On macOS, the build of libxxhash is performed by a bare `make`
launched by `MakePackage`s default `build` phase. As `prefix`
is not set at this point, the library gets compiled with a hard-coded
install name of `/usr/local/libxxhash.dylib`. Downstream clients,
in the tested case Root, will then fail to build as the install name
is incorrect.
Move setting of `prefix` from `install` to `edit` so that it
propagates to subsequent `build` and `install` phases.
* QE hdf5 is now a multi-valued variant. Update conflicts accordingly.
* Update dependencies for hdf5 multi-valued variant. Bug fix: high-level hdf5 library needs to be a hard dependecy.
* Support for HDF5 serial interface. Note that this is not supported in QE autoconf, but will be fixed in the next official release.
* Support in QMCPACK for CUDA_ARCH flag.
* No need to have explicit CUDA variant anymore.
* Internal QMCPACK CMake assumes a minimum GPU architecture if none is provided.
* cp2k: archive the Makefile after a successful installation
* cp2k: all checksums are now sha256
* libint: all checksums are now sha256 + added versions 2.4.X
* libxc: all checksums are now sha256 + added version 4.3.2
* cp2k: os.path.join instead of join_path + simplified openmp flags
* cp2k: turned into a MakefilePackage
* cp2k: refactored edit method so that Makefile writing happens last
* icu4c: Add cxxstd variant
For versions supported in Spack, ICU4C defaults to the C++11.
Provide a variant to default to this, plus options for C++14
and 17 if dependees require for ABI compatibility.
* icu4c: cxxstd variant cannot be multi
* Updated Geant4 and dependencies clhep and vecgeom to use cxxstd= variant so that it is applied consistently.
* Make standalone geant4 data packages that are used when data is not installed with geant4. This allows installing the data files are externals that can be common to many compiler version.
* move geant4 data package dependencies to umbrella package geant4-data
Fixes#7855Closes#8070Closes#2645
When searching for library directories (e.g. to add "-L" arguments to
the compiler wrapper) Spack was only trying the "lib/" and "lib64/"
directories for each dependency install prefix; this missed cases
where packages would install libraries to subdirectories and also was
not customizable. This PR makes use of the ".headers" and ".libs"
properties for more-advanced location of header/library directories.
Since packages can override the default behavior of ".headers" and
".libs", it also allows package writers to customize.
The following environment variables which used to be set by Spack
for a package build have been removed:
* Remove SPACK_PREFIX and SPACK_DEPENDENCIES environment variables as
they are no-longer used
* Remove SPACK_INSTALL environment variable: it was not used before
this PR
* Added a package for the MDAnalysis toolkit.
* Updated LBANN, Hydrogen, and Aluminum to not require CUDA-aware MPI
even when using CUDA. Fixed small bugs within both LBANN and Hydrogen
packages. Added requirements for breathe and cereal to LBANN.
* Update use case for doc dependencies.
* Cleaning up some of the LBANN software stack dependencies.
Add variant to support compiling against main C++ standards if needed
for ABI compatibility. Use 'default' value as Xerces-C autoconf system
does not enforce a minimum, or set a default, C++ standard.
* Extend current recipe for root package
* Add removed old version
* Add dependency type for Python and R packages
* Add compiler variants
* Adding new root version: 6.14.00
* Add back the minimum cmake version required
* LZ4 and xxhash support for old (using builtin options) and new versions (external packages)
* Set PYTHONPATH in setup_environment
* Address typos and issues pointed by HadrienG2
* Add newer ROOT versions
* Solve support for asimage
* Define incompatible variants
* Organize old dependencies
* Add support for minuit
* Add support for pythia6
* Remove unneeded list_url
* Update license header
* Remove block of commented old dependencies
* Contrain Qt version
* Comment pythia8 variant/dependency, not supported by spack
* Force not to build pythia8
* Adding a package for cardioid.
* Changing package to pass flake8
* Removing template leftovers per @ax3l 's advice
* Adding a homepage so tests pass.
* Removing whitespace for flake8
* Changing the name of master branch to match spack standards.
