from 7615e559 - sorry

This reverts commit 8f11c925.

And while at it, make it easier to work with distribution names with
less if statements and more case matching.

Better diff with the --ignore-all-space option.

Removed one paragraph which was really confusing and, I guess, wrong:
IIUC, backend references _always_ need to be taken with VRT_Assign_Backend,
because the referencing vmod can make _no_ assumptions about the assigned
backend.

Also I hope to have clarified use of VRT_StaticDirector().

I found the previous wording confusing and hope this is considered an improvement.

We should make clear that lack of onerror="continue" implies abort only if
the feature flag is set.

Refs varnishcache/pkg-varnish-cache#154

Thank you do Dridi for his work on the release documentation.

I have gone through all commits and reused some wording of his
where appropriate.

As before, the changelog should roughly match the commit order (from new
to old).

It was increased in 635f9f20 to avoid
sanitizer suppressions to scroll out of screen the lines we expect to
find, but on some 32bit ARM system it makes the test even less stable.

Refs #3788

Working on the release notes I realized that there was nothing
actionable to prepare for deprecated aliases in future releases.

Restructured so that:

* 'Upgrading' is limited to work that has to be done to upgrade from
  a current deployment to the new version.

* 'Changes' is a comprehensive, user-level description of changes and
  new features.

Conflicts:
	doc/sphinx/whats-new/index.rst

This reverts commit 6225c0af.

To be added back after we branch off for the next release.

In a different context, I questioned my understanding of C object
lifetimes and learned that I had given the wrong explanation for why
TOSTRAND() was used wrongly in
0c96fc65:

TOSTRAND() creates a compound literal whose lifetime is the enclosing
block.

Thus, the error was unrelated to temporary lifetime, but rather that
the compound literal was assigned in the if-block, but used outside
it.

Ref: https://stackoverflow.com/questions/71225472/more-questions-on-the-c11-temporary-lifetime-rule-and-undefined-behaviour/71226049#71226049

The fix did work, but the explanation was wrong.

This reverts commit a6c13064.

This was supposed to happen before the first release including the VMOD
but it got OBE at the time.

It is now possible to rename a symbol from a VMOD without breaking
existing code until the VMOD author decides to remove the deprecated
alias.

The VCC file syntax is simply:

    $Alias <alias> <function>

    [Optional description]

    $Alias <.alias> <obj.method>

    [Optional description]

The alias can only apply to a $Function or $Method.

The generated RST looks like this for a method:

    ALIAS x<object>.<name>()
    ------------------------

    Deprecated alias for ``x<object>.<method>()``.

    [Optional description]

It looks the same for a function without the $Object prefix. Aliases
don't have a reference label for sphinx docs, the goal is not to make
them prominent, but rather to have the ability to move them to a manual
section for deprecated symbols at the end of the document.

The script expects aliases to be defined like this in the RST docs:

    my.alias ``VCL <= 4.1``

        Type: DEPRECATED

	Alias of: something.else

	A little description is still necessary.

There is no DEPRECATED type in addition to the SYM_ALIAS kind, this
is a purely cosmetic suggestion.

Deprecated aliases should probably have a high VCL version limit like
in the example above, generate.py doesn't enforce it so far.

Aliases are a new strictly internal kind of symbols with zero runtime
cost. They resolve implicitly to the destination symbol, even when they
are created.

This should facilitate renaming things in the future. One thing we
could do is to have libvcc warn about deprecated aliases (subject to
a new vcc_err_alias parameter too) to make this visible not just in
the documentation.

There was already a macro called EXPERIMENTAL, so it clashes with the
new set of EXPERIMENTAL* macros. Anything directly referring to the
parameter keeps the experimental prefix, otherwise, experiment.

Spotted by Flexelint.

Instead of manually setting the default bits, give the parameters
specs actual default values and centralize the special handling of
the "default" argument.