* patch-aa: Use boost::none instead of the default none_t() constructor, which
  has disappeared in recent boost versions.
* Fix dependencies: add eigen3, libtool-ltdl and a bunch of boost-libs

Bump PKGREVISION

The public headers of this package are such that a direct dependency on log4cpp
and eigen3 must be added for all dependant packages.

patch-aa: drop and attribute((unused)) that g++ does not accept in a
BOOST_STATIC_ASSERT construct.

patch-ab: fix PLIST for older cmake by not installing versionned symlinks for a
module.

Bump PKGREVISION

Amend patch-aa fixing the CMakeList : creating a library with no source seem to
not be supported anymore. But this is in practice not needed, this was only
done for the "linux" target, while other target were building libraries
normally. I really don't get the reason to handle linux differently.

Bump PKGREVISION.

After commit:b014d103, one cannot just set PREFER_ALTERNATIVE.* to handle
enforced alternatives with PKGREQD. Instead, set directly the alternative
choice and check it's valid by restricting the selection list to this single
item.

Pass the correct sphinx-build exectuable (for the correct python version) to
the build system.

The blender version output by blender -v might not be on the first line.

This is EOL.

The packaged version was outdated, and all supported distributions provide a
newer alternative. No need to package it here, a sysdep is enough.

python33 is EOL, no package is using it and it should not be used.

Note: python32 has to be kept for compat with ubuntu 12.04, which is also EOL
but that still has to be supported.

Guilhem Saurel authored Apr 05, 2018 and Anthony Mallet committed Apr 05, 2018

Affected packages:

* gepetto-viewer
* gepetto-viewer-corba
* hpp-fcl

Guilhem Saurel authored Mar 22, 2018 and Anthony Mallet committed Apr 05, 2018

Changes since 1.0:

Cf. optimization/roboptim-core

Guilhem Saurel authored Mar 20, 2018 and Anthony Mallet committed Apr 05, 2018

Changes since 3.0:

* ColMajor/RowMajor support has been improved (cf. %89). Default is back to
  ColMajor since this is Eigen's default mode, but that can be changed with a
  CMake variable.
* Allocation checking has been improved (cf. %92).
* Multiplots are now available with the matplotlib plotting backend (cf. %94).
* Added vector_t and bool to the solver parameter types. As a consequence,
  std::string parameters should not rely on automatic conversion from const
  char* (cf. 7a0bbb7).
* Renamed scale[s]* to scaling* (cf. 434559c). Previous methods/typedefs are
  currently kept for backward compatibility, but marked as deprecated.
* Add new StructuredInput helper (cf. %96).
* Add support for matrix plotting with matplotlib (cf. 5abd27e / af48e9b).

Changes since 2.0:

* New features:
    * Lots of new functions (cos, sin, etc.), operators (plus, minus, scalar,
      map, etc.) and decorators (cached function etc.),
    * Callback system (logger, callback multiplexer),
    * Simple Matplotlib support for visualization,
    * Function pools,
    * Set argument names (useful for plotting).
* Improvements:
    * Support Eigen::Ref: now RobOptim functions accept blocks/segments of
      Eigen matrices as input,
    * Improved CachedFunction with LRU cache,
    * Automatic conversion of constraints/cost function types when creating
      problems,
    * Faster forward-difference rule for finite differences.
* Other changes:
    * Removed exception specifiers (void function(...) throw())
    * Storage order was changed from Eigen's default (column-major) to
      row-major. The storage order is available in the GenericFunctionTraits
      (StorageOrder).
    * Move metaprogramming magic to roboptim/core/detail/utility.hh
    * Merge roboptim/core/visualization/util.hh with roboptim/core/util.hh

Changes since 1.0:

The version 2.0 of roboptim-core now depends on Eigen for matrix computations
by default.
Traits allow the user to use its own type. Support for Eigen dense and sparse
matrices is built-in.

Alternative summary in `make show-options` can be quite obfuscated without
a manual. So display a short legend for each symbol present in the output of
the current alternative selection.

In combination with commit:b014d103, this allows to list all valid/existing
package names for all alternatives (read: python alternatives ...) instead of
just those allowed to build.

This also complements commit:d54a670a and will improve the bulk build results
for those packages that exist but are disabled (read: python versions not
available by default on a system).

A package was completely disabled if user preferences for alternative selection
(if any) could not be satisfied. This was useful in particular with python, to
not try building packages with python versions that are by default not
available on a system.

However, completely disabling a package because of this creates nasty side
effects if one just want to extract the package or otherwise check its
requirements. This could also break the "mirror distfiles" bulk builds.

Thus the alternative selection now considers all available alternatives (still
sorted with the user preference first) and two kind of errors can be generated:

 1. The first kind is when no alternative exist at all. This is still a fatal
    error, this indicates something really wrong with the dependencies.

 2. An alternative exists, but is not in the user preferences. This is a "soft"
    error (triggered only for cbbh (aka. build) targets and not cbeh
    (aka. extract) targets).

