- migrate to c++14 for most compilers *except* gcc.
There are still many systems in use with gcc-4.8.5, which does not
support c++14.
- initial rules for nvidia compilers (pgi is will soon be defunct).
Not fully tested...
CONFIG: provide fallback value for the user directory name
- in containers may have an unset USER env variable.
Default to 'user' to prevent ugly looking directory names.
- use orte-info to determine prefix/libdir for openmpi.
This removes a run-time dependency on mpicc, which is actually
only needed for building with MPI (not running with MPI).
The corresponding openmpi devel package (deb/rpm) will not
necessarily be installed on a particular system.
- retain mpicc logic if the new logic using orte-info does not
deliver an answer. Final fallback to using 'orterun' to
infer prefix/libdir.
- Additional logic for intel and msmpi to make it easier to
locate these vendor packages within ThirdParty
(ie, under ThirdParty/opt/...)
CONFIG: improve robustness
- add check for absolute path when adding PATH/LD_LIBRARY_PATH etc.
- prefix more variables with '_foam*' to prevent accidental overwrite
of userspace shell variables when sourcing
- largely as per patch from Jong-Gwan (Jason) Do
NB: the intel-one setup adds in paths for intelmpi.
Its mpicc version does not harmonize with the OpenFOAM
system openmpi setup (using mpicc --showme:link).
Needs adjustment, or use intelmpi instead.
- update name mappings for newer gcc, clang versions
- previously had a very old (likely irrelevant) setting for solaris
systems only.
- support site-specific customization.
Eg, using etc/config.{csh,sh}/prefs.fjmpi
- remove erroneous shell redirects present in cshell files
- prefix FOAM_MPI and library directories with 'sys-' for system
versions for uniform identication.
WM_MPLIB | libdir (FOAM_MPI) | old naming |
SYSTEMMPI | sys-mpi | mpi |
SYSTEMOPENMPI | sys-openmpi | openmpi-system |
- prefix preferences with 'prefs.' to make them more easily
identifiable, and update bin/tools/create-mpi-config accordingly
Old name: config.{csh,sh}/openmpi
New name: config.{csh,sh}/prefs.openmpi
- additional mpi preferences now available:
* prefs.intelmpi
* prefs.mpich
...
CONFIG: added hook for EASYBUILDMPI (eb-mpi), somewhat like USERMPI
- EasyBuild uses mpicc when compiling, so no explicit wmake rules are
used
ENH: support different major versions for system openmpi
- for example, with
WM_MPLIB=SYSTEMOPENMPI2
defines FOAM_MPI=sys-openmpi2 and thus creates lib/sys-openmpi2
ENH: centralize handling of mpi as 'mpi-rules'
Before:
sinclude $(GENERAL_RULES)/mplib$(WM_MPLIB)
sinclude $(DEFAULT_RULES)/mplib$(WM_MPLIB)
ifeq (,$(FOAM_MPI_LIBBIN))
FOAM_MPI_LIBBIN := $(FOAM_LIBBIN)/$(FOAM_MPI)
endif
After:
include $(GENERAL_RULES)/mpi-rules
- also allows variants such as SYSTEMOPENMPI2 to be handled separately
- dependency handling relocated from cmakeFunctions to wmakeFunctions
and reused for mpi-versioned builds. This allows more checks for
configuration parameters and removes hard-code build path
information.
CONFIG: remove spurious mplibHPMPI entries
CONFIG: remove ADIOS1 rules (antiquated)
- fix typo in makefiles/info that affected wmake -show-compile-c
- additional safeguard in src/OpenFOAM/Make/options against self-linking.
This is not normally required unless PROJECT_LIBS has been added into
the link stage.
- Note: mpich now builds libmpi.so instead of libmpich.so
- define both -DMPICH_SKIP_MPICXX and -DOMPI_SKIP_MPICXX regardless of
using openmpi or mpich. This simplifies the files and does not harm.
- in addition to managing different vendors and versions, it may also
be necessary or desirable to have a particular variant
(eg, profiling, release, etc).
Devise a new meaningful name for the variant and create a
corresponding wmake rule.
Eg, SYSTEMOPENMPI-profiling with a corresponding
"wmake/rules/linux64Gcc/mplibSYSTEMOPENMPI-profiling" file
that has suitable content for your system.
CONFIG: intel-mpi use intel64/ paths only for config and wmake rules (#1153)
- previously adjusted the config files, but missed the changes
required for the wmake rules too.
Now simply migrate to using "intel64/{include,bin,lib}"
instead of the older naming "{include,bin,lib}64"
These changes work since at least intel-mpi 2015 (5.x), but possibly
earlier as well
- foamCleanPath now only splits the environment variable on ':', which
allows other directories with spaces or '(..)' etc to pass through
without major issue.