Add patch to fix Fortran mangling in CMake test
This is a fix for builds with GCC 6/7/8 and CMake 2/3. It includes
a link to the issue where this is reported. For now, the Spack
package applies the patch to all releases >= 2.0.2
This package depends on both python 3.5+ and node-js which depends on
python 2.7/2.8 to build. Currently it can only be installed if the
node-js package's python dependency is manually removed, so node-js
will build with python found on the system rather than provided by
spack. Upcoming commits may allow spack to reconcile this
automatically.
* xsdk@develop: switch trilinos from @master to @develop
* xsdk@develop: dtk - fix trilinos dependency on dtk.
- dkt-3.0 should be used with trilinos-12.4*
- dtk/master should be used with trilinos@develop
A version >= 0.9.4 required for py-black. Added latest, which is
0.10.0. Manually verified that the .tar.gz for v0.9.3 used to generate
the sha256 matched the previous md5.
Add tribool.patch to add explicit casts from tribool to bool in two
places. Boost, starting with version 1.69, turned on explicit
conversions for tribool, which breaks the build for dyninst.
This is merged into the current dyninst develop and is expected to be
part of the next release, so the version range uses the current
release as an upper bound.
Closes#10168
* Add versions 4.1.0.0 and 4.0.12.0
* Replace md5 checksums with sha256
* Logic to generate a helper script for versions earlier than 4 was
not running. This corrects the branch which controls whether this
script is generated.
* Constrain Java dependency to version 8 (GATK does not work with
Java 11)
* Avoid adding scripts that have dependencies not handled by Spack
such as gcloud and google-chrome. These scripts are good candidates
for later extensions.
Closes#2622
This retrieves and installs the binary of Oracle's OpenJDK (an
open-source JDK implementation). The logic is largely based on the
Spack JDK package (Oracle's closed-source JDK implementation).
For now this is only supports Linux. If the download could be
customized based on OS (for the same version), binaries are also
available for MacOS and Windows.
Build system was changed to use a Makefile. Therefore older versions
(which don't support this) are removed. The new build system also
supports testing by default.
Fixes#7372
Added patch method which renames config.h in the fftw subdir of the
source tree. fftw 2.1.5 appears to ship with a copy of this file
with all defines commented out. This gets read by the #include
directives instead of the version in the build directory with the
correct defines. As a result, many C preprocessor macros left
undefined, including F77_FUNC_ which causes the bulk of fttwf77.c
to be skipped due to an #ifdef, so fftw_reverse_int_array et al
not included in library. Fixes#7372
Also fixed some inconsistencies with the handling of quad and
long_double in specs between the configure method and the build,
check, and install methods.
The Mercury package previously depended on libfabric@develop on
Cray platforms in order to pick up a fix for the GNI provider that
was only available in git. This is no longer needed now that the
libfabric package is at 1.7.0.
Add new versions for py-keras. Versions of py-keras at or after
2.2.0 require keras-processing and keras-applications packages.
Add these packages to Spack and add dependencies to the py-keras
package.
* Add version 11.2.68
* Add variant to disable AVX2 for older hardware
* Change the makefile so libraries and includes are not placed
under $PREFIX/usr
* Updated URL and homepage
Added versions 1.0.0-rc1, 0.6.1, 0.6.0, 0.5.1
hwloc dependency was removed after 0.5.1, so the related configure
option is not specified after 0.5.1
Require py-matplotlib dependency version 2.2.3 (version was not
specified before)
Fixes#8343#4154#7370Closes#4155
All but one of the backports packages were using "pkgutil-style"
namespacing, but configparser was using "pkg_resource-style".
A namespace needs to be one style or another. This adds a new
configparser version which switches to using the pkgutil-style
namespace approach.
With this update, one can now install flake8 with Spack and use
it by loading modules.
The PREFIX variable in the makefiles is directly used to construct the
install name on the library metadata on darwin. If not passed, the
default /usr/local is used and the resulting librhash.dylib contains
usage informaiton of /usr/local/lib/librhash.0.dylib. This adds the
PREFIX= variable directly to the make command so the resulting usage
metadata contains the correct valid library path.
* flux-core: add v0.11.0 and updated dependencies
libuuid hasn't been a direct dependency of flux-core since 2015 (well
before any of the versions listed in the package).
The yaml-cpp dependency moved to flux-sched when we moved the jobpsec
library from core to sched.
Some python3 incompatibilities were found in 0.11.0, so bump the python3
support back a version.
