…1.16

This is particularly important when TH is being used, as we need the
dynamic library built before the other ways try to do TH.

Additionally, change the search location for the package environment file to the
project directory instead of the sandbox directory.

Fixes haskell#1070.

This replaces the GHC/Hugs manual references in the Haddock comments
for the `Language.Haskell.Extension.Extension` enum constructors, as
in

```haskell
  -- | [GHC § 7.6.3.4] Allow overlapping class instances,
  -- provided there is a unique most specific instance for each use.
  OverlappingInstances
```

which urls pointing directly to the section containing further
documentation (assuming the urls are more immune to section
reorderings), e.g.:

```haskell
  -- | Allow overlapping class instances, provided there is a unique
  -- most specific instance for each use.
  --
  -- * <http://www.haskell.org/ghc/docs/latest/html/users_guide/type-class-extensions.html#instance-overlap>
     OverlappingInstances
```

Replace fragile GHC manual section references by urls

Fix for autoconf misbehaviour caused by cabal

Specifically,

 - `CApiFFI`
 - `DataKinds`
 - `DefaultSignatures`
 - `DeriveGeneric`
 - `InstanceSigs`
 - `InterruptibleFFI`
 - `LambdaCase`
 - `MonadComprehensions`
 - `MultiWayIf`
 - `ParallelArrays`
 - `PolyKinds`
 - `TraditionalRecordSyntax`
 - `Unsafe`

This addresses haskell#1050

Add a couple of new language extension flags known by GHC 7.[46]

See the discussion in haskell#1076.

See the discussion in haskell#1076.

This reverts commit df6fe40.

It broke the build on Windows, e.g.:

utils\hsc2hs\dist\build\autogen\Paths_hsc2hs.hs:45:34:
    Not in scope: `fromIntegral'

Partial Fix for haskell#543

Preprocessed files would be written to dist/build instead of
dist/build/my-test-suite/my-test-suite-tmp, causing them not to be
found during compilation.

Fixes haskell#1087.

This reverts commit 98e37ca.

This reverts commit 49de83a.

Fixes haskell#1081.

The main motivation for this change is to make it more convenient to use
GHCi.

…ype and TestType

…t 3 years instead of reimplementing it all over the place

…h qualified

…ncorrectly

…es. Instead, die immediately in Main.hs

See the discussion in haskell#1088.

This feature was disabled because it doesn't work with Cabal-1.16, but the
cabal-install in master has a '>= 1.17.0 & < 1.18' dependency.

This commit breaks the path module on Windows.

This reverts commit df6fe40.

Fixes haskell#1102

Conflicts:
	Cabal/Cabal.cabal

Conflicts:
	cabal-install/cabal-install.cabal

Currently only the haddock command has a flag to enable it

Control.Concurrent.QSem is deprecated and will be removed in GHC 7.8. The
documentation recommends using the SafeSemaphore package instead. Since this
package is not in the platform, we add the relevant module to
Distribution.Compat. Tested by running `cabal install -j2 ghc-core` on a fresh
~/.cabal.

SafeSemaphore uses the same 3-clause BSD license as Cabal.

It did reduce duplication, but made the code harder to read.

…ning sdist. Previously autogen files were only filtered from other-modules.

Instead of always using './.cabal-sandbox', load the sandbox path from
'cabal.sandbox.config'.

This is in place of the deprecated implementation from
Control.Concurrent. New implementation due to Simon Marlow. Note that
this has only been tested against stm-2.4 despite the permissive version
bounds given in the cabal file and bootstrap script.

This reverts commit ea1f338.

Conflicts:
	cabal-install/Distribution/Client/JobControl.hs

Code by Simon Marlow.

…ar solver.

The question is whether or not cabal-install will tell you why it tries
a particular goal (i.e., because it's the dependency of something else).

This regression was introduced between 0.14.0 and 1.16.0 of
cabal-install in order to fix a space leak. Restoring the information
actually costs some space again (although not as much as was required
back in 0.14.0).  I think the extra info is worth it though.

I think the extra parentheses in the type signatures help to
understand the intended use of functions.

Just a refactoring.

This way, the user won't be required to set things like 'split-objs'
individually for each sandbox.

This is a simplified version of Simon Hengel's patch for the 'base' library [1]
(without unsetEnv and the unix dependency). Tested on Windows and Linux.

[1] http://thread.gmane.org/gmane.comp.lang.haskell.libraries/18001

Fixes haskell#1120.

Fixes haskell#1007.