- The filter arguments are split on whitespace, colons or semi-colons.
- generalize some of the library extensions (.so vs .dylib).
Provide as wmake 'sysFunctions'
- added note about unsupported/incomplete system support
- centralize detection of ThirdParty packages into wmake/ subdirectory
by providing a series of scripts in the spirit of GNU autoconfig.
For example,
have_boost, have_readline, have_scotch, ...
Each of the `have_<package>` scripts will generally provide the
following type of functions:
have_<package> # detection
no_<package> # reset
echo_<package> # echoing
and the following type of variables:
HAVE_<package> # unset or 'true'
<package>_ARCH_PATH # root for <package>
<package>_INC_DIR # include directory for <package>
<package>_LIB_DIR # library directory for <package>
This simplifies the calling scripts:
if have_metis
then
wmake metisDecomp
fi
As well as reducing clutter in the corresponding Make/options:
EXE_INC = \
-I$(METIS_INC_DIR) \
-I../decompositionMethods/lnInclude
LIB_LIBS = \
-L$(METIS_LIB_DIR) -lmetis
Any additional modifications (platform-specific or for an external build
system) can now be made centrally.
- export library path for gmp/mpfr from CGAL config files.
This is required when non-system gmp/mpfr libraries are being
used, but not using a ThirdParty compiler installation.
- automatically handle lib/ vs lib64/ (eg, for central installations)
for packages such as boost, CGAL, etc. While the ThirdParty
compilation of these will normally land in lib64/, this may not be
the case when they are supplied by another means.
- reworked the handling of foamEtcFile and foamCleanPath for less
clutter in the configuration files.
Added the bin/tools/lib-dir script to handle logic that is
too complex to easily manage in csh.
- add note in BuildIssues about the I_MPI_CC variable, which is needed
when building with Intel-MPI and gcc/clang.
This additional setting is needed since the changes needed to solve
the issue of building scotch with Intel-MPI and icc (issue #434)
means that mpiicc is now being used as the wrapper when compiling
scotch.
- have the FOAM_MPI short name for INTELMPI start with 'impi-' instead
of just the version number.
Intel-MPI is often installed as /opt/intel/impi/4.1.3.049, which
results in 'FOAM_MPI=4.1.3.049' and the mpi flavour is lost.
Prefix these cases with 'impi-'
- suppress error messages that appear with zsh.
According to unset(1p), 'unset -f' unsets a function.
If the function was not previously defined, this is a no-op.
This is similar for zsh, but there it emits a warning if the
function was not previously defined.
- avoid 'local' in functions sources from etc/bashrc.
ksh does not support this.
- use 'command' shell builtin instead of 'type'.
Seems to be more consistent between shell flavours.
- permit SYSTEMMPI user adjustments via
etc/config.{csh,sh}/mpi-system
This can be a convenient place for setting up SYSTEMMPI for OpenFOAM
without adjusting bashrc, prefs.sh ...
- add a USERMPI type. This represents any generic mpi implementation.
The user is responsible for supplying an appropriate
wmake/rules/General/mplibUSERMPI file and managing all settings.
This type of setup can be useful in combination with specific build
systems (SPACK, EASYBUILD, etc) or module systems for which the MPI
variant is part of the installed configuration.
- handle sourcing bashrc with a relative path (issue #383)
- handle sourcing from bash and zsh.
Still need manual intervention when sourcing dash, sh, or ksh.
- replace grep in etc/cshrc with sed only
- logical instead of physical path for WM_PROJECT_DIR (issue #431).
Doesn't seem to be possible for csh/tcsh.
* Continue using physical locations when comparing directories,
but not for the top-level FOAM_INST_DIR, WM_PROJECT_DIR.
- relocate WM_CC, WM_CXX overrides from etc/config.*/compiler
to etc/config.*/settings to ensure that they are left untouched
when etc/config.sh/compiler is sourced while making third-party
packages (eg, gcc, llvm, CGAL).
- provide fallback FOAM_TUTORIALS setting in RunFunctions
STYLE: remove "~OpenFOAM" fallback as being too rare, non-obvious
Can now use this:
_foamSourceEtc config.sh/scotch
_foamSourceEtc config.csh/scotch
instead of this:
_foamSource $($WM_PROJECT_DIR/bin/foamEtcFile config.sh/scotch)
_foamSource `$WM_PROJECT_DIR/bin/foamEtcFile config.csh/scotch`
In the bash/sh version, leave the _foamSource function for now, since
ThirdParty is still relying on it.
STYLE: elminate while-loop for _foamAddPath etc since this type of
construct isn't readily possible for csh and isn't being used anywhere.
