC_MainValid(), VSC_IterValid() functions that
call VSM_StillValid on the fantom used to produce them.

Use the iter fantom when making the pointer lists, so VSM_IterValid()
returns the right result.

Clear the thread specific log pointer on log close

Reenable building of varnishtest

VSC type defined in vsc_types.h

Rearrange the VSC types according to relevance

Add global variables for the type name strings.

Compile in static VSC type descriptions in libvarnishapi

Don't do automagic reopening in VSC

Rename vf variables to fantom

Keep a list of fantoms for the points, and point the points to
these. This gets rid of the fantom struct on each point, and allows
for StillValid checking on groups at a time instead for each point.

Rename point's class to type to be consistent with VSM terminology

More class->type renames

Drop the type and ident pointers from VSC_point, and use the fantom's
pointers instead.

Reenable building of varnishstat

Don't assert on VSM file closed in VSC_Iter. It's safe to proceed
anyways and will produce a zero result in that case. This makes the
function consistent with how the other functions behave.

Rename VSM_FOREACH_SAFE as it isn't really safe in the vqueue sense,
and removing chunks from the VSM from the client library isn't
applicable anyway.

Make the VSM chunk marker sizes defines.

Check for shm open in VSM__itern

Revisit logic of VSM__itern: Make sure that the VSM_fantom is either
changed to the next, or not changed at all.

Add comments.

Keep copies of class, type and ident in VSM_fantom, and use these when
revalidating fantoms.

Do a proper VSM_Open in varnishstat

Add a VSM_ResetError function

Differentiate between missing HEAD marker and zero alloc sequence. The
latter is an indication Varnish isn't running, the former a corrupt
file.

Fix newline on diag messages for consistency.

Fix return value check inversion bug on stat call in VSM_Abandoned()

Don't assert on VSM_Open on an already open VSM. Report success instead.

segment.

The API will use this counter to check for abandonment.

Make use of the SHM age attribute in VSM_Abandoned.

This gives a less expensive way (without doing stat syscall) to check
for abandonment. If there's been more than 2 seconds between calls to
VSM_Abandoned and the age attribute hasn't changed in between, the
stat on the file will be performed to see if the SHM file has changed.

VSM_Abandoned will be the standard way to check for SHM changes in API
utilities, and when it returns true, the application is supposed to
call VSM_Close and then VSM_Open to reopen the SHM file.

Update Makefile.am in various places for build

Format: <type> <parent-vxid>

Add SLT_End records at the end of all vxid transactions.

Add SLT_Link records whenever one initiates a new child transation.
Format: <type-of-child> <child-vxid>

This is used by VSL api to form structured log output.

process uptime counter.

Basically the backend can drop the reason (OK) when responding to
a probe without Varnish declaring the backend sick.
Reported by astrosti on https://www.varnish-cache.org/forum/topic/961

first one created.  I would have skipped #0 if it was me...

bereq.* rather than the req.*, since that's what the backend saw.

only pass req to FetchHdr if we intend to send the body over.

Change args to http_FilterReq() to lower level.

After the purge, you will end up in vcl_purge{} from where it will
(at some future date) be possible to order a fresh fetch of the
object.

Doing purge from vcl_recv{} solves half of the object existence
problems in vcl_lookup{}.

	return (error(status_code [ , response_string ]));