Spotted by: Martin Boer

imposed by the kernel.

Spotted by:	ingvar (on PPC64)

From Martin Boer on #varnish

Spotted by Martin Boer on #varnish

The oc's busyobj isn't cleared in HSH_Fail (like it is in HSH_Complete).
This allows a thread to get a pointer to an already free'd busyobj.

Fixes: #1467

This file was named wrong, and the test does not make sense anymore, since the backend health information is in the fetch thread

This triggers 'shadows global' build errors.

This makes sure no extra 'transfer-encoding' headers should pop up
changing the byte counts.

This triggers 'shadows global' build errors.

This triggers 'shadows global' build errors.

These tags are now obsolete

Add per-backend VSC counters for traffic going in/out of this backend.

Add SLT_BereqAcct log records showing headers and bodybytes
received/transmitted when doing backend fetch operations.

Add VSC counters for pipe traffic. Traffic is split between header
bytes, and input and output piped bytes.

Log SLT_PipeAcct records showing per pipe session byte counts.

Charge ESI bodybytes to the parent request.

Log SLT_ESI_BodyBytes in each ESI-involved log transaction. This shows
how many bytes each request adds to the total byte count.

SLT_ReqAcct is not logged for ESI sub-requests.

Counters show bytes received/transmitted, split between header and
body bytes.

SLT_ReqAcct logs per request counter values.

We want to remove the req->acct_req dup counters that was feeding this
data to reduce the size of struct req. The statistics value of the
boolean flags was also small. So it's going away.

This makes it possible to anchor regex's at the end again.

Fix the meaning of '=' in logexpect to mean same value as that of the
last successfully matched log record.

the predictable panic if they are accessed from the backend side.

gcc 4.1.2 (and previous version I assume) doesn't support it.

The standard storage in our other Linux packages is 256 MB malloc.

Change the default on Redhat systems to match this.

Since we removed the bundled jemalloc, our packages have been
using system malloc on RHEL Linux.

As Ingar mentions in #1445, this is a workaround for the
dependency issue.

Originally attempt at a shorter fix in 6dc740f0 didn't work out.

find-provides did not pick up on libvarnishapi.so.1 for varnish-libs
automatically, and find-requires for the main package did. This meant
that package pair was uninstallable.

I don't quite know why this is happening, so I'm working around it
with a hardcoded Provides for now. This is on the critical path for
4.0 release.