STYLE: only use paraview settings when actually available
- this means executing makeParaView prior to building OpenFOAM itself,
but is consistent with the instructions given by makeParaView,
and elminates anticipating the source location from the paraview
config file, which increases the build flexibilty for ThirdParty
* boost 1_62_0 (Sept 2016)
* CGAL 4.9 (Sept 2016)
- now has headers-only mode that could be interesting
* FFTW 3.3.5 (Jul 2016)
* openmpi 1.10.4 (Sept 2016)
- the first openmpi 2.x release is also available, but too early to switch
* paraview 5.2.0 (Nov 2016)
- builds without additional patching
STYLE: removed unneeded CMake environment variables
- "$FOAM_USER_APPBIN" and "$FOAM_USER_LIBBIN" have been added to
"foamOldDirs" in "etc/bashrc" and "etc/config.sh/unset"
- "$OPAL_PREFIX" is now undefined in the option "SYSTEMOPENMPI" within
"etc/config.sh/mpi", but only if the path defined in this variable
is cleaned when using "foamCleanPath".
- "$OPAL_PREFIX" is now also conditionally undefined in
"etc/config.sh/unset" when the path is picked up by "foamCleanPath".
Patch contributed by Bruno Santos
Resolved bug-report http://bugs.openfoam.org/view.php?id=2210
- move unset mpi buffer size from config.csh/settings -> config.csh/mpi
- wmUNSET now also unalias wmREFRESH
Note: unsetenv WM_COMPILER_TYPE in csh variant?
- drop reference to paraview-3 in config files.
Cannot say anything definite about support for versions
this old (2012)
1. "foamCompiler" was changed to a more permanent "WM_COMPILER_TYPE"
environment variable, so that it can be used by 3rd party
installation scripts, such as "makeGcc", "makeLLVM" and so on. More
on this will be provided in issue #1215.
2. The script functions such as "_foamSource()" and "_foamAddPath()"
were moved to a new file "etc/config.sh/functions". It has the
ability to set or unset, depending on whether "WM_BASH_FUNCTIONS" is
defined or not. This allows for these functions to be reused by
other scripts, such as "makeGcc".
3. The script "etc/config.sh/CGAL" relies on whether a local
environment variable "SOURCE_CGAL_VERSIONS_ONLY" is defined or not,
so that it will load only the version settings if it's defined. This
is to make it easier to call this script from "makeCGAL". Although
it still feels a bit of a clunky hack, but I didn't manage to deduce
any other way we could do this :( I didn't add indentation within
the if-block, to make it easier to read the changes. In addition,
the local variable "common_path" is used to shorten the length of
the lines and use slightly less repeated code.
4. Added another new script "etc/config.sh/compiler", which has only
the version numbers for the compilers taken out from the "settings"
file. It currently depends on "WM_COMPILER_TYPE" for setting the
variables, the same way it did with "foamCompiler". This script is
now always sourced from the "settings" file, for the following
reasons:
- "makeGCC" and "makeLLVM" can now take advantage of this script file.
- The example "compiler" script (detailed next) can rely on this
script file and then override parameters on-demand, as well as
allowing for system compilers to have dedicated settings, such as
setting "WM_CC". This is similar to how the example environment
script for "paraview" works.
5. To the script "etc/config.sh/example/compiler" were added a few more examples:
- It now starts with a block where it first loads the default "compiler" script.
- Has a "WM_COMPILER=Gcc48u" case example for when we try to use GCC
4.8 in Ubuntu 15.10. This is just to give the idea that in a
particular system, we might have several system-wide compiler
versions. For example, in Ubuntu 15.10, there is GCC 4.7, 4.8 and
5.2, which could be used for testing performances or compatibility
with some other 3rd party library.
- Has the "WM_COMPILER=Icc" case example, related to the original bug
report, where "WM_CC=icc" and "WM_CXX=icpc", so that the user then
simply copies this file to their own local preferences folder.
6. Small bug fix in "etc/config.sh/mpi", where unsetting "minBufferSize" was missing at the end of the script.
7. Small change in "etc/config.sh/paraview", where "CMAKE_ROOT" is set
along with "CMAKE_HOME". This is due to a rare issue that occurs on
people's systems where they have a custom system-wide CMake version
installed and which is used by having "CMAKE_ROOT" set on that
environment. This can mess up OpenFOAM's custom ParaView builds,
given that conflicting CMake versions can lead to not building
ParaView at all.
- For more details about "CMAKE_ROOT":
https://cmake.org/Wiki/CMake_Useful_Variables [^]
8. The scripts "_foamAddPath _foamAddLib _foamAddMan" were not being
unset at the end of "settings". They are now unset at the end of
"bashrc", through a call to the new double-use "functions" script.
Additionally all references to "foamCompiler" have been changed to
"WM_COMPILER_TYPE".
See also http://www.openfoam.org/mantisbt/view.php?id=1232
