However, you can figure out what pages on a large public site like Wikipedia people are reading over HTTPS, based on statistical traffic analysis, because you can see the size of the request, page, and each of the images. Combined with link following analysis, you can make a fairly accurate guess as to what people are reading.
"User X browses Wikipedia"
and
"User X browses Wikipedia articles about topics A, B, C"
where topics A, B, C could be anything user X doesn't want people recording them reading about: for instance, various political articles, articles about mental illness, articles about LGBT issues, etc. Fill in the blanks.
One interesting idea I heard from someone recently is tuning compression so that you target a small number of total page sizes -- rather than padding to expand pages to the same sizes, just don't compress them all quite as well as you could have.
Both GnuTLS[1] and Nginx[2] have length hiding implemented. But AFAIK OpenSSL doesn't have it yet, so most users are still left in the dark.
[1]http://www.manpagez.com/info/gnutls/gnutls-3.2.10/gnutls_180...
It might still be challenging to get large sides to adopt padding that will increase the amount of traffic they send (of course, the idea of reducing the efficiency of compression has the same net effect). But it's great to know that there's already a tool in place for traffic padding, at least in some implementations.
How does that work, surely the gov can still see people accessing the information by monitoring network traffic and the info itself is still public. HTTPS doesn't encrypt the actual request traffic does it, and in any case the gov would still see which server the traffic is going to unless you're using something like tor [and possibly still then].