Follows the similar commit made on ghc-head branch

Herbert Valerio Riedel authored 10 years ago and Mateusz Kowalczyk committed 10 years ago

This is needed to keep the compilation warning free (and thus pass GHC's
./validate) regardless of whether Word is re-exported from Prelude or not

See https://ghc.haskell.org/trac/ghc/ticket/9531 for more details

Currently we only use it only for ‘since’ annotations but with these
patches it should be fairly simple to add new attributes if we wish to.

Closes #26. It seems to work fine but due to 7.10 rush I don't have the
chance to do more exhaustive testing right now. The way the meta is
output (emphasis at the end of the whole comment) is fairly arbitrary
and subject to bikeshedding.

Note that this makes test for Bug310 fail due to interface version bump:
it can't find the docs for base with this interface version so it fails.
There is not much we can do to help this because it tests for ’built-in’
identifier, not something we can provide ourselves.

This is to disambiguate them from markdown links and will be require
with a future release.

(this slows down reloads on modifications)

I get rid of the Monoid instance because we weren't satisfying the laws.
Convenience of having <> didn't outweigh the shock-factor of having it
behave badly.

There seems to be memory and speed improvement.

The benefit of this is that the ‘top-level’ element of such lists is
properly wrapped in <p> tags so any CSS working with these will be
applied properly. It also just makes more sense.

Pointed out at jgm/pandoc#1346.

I can not remember why they were wrapped in paragraphs to begin with and
it seems unnecessary now that I test it. Closes #307.

(so that we can use haddock-library with doctest)

I think you either add src to hs-source-dirs or the library to
build-depends.  But doing both does not make sense (AFAICT).

Update repository urls, use subdir property for haddock-library and use
a separate versioning scheme for haddock-library in preparation for release.