* Add new package for PaGMO/PyGMO
* Rename py-pygmo to pagmo
* Fix concretization bug
* Correct Python detection, use correct flags
* Install PyGMO to the pagmo prefix, not the Python prefix
This commits address build issues reported in issue #3268.
In particular, makes cleaner the dependency on parallel-netcdf
when +ncmpi variant is requested and adds CFLAGS=-D H5_USE_16_API
to handle the fact that Ior uses old 1.6 style APIs when +hdf5 is
requested.
These changes make Ior build fine and create modules for HDF5 versions
1.8.x and 1.10.0.
Even though build is successfull the modules haven't been tested in a
run.
* Make libgcc_s relocatable
* spack specifies full path to lib64
* don't need trailing /
* setting rpath is different on macos
* add -headerpad_max_install_names linker options
* formatting fixes
* Add Adam's recommendation
* add various variants to vim
* default to --enable-fail-if-missing
* vim package: -{python,ruby,lua,perl} sets build option to `no`
* vim's interpreter build options are `--enable-XXXXinterp={yes,no,dynamic}`
* dynamic seems to work only on windows builds
* `=no` is the default => use it in spack if the variant is disabled
- Added a new interface for Specs to pass build information
- Calls forwarded from Spec to Package are now explicit
- Added descriptor within Spec to manage forwarding
- Added state in Spec to maintain query information
- Modified a few packages (the one involved in spack install pexsi) to showcase changes
- This uses an object wrapper to `spec` to implement the `libs` sub-calls.
- wrapper is returned from `__getitem__` only if spec is concrete
- allows packagers to access build information easily
Declare that (i) Trilinos can be only built against 32bit Hypre
(ii) SLEPc can not be built with Arpack when 64bit indices are used
(iii) reflect those constraints in deal.II's depends_on
While there, add extra release flags for best performance.
* Updating SAMRAI package file with new urls and new versions. Also using Spack's MPI compilers.
* Using default optimization for Samrai and setting type to build for m4 dependency.
* Adding debug variant to Samrai.
* Converting Samrai to AutotoolsPackage.
* Add git-review
* Fix flake8 errors
* use PythonPackage for git-review
* added dependencies for git-review
* fix flake8
* moved url's to pypi.io based urls
* fixed flake8... again
* mostly url fixes, not tested, will do tomorrow
* flake8 fix
* changed URL for py-pysocks
* Fixed URL for pysocks
* fixed url
The new repository for the development version points to the most
up-to-date official source for Adol-C. The previous SVN repository was
only occasionally sync'd with this repository.
* Fix build issue #3190 on bg-q :
- disable parallel build as it produces weired linker errors
in cross compiling environment
- update ldflags for bg-q as per Makefile provided Makefile.inc.ppca2_ibm_bgq
* Added note about -lpthread vs -pthread
* Corrected extrae package homepage and url
* Corrected paraver package homepage and url
* To fix 'undefined reference to libintl_dgettext' appearing in extrae package when using binutils+libiberty
* Adding support to previous paraver version and reverting binutils changes to open another PR
* Add installer for petsc4py and slepc4py
* @adamjstewart requested changes on this pull request; make these type=('build', 'run')
* @citibeth suggestion Remove unneeded indirect dependencies
Zoltan expects mpi libraries with option --with-mpi-libs.
If this option is not provided then it uses '-lmpi' which
may not be correct. To avoid this, we have to explicitly
pass empty string.
This adds notes to the source explaining the change.
* Add log4cxx package
* add missing patch files to log4cxx
* remove unnecessary dependencies
* this fails at 'make check'
* this is the AutotoolsPackage version
* fixed install bug
* Update the krell institute products to use the latest features of spack for building on cluster platforms.
* Address travis error messages and resubmit the pull request.
* Update the contents of openspeedshop package.py so it passes the flake8 tests.
* Fix flake8 error-whitespack issue in mrnet package.py file.
* Add updates based on spack reviewer feedback.
* More fixes based on comments from reviewers. Switch using extend to using append, remove additional setting of PATH and LD_LIBRARY_PATH that should not be required due to RPATH.
* More review related changes. Update MPIOption.append lines and take out xercesc references.
* Create a base options function for common openspeedshop base cmake options to reduce redundencies.
* Add libxml2+python depends on to get around issues with the libxml2 package file.