* flux-sched: add v0.7.0 and make boost dependency more precise
* llvm: Bump version to 7.0.1
* llvm: Added perl-data-dumper build dependency for openmp
* llvm: Enable exception handling and RTTI
Useful to have turned on in general with RTTI but also necessary
to workaround some lldb stability issues with some versions of
libstdc++.
* Update the PMIx package
- Add latest releases to the list.
- Add dependency on hwloc for versions >= 3.0
- Flag preferred versions for libevent and hwloc
- Improve the description
- Point the project URL to the correct site
Signed-off-by: Ralph Castain <rhc@pmix.org>
* Only install the hwloc dependency for versions >= 3.0.0
Thanks to @ax3l for the suggestion
Signed-off-by: Ralph Castain <rhc@pmix.org>
* Add myself as a listed maintainer
Signed-off-by: Ralph Castain <rhc@pmix.org>
* Add warnings for existing conflicts. New versions of QMCPACK require a compiler with C++14 support.
* QMCPACK now requires a newer release of CMake.
* Switch QMCPACK to using FFTW-api provider.
* Relax dependencies on QMCPACK +da variant.
* Add QMCPACK 3.6.0 release.
* QMCPACK compiler-CUDA conflicts.
* QMCPACK fix typo in expressing conflict.
* Conditional needed since these variables are only used when FFTW is a dependency to QMCPACK.
* QMCPACK QE variant dependency was underspecified. Must include MPI as a hard dependency on QE.
* Update QMCPACK dependencies for HDF5, CMake and Boost.
* QMCPACK clean-up, more consistent use of spec vs. self.spec.
* Add support for collective I/O variant of QMCPACK.
* Update QMCPACK conflicts for QE and PHDF5 variants.
* Use built-in install method for QMCPACK starting with 3.6.0
* QMCPACK compiler C++14 conflicts were off by one version.
* QMCPACK package needs to setup_environment method.
* In support of PR#10252, get rid of slashes in string.
* Migrate these to the CUDA package and make a seperate PR.
* In support of PR#10252, get rid of join_path for nexus.
* In support of PR#10252, work around concretizer limitation.
* In support of PR#10252, convert to conditionals to if-else clauses for compactness.
* In support of PR#10252, get rid of extra space and new line in warning messages.
* Updated MPI and TAU versions in the TAU package
* Updated PDT versions
* Added support for Python (+python), default to `False`
* Cleaned up binutils, python, mpi usage.
* Updated binutils install to add additional header files to the include/extra directory.
* Resolving review problems with TAU and binutils packages.
The binutils package was updated to include support for installing
the ELF headers as a +headers variant. The TAU package was updated
to use all known Spack dependencies.
* r-gamlss-data: new package at 5.1-0
* r-gamlss-dist: new package at 5.1-1
* r-gamlss: new package at 5.1-2
* r-gamlss: add mass, survival, nlme imports
Fixes#10361Fixes#6634
When mkl isnt in the spec, still write an explicit mkl entry to
prevent numpy from looking in system dirs and using a system install
of mkl. Do the same for openblas and atlas (although mkl was the only
system install reported to interfere with Spack's choice).
* Updating nalu-wind, nalu, openfast, and tioga packages. Includes
improved handling of +shared variant.
* Add fftw and catalyst support to nalu-wind.
* Restrict ParaView to use matplotlib versions before 3 (which
require Python 3)
* py-modred: New package
Parallelized python module for finding modal decompositions and
reduced order models.
* fix license
I think?
* Add dependency on py-numpy
and remove commented out code
* vigra: new package
* Better constraints and types on some depends_on. Use site_packages_dir variable for Python install dir. Explicit switch-offs where available. Change some local variables to suit style checks.
* Take care of lines too long
* Explicit WITH_HDF5=ON
* docbook processing works correctly for gdk-pixbuf
1. The various bits of documentation in gdk-pixbuf include hardcoded
references to dtd's and stuff at their canonical, Internet,
locations. BUT, gdk-pixbuf runs xslt-proc with the `--nonet`
option, which forbids it from using the network. Sadness ensues.
Traditionally folks use XML Catalogs to map these to local copies.
Our docbook-xsl package wasn't setting the appropriate env var in
its dependents environments to use our catalog.
Now it does. Less sadness ensues.
2. If we're going to use these things, we should depend on them.
* Add gdk-pixbuf's bin to build environment
The "post-install.sh" script uses gdk-pixbuf-query-loaders, which was
installed earlier.