This solves the "mirror distfiles" issue while preserving the good old
behaviour of not building a package with a user-disabled alternative.

When dealing with multiple constraints on a package (e.g x>n x>m), it could
happen that the lower/upper bounds on the constraint would be incorrectly
computed if the different version numbers would have a different number of
comonents AND the other side of the bound would be longer as well
(e.g: x>=1<3.0 x>=2.1). This was in practice not triggered by any existing
package.

Fix that by making sure to reset the array encoding the min/max constraint
before updating it.

This supports only ros <= groovy, which is almost obsolete. This package is
probably not used anymore.

OK Florent

This depends on graphics/rcpdf-interface, which supports only ros <= groovy,
which is mostly obsolete. This package is probably not used anymore.

OK Florent

Strict sets skip packages not explicitely listed in their configuration. But
they still process packages dependencies to keep the build order topologically
sorted. The packages outside the strict set are just skipped.

The counter of processed packages must be incremented no matter if they are
skipped or not, otherwise the displayed "% done" is incorrect. This could be
seen on bulk builds for the wip/ directory.

Changes since 20170123:

 * Discard bulk logs if they cannot be inserted in the result database
   (avoiding to have them inserted in the wrong database by a future
   session). Database update failure almost never happen, and the logs are not
   critical, so loosing them is actually a better option than manually cleaning
   a wrong database update.

 * Make sure rbulk-dispatchd starts jobs with a clean log file. This could
   happen after an unclean shutdown of the daemon which could leave old files
   around.

The installed .pc and headers of this package leak many internal
dependencies. Thus, all of them has to become direct dependencies of depending
packages.

This fixes in particular motion/sot-core not finding bullet at build time.

Do not bump PKGREVISION. The missing entries were just about doc, nobody
wants to read this ;)

ros-gencpp is now at version 0.6 for ros kinetic and ros-lunar.

Changes since 2.8.1 not available upstream.

Changes since 2.8.1 not available upstream.

However, visible changes include:

 * Fix build with recent g++ and boost

Changes since 2.8.1 not available upstream.

However visible changes include:

 * Fix build with recent g++ and boost
 * patch-ac for boost >= 1.61 applied upstream

Still need patch-ab to fix gcc detection for gcc >= 7

Changes since 2.8.1 not available upstream.

However, visible changes include:

 * Renamed headers from include/log4cpp to include/orocos/log4cpp
 * Renamed libraires from liblog4cpp.so to liborocos-log4cpp.so
 * Renamed pkg-config file from log4cpp.pc to rocos-log4cpp.pc

When a package is "not available" for a given platform, the print-pkgnames will
(correctly) return an empty list. This makes tsort-set think that no package
is matching here.

This is bad because:
  * bulk builds results won't be triggered for this package (and appear as
    "unknown" instead of "cannot be built here" / "not available").
  * package sets without NULLGLOB option will incorrectly raise an error

Fix by using the default PKGNAME in tsort-set when print-pkgnames returns the
empty list.

Should address issue #154 and repair the dependencies of the debian packages
from wip/.

Changes since 1.2:

 * Honour PKG_PATH when resolving robotpkg dependencies (#154)
 * Skip directories when running dpkg-query -S (to avoid getting a huge
   list of unrelated dependencies).

It was initially (and still is) a feature to allow, in a in package set
definition, a package pattern expanding to the empty list. However, this can in
practice hide a typo which may be hard to track down if the package set is
otherwise not empty.

This commit adds a new directive PKGSET_NULLGLOB, which can be specialized per
package set with PKGSET_NULLGLOB.<set-name>, that controls whether a set should
allow patterns exanding to an empty list.

PKGSET_NULLGLOB defaults to no, thus now forbidding empty expansions by
default. It can be turned back to 'yes' to restore the previous behaviour.

If, for whatever reason, the packaging of a file and repository updates fails
before completion (e.g. because the building machine is powered down), even
with commit:25f649f the package repository may be left in an inconsistent
state.

To prevent further bulk builds to think that the build was successfull, a
marker file is left along with the binary package file, and removed only at the
very end of the packaging phase. If the file is present when the bulk target
starts, this indicates a previous failure that could not be recovered and the
packaging will be redone.

This should address issue #151

The 'package' target writes outside $WRKDIR. After building the binary package
file itself, a few admistrative tasks must be performed to update the binary
package repository and invoke additional targets like building a debian package
from the binary package.

In case of error in the middle of these administrative tasks, it is safer to
rollback all modifications that were done to the packages repository, so that
it stays in a consistent state.

This is done like for the `install` target, by using a recursive make
invocation, tracking any error and invoking the rollback target if necessary.

As a side effect, a new "depackage" toplevel target is added, which is
available to the users to manually remove a package from the repository.

* patch-aa: Disable tagfiles, this triggers a segfault in doxygen-1.8.6 while
  processing the gepetto-viewer-corba.doxytag. This does not change the
  documentation anyway: the generated "related pages" is empty even with
  tagfiles on.

Fix the -l directive for the boost_python library.

Bump PKGREVISION