* Using boost over 1.60.0 causes compile errors. This is a known boost bug. Also, dyninst-9.2.0 is set to be the vesrion of dyninst to use with OSS, as of now. The newer version fails to build.
* Fix bad syntax in specifying the boost version range.
* Update the version numbers for the krell institute components and tools: cbtf and openspeedshop.
* Do not build glib for qt3, it is not needed and causes build problems at this time anyway.
* A fix was added for setting LD_LIBRARY_PATH in the qt3 build, but if LD_LIBRARY_PATH is not set the qt build fails. So so check and set LD_LIBRARY_PATH if not set, update if it is set.
* Update the fix for qt3 build by setting LD_LIBRARY_PATH instead of checking for whether it is set or not per Adams comment that spack clears LD_LIBRARY_PATH.
* A fix was added for setting LD_LIBRARY_PATH in the qt3 build, but if LD_LIBRARY_PATH is not set the qt build fails. So so check and set LD_LIBRARY_PATH if not set, update if it is set.
* Trim comments to fit more concisely.
* Fix tabs versus spaces and swap if and else clause check from a negative to a positive check.
Update package file for netlib-lapack to work with IBM XL compiler
The flag -qzerosize is redundant, as the IBM XL compiler should set this flag
by default. However, at this time the default flag appears to be ignore.
Hence, I am setting it in the package file as well, as it has no negative
effect if set twice.
graphviz:
* Download from Fedora projet, as main graphviz site not working.
* Disable java because Spack does not yet support Java, and the system might not have it installed.
* Added all language binding variants; disabled enough in the default configuration to avoid dependencies.
* Removed alternate download location (turned into comments).
* Turn off all language bindings by default.
* Raise an exception on bindings that have not been verified to work.
* Added text indicating what works and doesn't work when user runs `spack info`.
* New package.py for ExM C-Utils: An initial package dependency for Swift/T
* New package.py for ADLB/X: A 2nd package dependency for Swift/T
* New package.py for Turbine
* New package.py for STC
Spack wants URL info even for external packages. Without it, I get
the following error:
NoURLError: Package SpectrumMpi has no version with a URL.
File "/home_local/serbanspack/spack/lib/spack/spack/repository.py", line 580, in get
self._instances[key] = package_class(copy)
File "/home_local/serbanspack/spack/lib/spack/spack/package.py", line 562, in __init__
f = fs.for_package_version(self, self.version)
File "/home_local/serbanspack/spack/lib/spack/spack/fetch_strategy.py", line 878, in for_package_version
url = pkg.url_for_version(version)
File "/home_local/serbanspack/spack/lib/spack/spack/package.py", line 682, in url_for_version
raise NoURLError(cls)
* Creating a spack package for LLNL's LBANN (Livermore Big
Artificial Neural Network) training toolkit.
* Recipe for building LBANN toolkit. Contains limited feature set and
is optimized for building with GNU gcc and OpenBLAS.
* Removed unnecessary dependencies based on reviewers feedback.
* Added support for the int64 data type in the Elemental library. This
is required for supporting indices for large matrices.
* Added a variant to force a sequential weight matrix initialization.
This is slow, but provides an initialization that is independent of
model parallelism.
* Added a guard to prevent building Elemental with the Intel compiler
for versions that have known bugs.
* New package.py for ExM C-Utils: An initial package dependency for Swift/T
* New package turbine
* Fix package.py as requested by @adamjstewart
* New package.py for ADLB/X: A 2nd package dependency for Swift/T
* Add latest version of GNU Parallel (#3106)
* Address formatting guidelines from @adamjstewart
* WIP on new Turbine package.py
* Formatting fixes
* Complete Turbine package.py
[The fix](https://github.com/golang/go/issues/17986) for the small buglet addressed by `misc-cgo-testcshared.patch` has been merged into the tree a while back. I was surprised to see that it wasn't in 1.7.5 and did a bit of digging. It is *has not* been merged into the 1.7 branch but it *has* been merged into 1.8 (and therefor the patch will no longer be necessary).
Figured I'd document my digging for the next person to come along.
* spectrum-mpi: Add new package file for external package
IBM Spectrum MPI is a commercial implementation of MPI based on
OpenMPI. It is usually install in /opt/ibm/spectrum_mpi.