If py-psyclone can set its own bin on its PATH, so can we...
* Make gdk-pixbuf use recognizable docbook location
Our docbook-xsl package assumes that the canonical location is
`http://cdn.docbook.org/release/xsl/current/manpages/docbook.xsl`, but
the gdk-pixbuf's meson build script uses
`http://docbook.sourceforge.net/release/xsl/current/manpages/docbook.xsl`.
This means that our XML Catalog doesn't fix the reference and sadness
happens.
Just patch the build that we see what we want to see, then we can make
it all go away.
* Add commentary re docbook patch
* Make catalog helper a property
Thanks @adamjstewart!
* Run tests if/when run_tests is true
Thanks @adamjstewart!
* Tune up dependencies
Thanks @adamjstewart!
* Wordsmit commentary
This adds a lot of fixes and updates for alpine ascent and its dependencies:
* add support for static (via ~shared) and use vtk-m 1.2
* update vtkh package to output cmake configure file and pinning it to vtkm 1.2
* make conduit respect ~python
* fix ascent python logic
* consistant cmake usage
* conditionally add tbb in ascent if vtkh
* enable openmp
* more robust handling of variants
* update ascent and conduit for static builds
* add optional support for mfem in ascent
* enable mfem conduit support for ascent
* add optional adios dep to conduit
* remove ver req from conduit
* ascent: remove confusing comment
* tweaks to conduit and ascent pkg recipes
* fix typo in conduit package
* pref conduit master
* changing mfem to depend on conduit@master to get updated relay
* restore use of conduit 0.3.1 or greater for mfem
* set master as preferred conduit version
* allow mfem to use conduit master
* adding rover package and editing ascent
* updating vtkm cmake dep
* guard ascent python support on +shared
* removing rover to simply ascent package
* add fortran variant to conduit, to allow us to turn off conduit support even when a fortran compiler is specified
* fix fortran compiler check so it can work on cray systems
* working towards cuda fix for vtkm lagrange filter
* update ascent package with more variants, and patch to avoid nvcc issue
* hdf5 flags fix for BGQ
* add post install test
* add testing to ascent
* add testing of the using-with-make example
* add ctest output on error for run_tests
* dealii: fix concretization of xsdk package
* tests: add concretization tests for deal.II and xSDK, which are often broken due to limitations in the concretizer
* use pytest.mark.parametrize
Updated LBANN, Hydrogen, and Aluminum to not require CUDA-aware MPI
even when using CUDA. Fixed small bugs within both LBANN and Hydrogen
packages. Added requirements for breathe and cereal to LBANN.
Fixes issues #9613 and #3209.
Remove the spack wrapper directories from PATH for the bootstrap step.
This was breaking the build for Cray (and other cross-compile) because
bjam was built for the BE and died on SIGILL on the FE.
This only affects building bjam. The boost libraries are still built
the normal spack way with the spack wrappers.
Add variants 'cray' and 'bgq' to build for Cray and Blue Gene compute
nodes. These systems need separate options to find the MPI compiler
for hpcprof-mpi.
The 4.0.1 release is the culmination of the SUNDIALS 4.0.0 preview releases,
i.e. the 4.0.0-dev versions. As such, the 4.0.0-dev releases are now
being removed.
* Explicitly listing blas and lapack sections in site.cfg for numpy when using netlib-lapack so that scipy can find them as well.
* Reducing code to use a function call instead of copying and pasting.
* Fixing flake8 errors.
* Fixing python 3.6 error when site.cfg lists library_dirs multiple times.
* Formatting.
* Verifying blas and lapack are enabled before writing to site.cfg.
* Fixing flake8 errors.
* Fixing conflicts since this package file has been updated.
* Fixing flake8 stuff.
* Handling blas and lapack variants for netlib-lapack in site.cfg for py-numpy.
* Don't write netlib-lapack site.cfg file if no blas or lapack variants enabled in py-numpy.
* Removing redundant if statement.
* Fixing mistake in py-numpy site.cfg generation.
* Separating blas and lapack further in site.cfg for netlib-lapack in py-numpy.
Allow customizing views with Spec-formatted directory structure
Allow views to specify projections that are more complicated than
merging every package into a single shared prefix. This will allow
sites to configure a view for the way they want to present packages
to their users; for example this can be used to create a prefix for
each package but omit the DAG hash from the path.