This reverts commit af86c76.

Fix a typo in Cabal/README

Fixes haskell#1113.

Plus add functions for finding all the deps of one or more
components, in dependency order. This will be useful for
building individual targets, e.g. cabal build foo.

This will be used for the new "cabal build [target]" feature and
similarly for "cabal repl [target]". That is we can actually specify
what we want to build, like a proper build system.

Targets can be components, modules or files. We go to some effort to
allow targets to be specified in a short convenient form. We allow
case insensitive matching. We only complain when the given form is
ambiguous (and note that we don't let the case insensitive matching
add artificial ambiguity). If it is ambiguous we report the
non-ambigious qualified forms of the targets it could be, e.g.:

  Ambiguous build target 'darcs'. It could be:
  exe:darcs (component)
  lib:darcs (component)

We note the kind of target here (component, module, file) since it's
possible to have ambiguities between them, like a component called
'Foo' and a module 'Foo' if the user entered 'foo'.

Needed to support the "cabal build [target]" feature.
Also move an existing util function so we can reuse it.

Allows users to specify what targets to build. It builds the given
target and the dependent components.

Currently only component targets are fully supported. Currently using
an individual module or file target will give a warning and it will
build the whole component the module or file belongs to.

GHC on ARM doesn't have a threaded RTS, which makes it impossible to build
cabal-install without editing the source by hand (see haskell#1129). This patch makes
cabal-install compile out of the box on ARM.

Do not link with threaded RTS on ARM.

Show '--config-file' in 'cabal help' output.

We must not pass '--disable-benchmarks' to setup scripts built with Cabal <
1.14; the same is also true for '--disable-tests' and Cabal < 1.10.

Fixes haskell#932 and haskell#1004.

Without this patch:

    $ cabal install -w /path/to/ghc-7.2.1/bin/ghc syb
    [...]
    unrecognized option `--disable-benchmarks'
    Failed to install syb-0.3.7

With this patch:

    $ cabal install -w /path/to/ghc-7.2.1/bin/ghc syb
    [...]
    Registering syb-0.3.7...
    Installed syb-0.3.7

Was accidentally using the targets rather than the dependencies
and the targets.

…oo/cabal

We set both globalInstallDirs and userInstallDirs to the same value, but using
'--global' makes cabal-install want to do unnecessary things like invoking
itself with 'sudo'.

…ixes haskell#1052)

And move compiling of C files after the compile of Haskell modules,
just like we do for libraries. Reversing the order is important because
C files can depend on headers generated for Haskell FFI exports.

This is based on a patch by Edward Z. Yang <ezyang@mit.edu>

The name 'InstallContext' will be used for common data shared between
lower-level install functions.

Splits 'D.C.Install.install' into three parts:

    * makeInstallContext - load common data
    * makeInstallPlan    - produce the install plan
    * processInstallPlan - actually perform the installations

This allows to manipulate the install plan produced with 'makeInstallPlan'
before performing the installations with 'processInstallPlan'. The high-level
'install' action is still present; most clients should use it instead.

Implemented by creating an install plan for ["add-source-dep-1", ...,
"add-source-dep-N", "."], pruning "." from this plan and then doing all
remaining installs in the plan before building the current package. This way,
all reverse dependencies of add-source packages needed to install the current
package are also reinstalled.

This allows specifying the main-is field as a C file.

This is closely based on patches by Edward Z. Yang, who in turn credits
and earlier set of patches by Irene Knapp. The slight difference in
this version of the patch is that it is adjusted to work with the new
approach where we have separate hs compile; c compile; and link phases.

For compatability, we need to make sure people using C files for the
main-is field, also have to specify the right cabal version.

This lets us use it for the package check in configure.

cabal init: parse module names with simpleParse instead of fromString

This will allow it to be set in config files too.

Sync bootstrap.sh with platform 2012.4.0.0

Now the warning message includes mention of the input data that is being
ignored because hpc can't deal with it.

This involves a change of behaviour: when the HPC version cannot be
determined, this is now a fatal error, rather than just assuming an old
version.

Fix path (mis)handling when scanning for modules (fixes haskell#1160)

Makes it possible to write unit tests for code under Distribution.Client.*.

Split cabal-install into library and executable components.

Improve warning for old versions of HPC

s/AmbigousParse/AmbiguousParse/

`+++` and `<++` are right-associative, so `a +++ b +++ c <++ d` is parsed as
`(a +++ (b +++ (c <++ d)))`.

Fixes haskell#1159.