Users need to add the Spectrum MPI package in their packages.yaml
file as follows:
packages:
spectrum-mpi:
version: ['10.1.0.2']
paths:
spectrum-mpi@10.1.0.2: /opt/ibm/spectrum_mpi/
buildable: False
all:
providers:
mpi: [spectrum-mpi@10.1.0.2]
* spectrum-mpi: Added license files and removed the versions
No need for versions as the package is external.
* spectrum-mpi: Remove extraneous defines
Keep only the defines that other MPI implementation define in their
package file:
self.spec.mpicc
self.spec.mpicxx
self.spec.mpif77
self.spec.mpifc
Update the go package to v1.7.5.
- This release no longer needs the time-test patch (it's been merged upstream).
- This release still seems to need the cgo-testcshared patch.
- Also add a comment about environment set up that I need to build it successfully on a very large server.
Tested on CentOS 7.
The repository used by the texlive installer defaults to a redirector
that sends one off to various URLs depending on <fill in the blank>.
This is problematic because all of the mirrors do not update in
synchrony and bad tarballs often hang around for a while.
This leads to problems that are particularly hard to diagnose because
you're likely to end up using a different repo the next time around.
This commit constraints the package to a particular, mainstream,
repository. It's not fast, but it's consistent and usually correct.
It also updates the installer digest, because no day is complete
without updating it at least (sigh) once.
* Different versions of cmake need diff vers of openssl. See Issue https://github.com/LLNL/spack/issues/2990 for background.
Versions of cmake through 3.6.9 seem to need OpenSSL up to 1.0.99. Later
versions can use the current release (thanks to @citibeth for
[digging up details](https://cmake.org/pipermail/cmake/2016-November/064631.html)).
@davydden suggested this change.
Without it I *am not* able to build `cmake@3.6.1` on CentOS 7 and I *am*
able to build `cmake@3.7.2`.
Tested with `cmake@3.7.2` and `cmake@3.6.1` on CentOS 7.
With this change I am able to build both `cmake@3.6.1` and `cmake@3.7.2`
on CentOS 7.
There was a new release of Ant (1.9.8) which led to the v1.9.7
tarball disappearing.
This changes the URL to Ant's archive dir, which seems to contain
*everything* including the two current releases (1.9.8 and 1.10.0)
It adds a digest for 1.9.8.
It adds and comments out a digest for 1.10.0 (which requires Java 8),
as I have not tested it.
We just released 2.0.2 yesterday, so add in that release for
spack. Don't need the PMI patch for this release.
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
cran.r-project.org runs on a single old-school server in Austria
and could potentially be overloaded if "everyone" used it.
cloud.r-project.org is a cloud-based repository that "automatic redirection to servers worldwide [...]", cf. https://cran.r-project.org/mirrors.html.
I assume, that cloud.* can be scale up as needed. Out of the official CRAN mirror, this should be the safest one to pick if a static CRAN mirror is needed.
* Add several new R packages
* Add a few more R packages
* Update more versions
* Convert Package to RPackage
* Add a few more packages
* Add missing dependencies
* AutotoolsPackage: added configure_directory to permit build out of source. The configure script executable is now invoked with an absolute path. Modified a few packages accordingly.
* build_systems: functions returning directories are now properties
* build_systems: fixed issues with tcl and tk
* AutotoolsPackage: reworked recipe for autoreconf
* Spec.satisfies accesses Spec.concrete as property
Fixes#2760
When copying a spec, _concrete is always set to False for each
dependency. "Spec.satisfies" was accessing the member "_concrete"
directly instead of using the property "concrete". This means that
if you copy a spec, the dependencies will be considered equal, but
did not necessarily satisfy one another. Spec.satisfies is a
prerequisite for a package to be considered an extension; as a
consequence, an extension with run-time dependencies that were also
extensions did not activate those extensions. This updates
Spec.satisfies to avoid checking the cached member "_concrete"
directly.
* Added test to check for activation of dependency extension
* Added test to check for transitive satisfiability between a spec and its copy
* Update libquo's homepage, url, and description.
Recently LANL moved to lanl (from losalamos).
* Fix formatting.
Conform to style guidelines.
* More style updates.
Doesn't appear to be necessary, but for good measure.
* oce: fix fetching
apparently the fetcher is confused when there is a mixture of 3-digits
versions and 2-digits versions, i.e. 0.18 and 0.17.2
* flake8
* Updating udunits.