This includes a new YAML format file for specifying the simplified
prefix for a spec in a view. This configuration allows the use of
different prefix formats for different specs (i.e. specs depending
on MPI can include the MPI implementation in the prefix).
Documentation on usage of the view projection configuration is
included.
Depending on the projection configuration, paths are not guaranteed
to be unique and it may not be possible to add multiple installs of
a package to a view.
* elfutils: option to install elf.h header file
Add variant 'elfh' (default False) to install the elf.h header file to
the include directory. This provides access to a newer version of
elf.h and allows an elf application to build with the same elf.h that
libelf.so was built with.
Add version 0.175.
* Drop the variant 'elfh' and always install elf.h instead.
* new package: git-fat
* backport of rsync -> older versions no longer available
Change-Id: I3a41f17df0f88ca11e08ea6fd2f2d381435e0f09
* fix legal stuff and use new git attribute
Change-Id: Ice156dfe8ad4137a861f7b2938ee92f4024142be
* use date as version number rather than hash
Change-Id: Ib537659e368f69124895306d6a6ac0c5e36061a7
* only use the date once
Change-Id: I2592fa28aaa53fed949ddc82959d712ff3f6f752
* Initial compiler support
* added arm.py
* Changed licence to Arm suggested header
* Changed licence to the same as clang.py
Main author of file is Nick Forrington <Nick.Forrington@arm.com>
Minor changes by Srinath Vadlamani <srinath.vadlamani@arm.com>
* compilers: add arm compiler detection to Spack
- added arm.py with support for detecting `armclang` and `armflang`
Co-authored-by: Srinath Vadlamani <srinath.vadlamani@arm.com>
* Changed to using get get_compiler_version
* linking to general cc for arm compiler
* For arm compiler add CFLAGS to use compiler-rt rtlib.
* Escape for special characters in rexep
* Cleaned up for Flake8 to pass.
* libcompiler-rt should be part of the LDFLAGS not CFLAGS
* fixed m4 when using clang to used LDFLAGS. Fixed comments for arm.py to display compiler --version output with # NOAQ for flakes pass.
* added arm compilers
* proper linked names
* add the tfel package
* fix the tfel package
* fix the tfel package
* fix the tfel package
* Taking Adam J. Steward' remarks into account
* fixes trailing white spaces
* Update description
* Update dependencies following @adamjstewart adices
* Style fixes
* Style fixes
* Add java optional support
* add the maintainers attribute (following @alalazo advice), disable interface not selected (following @adamjstewart advice)
* flake8 fixes
* Fix Cast3M and python-bindings support. Python detection is made compatible with cmake'FindPythonLibs module (at least how it is used in TFEL)
* Style fixes
* Style fixes
* Fix test on python version
* Follow @adamjstewart advices: code is much cleaner and readable
* Small fix
* Small fix
* Add comment
* Small fix in cmake option
* try again (trying to overcome Travis CI unstable build process)
* Add support for the MFrontGenericInterfaceSupport project (MGIS)
* Style fixes
* Package documentation update
* Package documentation update
* Fix a typo thanks to Andreas Baumbach review
* Follow Adam J. Stewart advices
* Fix type
* bowtie: new versions and %gcc@8.0.0: fix
Bowtie development shifted from Sourceforge to GitHub.
This commit adds several versions from GitHub, using the archive
tarballs. Note that the URL for 1.2.2 tarball is actually an '_p1'
tag....
It leaves the old 1.2 release download info in place.
Versions 1.2.0 and newer come from GitHub (I'm unsure if the 1.2 and
1.2.0 are equivalent).
Include a fix that enables %gcc@8.0.0: from:
https://github.com/BenLangmead/bowtie/issues/87
But, v1.2.2 has trouble with "newer" gcc's, so it only adds v1.2.2 for
%gcc@6.0.0:.
Feel free to tighten this. I know that:
- 1.2 -> 1.2.2 work with %gcc@5.5.0;
- 1.2 -> 1.2.1.1 work with %gcc@8.2.0; and
- 1.2.2 fails with %gcc@8.2.0
* Tighten to `conflicts('%gcc@8:', when='@1.2.2')`
* Point 1.2.2 and 1.2.2_p1 at the 1.2.2_p1 tarball
* Update package.py
Doxygen has migrated from a private SVN to GitHub. This PR updates the URLs and adds versioned commit hashes from GitHub. It also includes version 1.8.15 as the latest option.
