gzip_disable doesn't disable gzip_vary

[Edho Arief]

2012/3/2 Maxim Dounin <mdounin at mdounin.ru>:
>> Shouldn't there be no change to header at all when client user agent
>> matches the gzip_disabled (ie. just like with gzip off)?
>
> http://tools.ietf.org/html/rfc2616#section-14.44
>
>   An HTTP/1.1 server SHOULD include a Vary header field with any
>   cacheable response that is subject to server-driven negotiation.
>
> For sure with gzip enabled the response "is subject to server-driven
> negotiation".
>

So, the "gzip enabled response" includes cases with non-gzip content as:
- no "Accept-Encoding: gzip,deflate"
- matches gzip_disabled
and therefore "Vary: Accept-Encoding" will always be added even though
the response isn't compressed?

Or in other words, does "gzip on; gzip_vary on" means "add 'Vary:
Accept-Encoding' header on any responses matching gzip_types
regardless whether the response actually compressed or not"?