Udunits 2.2.21 is no longer on the unidata ftp site.
The latest is 2.2.23, adding that and it's md5sum.
* Updating udunits2 download URL.
The Unidata ftp site does not keep previous versions of udunits.
However all the tagged versions are in their github.com repository.
Updating the URL to use github.
* Updating udunits2 with a autoreconf def.
The udunits packages on github do not contain a configure script.
One has to run autoreconf to generate it, so adding that to
package file.
Also updated to the latest version and all md5 checksums.
* PackageMeta: `run_before` is an alias of `precondition`, `run_after` an alias of `sanity_check`
* PackageMeta: removed `precondition` and `sanity_check`
* PackageMeta: decorators are now free-standing
* package: modified/added docstrings. Fixed the semantics of `on_package_attributes`.
* package: added unit test assertion as side effects of install
* build_systems: factored build-time test running into base class
* r: updated decorators in package.py
* docs: updated decorator names
* Add dependency on perl
The build process uses perl and also needs `Test::More`.
Some distros, e.g. CentOS, break the core Perl distribution
into separate packages, so it's possible to "have perl" but
not have all the bits one needs to build OpenSSL.
We'll just install one of ours, which comes with all of its
factory parts included.
* Remove uninformative comment
> # Also requires make
doesn't really add any value...
* Fix configure's zlib version check
R wants a version of zlib that is 1.2.5 or newer.
The version checking code just does a lexicographic comparison
of the first 5 characters of the string, so it seesthat the latest
zlib version, 1.2.10, as 1.2.1 and fails.
This patch changes the comparison to use zlibs' hex ZLIB_VERNUM
so that it does not suffer from this problem.
A version of this patch is wending it's way through the R comunity
community and will/should be included in a future release.
I tested the patch with the current R, 3.3.1.
* Tighten zlib dependency version (>= 1.2.5)
* Convert patch to level=1 format.
* libmonitor does not exist at
http://libmonitor.googlecode.com/svn/trunk/
Change location to HPCToolkit version at github. Specify the hash corresponding
to the 20130218 version.
* When xcb is version 1.11, patch the configure file so it doesn't trip on
pthread-stubs and xau.
* Add os.getcwd()/lib to LD_LIBRARY_PATH, instead of nuking
existing LD_LIBRARY_PATH.
* cfitsio: Allow building as shared libraries
- Allow building as shared libraries
- Also add optional bzip2 variant
- Also update to newest version
This requires the recent correction to “fix_darwin_install_name”.
* cfitsio: Simplify code, default to +shared
* Adding 'self.' to the spec call for configure_args
* Updating this to set spec to self.spec
* More updates to the configure_args spec calls
* Another spec issue
* Another spec issue.
* And another spec issue
* Switching cmor back to self.spec.
- Add a PythonPackage class with build system support.
- Support build phases in PythonPackage
- Add a custom sanity check for PythonPackages
- Get rid of nolink dependencies in python packages
- Update spack create to use new PythonPackage class
- Port most of Python packages to new PythonPackage class
- Conducted a massive install and activate of Python packages.
- Fixed bugs introduced by install and activate.
- Update API docs on PythonPackage
* Set MPI[lang] env var to compiler wrappers on cray
Sets MPICC, MPICXX, MPIFC etc to point to the compiler wrappers on Cray systems.
Some packages look for MPIs via environment variables - e.g adios
* Use spec to check for Cray
* os is no longer used.
* petsc: add 64bit variant
* hypre: add int64 variant
* superlu-dist: add int64 variant
* petsc: add int64 variant
* metis: rename idx64 to int64 to make it consistent with other packages
* mumps: rename idx64 to int64 to make it consistent with other packages
* dealii: rename 64bit to int64 to make it consistent with other packages
added latest version of isl. Needed to get llvm/clang in the latest
trunk to build via spack, but there are more versions available.
what is the policy regarding those? Add what we need for now?
* Updating BoxLib with the ability to build again.
* Using Spack MPI compilers directly.
* Fixing typo.
* Using tarball from github archive for boxlib.