* Update package.py
This enforces conventions that allow for correct handling of
multi-valued variants where specifying no value is an option,
and adds convenience functionality for specifying multi-valued
variants with conflicting sets of values. This also adds a notion
of "feature values" for variants, which are those that are understood
by the build system (e.g. those that would appear as configure
options). In more detail:
* Add documentation on variants to the packaging guide
* Forbid usage of '' or None as a possible variant value, in
particular as a default. To indicate choosing no value, the user
must explicitly define an option like 'none'. Without this,
multi-valued variants with default set to None were not parsable
from the command line (Fixes#6314)
* Add "disjoint_sets" function to support the declaration of
multi-valued variants with conflicting sets of options. For example
a variant "foo" with possible values "a", "b", and "c" where "c"
is exclusive of the other values ("foo=a,b" and "foo=c" are
valid but "foo=a,c" is not).
* Add "any_combination_of" function to support the declaration of
multi-valued variants where it is valid to choose none of the
values. This automatically defines "none" as an option (exclusive
with all other choices); this value does not appear when iterating
over the variant's values, for example in "with_or_without" (which
constructs autotools option strings from variant values).
* The "disjoint_sets" and "any_combination_of" methods return an
object which tracks the possible values. It is also possible to
indicate that some of these values do not correspond to options
understood by the package's build system, such that methods like
"with_or_without" will not define options for those values (this
occurs automatically for "none")
* Add documentation for usage of new functions for specifying
multi-valued variants
* PARALLEL-NETCDF: Update new version and location
PnetCDF-1.11.0 is released.
Also, the canonical download area has been changed and they are now using git, so can also provide a develop and master checkout.
One issue is that they changed the name of the tar files, so 1.11.0 needs special handling (and future versions will also).
All checksums at new location match the checksums from the old location.
* Address concerns in review
Added a `url_for_version` function to handle the name change in the tar files at version 1.11.0 and later. Updated description to match current shown on website. Updated `homepage` setting since it has recently moved.
* Address issues in second review
- in many files, regular strings were used in places where raw strings
should've been used.
- convert these to raw strings and get rid of new flake8 errors
* express: new version, use tags and fix gcc@6.0.0:
Express fails to build with gcc@6.0.0:.
The debian project [has a
fix](https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=811859) but
they don't seem to have pushed it upstream.
I've opened an issue and a PR in eXpress repo, but eXpress isn't
actively developed, so I'm fixing it here too.
Since the Spack package was created, the eXpress team tagged their
releases. I've updated the package to use the tags.
Version 1.5.1 used to be known as 2015-11-29 (same commit). 1.5.2 is
new(er).
* Make flake8 happ{y,ier}
Apply fix from aspell issue (519) for a pointer dereference bug that
newer versions of gcc won't let slip past.
There hasn't been a release that includes the fix, this applies the
change to the latest release.
There's a missing break in a switch statement that newer gcc's
dislike.
Our #4696 simply disallowed newer gcc's.
This fixes the problem instead.
It's been [PR'ed upstream](https://github.com/agordon/fastx_toolkit/pull/22).
Tested with gcc@5.5.0 and gcc@8.2.0 on CentOS.
The most recent release of bamutil that we support uses an embedded
copy of libStatGen that has several issues that keep it from building
with newer releases of gcc.
They've all been fixed upstream and the latest release of bamutil
would pick them up if/when we support it. The build process has
changed though, plus my team needs *this* version.
This commit backports those fixes.
- cleanup patch.py:
- make patch.py constructors more understandable
- loosen coupling of patch.py with package
- in Package: make package_dir, module, and namespace class properties
- These were previously instance properties and couldn't be called from
directives, e.g. in patch.create()
- make them class properties so that they can be used in class definition
- also add some instance properties to delegate to class properties so
that prior usage on Package objects still works
* Always build glib with iconv
My early PR, #10165, which added a variant to configure glib to use
libiconv and defaulted to false, seems to be causing more trouble than
the knob is worth.
This changes the glib package to always depend on and use libiconv.
* libiconv depends_on is no longer conditional
