Some tests were moved under parseString as they weren't about paragraph
level markup.

Conflicts:
	src/Haddock/Parser.hs
	test/Haddock/ParserSpec.hs

The nesting rules are similar to Markdown's with the exception that we
can not simply indent the first line of a hard wrapped indented
paragraph and have it treated as if it was fully indented. The reason is
differences in markup as some of our constructs care about whitespace
while others just swallow everything up so it's just a lot easier to not
bother with it rather than making arbitrary rules.

Note that we now drop trailing for string entities inside of lists. They
weren't needed and it makes the output look uniform whether we use a
single or double newline between list elements.

Conflicts:
	src/Haddock/Parser.hs
	test/Haddock/ParserSpec.hs

There were some breaking changes over the last few patches so we bump
the interface version. This causes a big problem with testing:

1. To generate cross package docs, we first need to generate docs for
the package used.
2. To generate package docs with new interface version, we need to use
Haddock which has the version bumped.
3. To get Haddock with the version bump, we first need to test cross
package docs
4. GOTO 1

So the problem is the chicken and the egg problem. It seems that the
only solution would be to generate some interface files on the fly but
it is non-trivial.

To run this test, you'll have to:

* build Haddock without the test (make sure everything else passes)
* rebuild the packages used in the test with your shiny new binary
  making sure they are visible to Haddock
* remove the ‘_hidden’ suffix and re-run the tests

Note: because the packages currently used for this test are those
provided by GHC, it's probably non-trivial to just re-build them.
Preferably something less tedious to rebuild should be used and
something that is not subject to change.

Any extensions that are not enabled by a used language (Haskell2010
&c) will be shown. Furthermore, any implicitly enabled are also going
to be shown. While we could eliminate this either by using the GHC API
or a dirty hack, I opted not to: if a user doesn't want the implied
flags to show, they are recommended to use enable extensions more
carefully or individually. Perhaps this will encourage users to not
enable the most powerful flags needlessly. Enabled with show-extensions.

Conflicts:
	src/Haddock/InterfaceFile.hs

LaTeX will treat the h3-h6 headings the same as we'd have to hack the
style file heavily otherwise and it would make the headings tiny
anyway.

Hoogle upstream said they will put in the functionality on their end.

Conflicts:
	src/Haddock/Interface/Rename.hs
	src/Haddock/Types.hs
	test/Haddock/ParserSpec.hs

Conflicts:
	src/Haddock/Backends/Hoogle.hs
	src/Haddock/Interface/Rename.hs
	src/Haddock/Parser.hs

Also simplify specs and parsers while we're at it. Some parsers were
made more generic.

This commit is a part of GHC pre-merge squash, email
fuuzetsu@fuuzetsu.co.uk if you need the full commit history.

We also check that we have parsed everything with endOfInput.

We remove the HTML test as it is no longer necessary. We cover the
test case in spec tests and other HTML tests but keeping this around
fails: this is because the new parser has different semantics there.
In fact, I suspect the original behaviour was a bug that wasn't
caught/fixed but simply included as-is during the testing.

Instead, discard any static flags before parsing the command line using
GHC's DynFlags parser.

See http://ghc.haskell.org/trac/ghc/ticket/8276



Based off a patch from Simon Hengel.

Signed-off-by: Austin Seipp <austin@well-typed.com>

(so that themes are found during development)

Mateusz Kowalczyk authored 11 years ago and Simon Hengel committed 11 years ago

We inject -dynamic-too into flags before we run all our actions in the
GHC monad.

As `Traversable` needs at least one of `traverse` or `sequenceA` to be
overridden.

Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>

quchen authored 11 years ago and thoughtpolice committed 11 years ago

Signed-off-by: Austin Seipp <aseipp@pobox.com>

This is motivated by the fact that Haddock tests are not ran by the
GHC's ‘validate’ script so we're pretty liberal on dependencies in that
area. Full Hspec gives us some nice features such as Quickcheck integration.

I do the same thing that the XHTML backend does: give these no special
treatment and just act as if they are regular functions.

Mateusz Kowalczyk authored 11 years ago and Simon Hengel committed 11 years ago

This adds tests for all elements we can create during regular
parsing. This also adds tests for text with unicode in it.

Mathieu Boespflug authored 11 years ago and waern committed 11 years ago

This information is as relevant in the documentation as it is in the
source files themselves.

Signed-off-by: David Waern <david.waern@gmail.com>