* Added customization for make targets in 'build' and 'install' phases for CMakePackage
* Use rst in build system docs so that Sphinx generates nice API docs
* Allow AutotoolsPackages to be built in a different directory
* Flake8
* Fix missing import
* Allow configure to be located in different directory
* Update espressopp to use build targets
* Flake8
* Sphinx fix, lists must be a new paragraph
* Back out change that allowed a configure script in a different directory than build_directory
* Add missing deps, build in parallel
* Missing space for rst list
Although it hurts a little, officially pre-compiled headers in
boost are only supported for gcc & msvc and the latest clang
releases still fail to build boost with it.
Therefore, I disabled building those to get boost build with
clang 3.9.0 on an Ubuntu 14.04 (x86).
Links to documentation and boost bug reports are inline, so
people can later on check if they still apply. Seems just to
be a bug in `Boost.Build` that tries to set `-o` with multiple
output files.
* Added support for Octopus 6.0 and the optional libraries: metis, parmetis, netcdf, arpack-ng, and scalapack.
* Fix PEP 8 line too long error.
* Changed format syntax {} to {0} to support Python 2.6.
* Adding the py-ipdb package for more stable debugging with iPython.
* Removing deprecated python support, adding missing dependencies.
* Adding version checks for (i)python
* Update the krell institute products to use the latest features of spack for building on cluster platforms.
* Address travis error messages and resubmit the pull request.
* Update the contents of openspeedshop package.py so it passes the flake8 tests.
* Fix flake8 error-whitespack issue in mrnet package.py file.
* Add updates based on spack reviewer feedback.
* More fixes based on comments from reviewers. Switch using extend to using append, remove additional setting of PATH and LD_LIBRARY_PATH that should not be required due to RPATH.
* More review related changes. Update MPIOption.append lines and take out xercesc references.
* Create a base options function for common openspeedshop base cmake options to reduce redundencies.
* Add libxml2+python depends on to get around issues with the libxml2 package file.
* Using boost over 1.60.0 causes compile errors. This is a known boost bug. Also, dyninst-9.2.0 is set to be the vesrion of dyninst to use with OSS, as of now. The newer version fails to build.
* Fix bad syntax in specifying the boost version range.
* Update the version numbers for the krell institute components and tools: cbtf and openspeedshop.
* Do not build glib for qt3, it is not needed and causes build problems at this time anyway.
* Removing the nobuild, nolink, and alldeps dependency types in favor of being explicit.
* This will help with maintenance going forward, as adding more dependency types won't affect existing declared dependencies in weird ways.
* default deptype is still `('build', 'link')`
* Remove +mpi variant from NCO
* Update NCO deps to AutotoolsPackage
* Update the other NetCDF packages to AutotoolsPackage
* ANTLR has a nolink deptype on java, don't worry about csharp
* Remove restrictions from NCO package, fix homepage
* Add version 4.6.3 of NCO
* Missed AutotoolsPackage for NetCDF-CXX4
* NetCDF tests fail when run in parallel
* Remove commented out dependency
* Add documentation dependency
* espressopp: package for the ESPResSo++ software
This commit adds a package for the ESPResSo++
simulation software.
* Move Espressopp package to CMakePackage
This commit moves Espressopp package to CMakePackage.
Addresses some comments in the PR.
* Remove run_tests and changed type of dependencies
Addressed some comments from @adamjstewart.
Removed global run_tests setting and removed type from many
dependencies. Set type for py-mpi4py to 'nolink'.
* Updated to last version and fix mpi4py dependency
Added latest espressopp version and changed mpi4py (version) dependency for
different espressopp versions.
* Changed mpi4py version requirements for espressopp
Changed py-mpi4py version requirements for espressopp@1.9.4 according
to the package build requirements for that version.
Adds the following packages:
node-js py-backports-abc py-functools32 py-pycurl py-vcversione
npm py-certifi py-jsonschema py-tornado py-zmq
- Added python packages that are dependencies for the Jupyter suite
- Update new python packaages to use extension package install function.
- Added npm and node-js packages
* Rename packages
* Upcasing depends_on() in packages.
* Downcased extends('r')
* Fixed erroneously changed URL that had slipped through.
* Fixed typo
* Fixed link from documentation into package source code.
* Fixed another doc problem.
* Changed underscores to dashes in package names.
* Added test to enforce lowercase, no-underscore naming convention.
* Fix r-xgboost
* Downcase more instances of 'R' in package auto-creation.
* Fix test.
* Converted unit test packages to use dashes not underscores
* Downcase `r` in the docs.
