Move template specializations of matrix capabilities into header. Restrict call to methods solve(), invert() to invertible matrices.


[[Imported from SVN: r7107]]

I shouldn't be doing patch reviews after 2300

Here is Andreas' commit message:

>From 1049843fa0dbc6b340817b0cd69959e742e80d7a Mon Sep 17 00:00:00 2001
From: Andreas Lauser <Andreas.Lauser@iws.uni-stuttgart.de>
Date: Wed, 6 Feb 2013 23:33:28 +0100
Subject: [PATCH] once again: fix dune_unused.m4

my previous attempt was only partially correct: the compiler did not
chocke because the unused attribute was placed behind the variable,
but because the third argument of AC_TRY_COMPILE is placed inside the
main function. it seems like c++ does not support nested functions, so
the test still failed, leading to the equally undesired result that
DUNE_UNUSED did not have an effect anymore. Mea culpa!

(this time I have at least double verified, that the compiler does not
bail out. last time I probably mis-interpreted the wrong lines as the
output of the dune_unused test in config.log.)


[[Imported from SVN: r7106]]

Thanks to Andreas Lauser for this patch.

Here is his own commit message:

From: Andreas Lauser <and@poware.org>
Date: Mon, 4 Feb 2013 15:34:22 +0100
Subject: [PATCH] fix the test for __attribute__((unused))

it contained two errors which "anihilated" themselfs: the first was a
syntax error in the test itself (the UNUSED macro must be placed
before the identifier of the variable), which caused the compiler to
always return an error. this was compensated by an incorrect
invocation of the AC_TRY_COMPILE macro which featured an empty
argument as third argument instead of the action to be executed if the
test succeeded. this in turn always lead the forth argument (which was
intended by the test to be executed if the test succeeds but was
actually executed if it failed) to be always triggered. In turn this
leads me to the conclusion that DUNE was never compiled using a
compiler which does not support __attribute__((unused)) since this
macro was introduced...


[[Imported from SVN: r7105]]

[[Imported from SVN: r7104]]

[[Imported from SVN: r7103]]

[[Imported from SVN: r7102]]

[[Imported from SVN: r7101]]

[[Imported from SVN: r7100]]

[[Imported from SVN: r7099]]

[[Imported from SVN: r7098]]

(original patch provided by tomalk)

[[Imported from SVN: r7097]]

[[Imported from SVN: r7096]]

[[Imported from SVN: r7095]]

[[Imported from SVN: r7094]]

[[Imported from SVN: r7093]]

[[Imported from SVN: r7092]]

[[Imported from SVN: r7091]]

[[Imported from SVN: r7088]]

[[Imported from SVN: r7087]]

[[Imported from SVN: r7086]]

Merge rev. 7083 from branches/cmake

[[Imported from SVN: r7084]]

[[Imported from SVN: r7083]]

[[Imported from SVN: r7082]]

[[Imported from SVN: r7079]]

[[Imported from SVN: r7078]]

Patch by Andreas Lauser.  Thanks!

[[Imported from SVN: r7076]]

[[Imported from SVN: r7075]]

[[Imported from SVN: r7074]]

[[Imported from SVN: r7073]]

[[Imported from SVN: r7072]]

[[Imported from SVN: r7071]]

[[Imported from SVN: r7070]]

BTW: Currently they are not at all activated in dune-common.

[[Imported from SVN: r7069]]

Fixed function call in test code, because indexSet_ is a pointer. Clang complained.

[[Imported from SVN: r7068]]

Add support for boost::hash

This patch adds boost::hash as a fallback option for Dune::hash
if neither std::hash nor std::tr1::hash can be found. We only
have to include the appropriate header and import the name into
the Dune namespace, as the built-in extension mechanism of boost::hash
will automatically pick up the global hash_value() functions.

To test whether the mechanism actually works, the patch also adds
boost::hash to the list of hash implementations tested in
bigunsignedinttest.

Kudos to Steffen Müthing, see FS#1192

[[Imported from SVN: r7067]]

Make bigunsignedint hashable using Dune::hash

This patch adds the required hooks to enable hashing of bigunsignedint
and makes sure the hasher can be invoked in bigunsignedinttest.

Making bigunsignedint hashable also makes it possible to use hash-based
containers for EntityIDs with YaspGrid and SGrid, as those grids implement
the EntityID as a plain bigunsignedint.

The patch also adds a little test to bigunsignedinttest, trying to hash
a bigunsignedint with Dune::hash as well as any of the detected backends
(std::hash and std::tr1::hash).

Kudos to Steffen Müthing, see FS#1192

[[Imported from SVN: r7066]]

Add basic support for hashing using Dune::hash

This patch introduces the functor Dune::hash in a similar manner to
other C++11 functionality by importing the definition from namespace std.

Apart from that, the patch also provides some basic functionality to simplify
making user-defined types hashable, in particular functions for combining hashes
and hashing iterator ranges as well as a mechanism for defining the struct
specializations required by std::hash and std::tr1::hash in the correct namespace.

The current implementation will first try to use std::hash and, if that fails,
attempt to fall back on std::tr1::hash. If that is not available either, hash
support is disabled as there is no homegrown fallback for now.
The extension mechanism will register the type with both std::hash and std::tr1::hash,
if available. This should make it possible to use unordered_{set,map} from both C++11
and TR1 without having to specify a custom hasher.

Kudos to Steffen Müthing, see FS#1192

[[Imported from SVN: r7065]]

Remove cast operator that caused Clang warnings.
This fixes FS#1207.

Kudos to Ansgar Burchardt for the patch.

[[Imported from SVN: r7064]]

Ignore eigenvaluetest.
Reordner member variable assignments to avoid pedantic warnings in
shared_ptr.
Only activate boundary checking in dynmatrixtest if not already active.

[[Imported from SVN: r7063]]

[[Imported from SVN: r7062]]