Verify that s_resp_bodybytes is indeed incremented the correct number of bytes on partial content responses.

Pointed out by:	Dridi Boukelmoune

Fixes #1510

is a contention point and scales negatively with number of thread-pools.

Sum into a per-pool dstat structure instead, and have the first
idle thread summ that into the global counters without blocking the
pool mutex.

Fixes	#1495

(I hope)

also takes account of fetch jobs.

what the real solution is going to be.

When we in the future introduce something new and shiny here, don't
automatically start putting it into the tarball release.

This may seem a bit backward, the main idea is to avoid hardcoding
the section/subdirectory names yet another place.

These are just intermediate files written by Sphinx to
build the HTML documentation. We don't need to distribute
them.

With parallell builds, building this file has usually not happened yet
in the build directory.

Suggested by: Martin

We don't need to add the RST2ANY_FLAGS arguments twice.

Fixes	#1501

pick a backend.

Add a reference from the init function to any non-anonymous match_acl_*()
functions to ensure the compiler does not barf.

Fixes	#1504

This follows the format Apache logs in.

One of the examples was still referring to the old way of defining
directors, which I am sure is confusing for the new users this
document is meant for.

Noticed by: coredump on irc.

Used the occation to do some minor scrubbing of the text.

None of our automatic build environments have Python 3 currently.

Use the strict parsing mode (still infant) to be sure
that our own vmod.vcc files are valid.

This description isn't used anywhere, and it doesn't contain
much useful information.

This is one of those tiny jobs you start on (add --strict),
and then it just goes downhill from there.

Main changes:
* Use optparse to parse command line arguments.
* With --strict it aborts if a parse or format error is seen.
* A basic test case for the parser has been added.

No changes were made to the parser or output format.

Adding the test case meant reordering the execution flow.

Since Varnish code standard is using tabs, I've kept this to my
best ability. Pylint is not happy about this and give the code
a score of -4 out of 10 points. :-)

These hard-code references were forgotten when I
removed this file in ebbd8288.

The sphinx version on git.varnish-cache.org is v1.1.8 which
hasn't got -j.

In sphinx v1.2.2 (debian jessie) it is in place.

Guess we'll have to live with those extra 820 milliseconds
per build for now.

Warnings are now considered build errors.

We've been shipping documentation with sections missing
because we didn't take warnings seriously.

Bonus change: build in parallel, save 82 centiseconds per build!
(html target on my laptop, ymmv)

Squashing Sphinx warnings revealed that the vmod_std and
vmod_directors includes wasn't working.

Move the generated files a bit so Sphinx is happy to load them.

I'm not using include (where relative paths seem to work)
since that works differently in a TOC listing.

Sphinx complains that this file isn't referenced
anywhere, and a quick glance indicate that the contents
have been merged into other rst files.

We don't allow compiler warnings in the C code, so I don't see any
reason why we should allow warnings in the documentation.

Warnings in the documentation are now build errors.

Empty RST comments need an extra empty line afterwards, re this output
from rst2man:

    (WARNING/2) Explicit markup ends without a blank line; unexpected unindent

This made it easier to see where objects started/stopped, but
vmodtool.py don't believe in this comment format. Especially rst2man
wasn't impressed, so roll back for now.

Write a header in generated .rst files that it was generated from the
.vcc file.

Also softly sprinkled with some python formatting:
* use a python 2.5 context manager for the filepointer
* and make the output easier to read by avoiding concatination.

And avoid hardcoding it, in case we change it in the future
some time.