* Update module_file_support.rst
Fix r->R for class R.
* Use zlib's "fossil" site for old tarballs
Following citibeth's suggestion in #2732, use zlib's "fossil"
site (not to be confused with the sqlite team's VCS...) for retrieving
old tarballs.
Digests for 1.2.{8,10} match and both install for me on CentOS 7.
* Use zlib's "fossil" URL as the one true URL
Everything seems to be available at zlib's "fossil" URL, so just use
it as the one and only url.
(and fix a flake8 complaint about a comment)
* add package for conduit
* try to fix main conduit docstring
* use join_path instead of pjoin
* address a few requests in pr #2670
change name of 'github-master' to 'master'
change 'docs' variant to 'doc', set default to False
remove explicit +shared variant spec for silo and hdf5 deps
(in the conduit +shared case) cases since they default to True
add reference to static rpath issue
(https://github.com/LLNL/spack/issues/2658)
* address pr #2670 requests
add todos and more info on why variants for deps where selected
use python module install python to enable spack activate
use .format instead of %s
* zlib@1.2.10 and R do not get along, work around it
R's configure script has trouble with version numbers > 1.2.9.
This works around it by constraining R to 1.2.8.
I'm working with some R folk on getting it fixed going forward.
* Add adamjstewart's change.
* Update to latest zlib version, server no longer provides older version
Funded-by: IDEAS
Project: IDEAS/xSDK
* Add alternative URL for previous release of zlib
* inheritance of directives: using meta-classes to inject attributes coming from directives into packages + lazy directives
* _dep_types -> dependency_types
* using a meta-class to inject directives into packages
* directives are lazy
fixes#2466
* directives.py: allows for multiple inheritance. Added blank lines as suggested by @tgamblin
* directives.py: added a test for simple inheritance of directives
* Minor improvement requested by @tgamblin
CMakePackage: importing names from spack.directives
directives: wrap __new__ to respect pep8
* Refactoring requested by @tgamblin
directives: removed global variables in favor of class variables. Simplified the interface for directives (they return a callable on a package or a list of them).
The upstream luafilesystem tarball/version had a wayward/inconsistent
underscore in their more recent version tag. The played badly with
our package fetching machinery (due to recent changes?).
Upstream cleaned up their bit which required some touchups here.
- updated the url
- updated the version (digest)
- updated the format statement for the path to the rockspec.
* Adds catch, cppunit, spdlog, tinyxml(1 and 2), google benchmark
Tinyxml comes in two flavors, 1 and 2. Each comes in several
versions... So they cannot be easily united into a single package.
* Use CMakePackage and friends, add copyright
Also eleminate debug/release variants, since it no longuer fits in the
CMakePackage format.
* Remove unnecessary url
* spdlog now has tagged releases
* Remove unnecessary url argument
* Fewer quotes in cmake args, because magic
* Incorrect base class for tinyxml
* Ensure that every package has a license
Also fixes URLs with http://http:// doubled.
This is a continuation of #2656.
* Add license to every file in Spack
* Make sure Todd is the author of all packages
* Fix flake8 tests
* Don't license external Sphinx docs
* Don't display licenses in tutorial example packages
Also fixes typos and converts command-line examples
from tcsh to bash, which is more common
* Added uuid: OSSP uuid is a ISO-C:1999 application programming interface
* Fixed install error in cmor package
* Added uuid: OSSP uuid is a ISO-C:1999 application programming interface
* Fixed install error in cmor package
* Modifiying fontconfig file to allow install in OpenSuse 13.2
* Adding pkg-config dependency to freetype and libxml2
* Removed first possible solution. With @adamjstewart selecting the one adding pkg-config to dependencies
* Update digest for vim@8.0 tarball.
The digest appears to have changed.
The new digest value matches the value in their
[MD5SUMS](ftp://ftp.vim.org/pub/vim/unix/MD5SUMS)
file.
* Change ftp.vim.org -> github.com/vim/vim/archive
Vim seems to do weekly releases, but the name of the tarball on
ftp.vim.org doesn't change so we have to regularly play 'update the
digest' (aka Internet whack-a-mole).
This commit changes the url so that we are now downloading particular
versions from the Vim project's github archive.
I didn't walk back through of time to get all of the versions that
used to be explicit. I grabbed the final `7.4` and the current `8.0`
releases. If people need more we can add them.
