One of the mottoes of the Sage project is to not reinvent the wheel: If an algorithm is already implemented in a well-tested library then consider incorporating that library into Sage. The current list of available packages are the subdirectories of SAGE_ROOT/build/pkgs/. The management of packages is done through a bash script located in SAGE_ROOT/local/bin/sage-spkg. This script is typically invoked by giving the command:
[user@localhost]$ sage -i <options> <package name>...
options can be:
Not all packages are built by default, they are divided into standard, optional and experimental ones. Standard packages are built by default and have much more stringent quality requirements.
The section Directory Structure describes the structure of each individual package in SAGE_ROOT/build/pkgs. In section Manual package build and installation we see how you can install and test a new spkg that you or someone else wrote. Finally, Inclusion Procedure for New and Updated Packages explains how to submit a new package for inclusion in the Sage source code.
Third-party packages in Sage consists of two parts:
As an example, let us consider a hypothetical FoO project. They (upstream) distribute a tarball foo-1.3.tar.gz (that will be automatically placed in SAGE_ROOT/upstream during the installation process). To package it in Sage, we create a subdirectory containing the following:
SAGE_ROOT/build/pkgs/foo |-- patches | |-- bar.patch | `-- baz.patch |-- checksums.ini |-- package-version.txt |-- spkg-check |-- spkg-install |-- spkg-src `-- SPKG.txt
When installing Sage these files are used to patch the tarball and to start the build and install process of the package.
We discuss the individual files in the following.
The spkg-install file is a shell script installing the package, with PACKAGE_NAME replaced by the the package name. In the best case, the upstream project can simply be installed by the usual configure / make / make install steps. In that case, the build script would simply consist of:
#!/usr/bin/env bash cd src ./configure --prefix="$SAGE_LOCAL" --libdir="$SAGE_LOCAL/lib" if [ $? -ne 0 ]; then echo >&2 "Error configuring PACKAGE_NAME." exit 1 fi $MAKE if [ $? -ne 0 ]; then echo >&2 "Error building PACKAGE_NAME." exit 1 fi $MAKE -j1 install if [ $? -ne 0 ]; then echo >&2 "Error installing PACKAGE_NAME." exit 1 fi
Note that the top-level directory inside the tarball is renamed to src before calling the spkg-install script, so you can just use cd src instead of cd foo-1.3.
If there is any meaningful documentation included but not installed by make install, then you can add something like the following to install it:
if [ "$SAGE_SPKG_INSTALL_DOCS" = yes ] ; then $MAKE doc if [ $? -ne 0 ]; then echo >&2 "Error building PACKAGE_NAME docs." exit 1 fi mkdir -p "$SAGE_LOCAL/share/doc/PACKAGE_NAME" cp -R doc/* "$SAGE_ROOT/local/share/doc/PACKAGE_NAME" fi
The spkg-check file is an optional, but highly recommended, script to run self-tests of the package. It is run after building and installing if the SAGE_CHECK environment variable is set, see the Sage installation guide. Ideally, upstream has some sort of tests suite that can be run with the standard make check target. In that case, the spkg-check script would simply contain:
#!/usr/bin/env bash cd src $MAKE check
The package-version.txt file containts just the version. So if upstream is foo-1.3.tar.gz then the package version file would only contain 1.3.
If the upstream package is taken from some revision other than a stable version, you should use the date at which the revision is made, e.g. the Singular package 20090818 is made with the revision as of 2009-08-18.
If you made any changes to the upstream tarball (see Directory Structure for allowable changes) then you should append a .p1 to the version. If you make further changes, increase the patch level as necessary. So the different versions would be 1.3, 1.3.p1, 1.3.p2, ...
The SPKG.txt file should follow this pattern:
= PACKAGE_NAME = == Description == What does the package do? == License == What is the license? If non-standard, is it GPLv3+ compatible? == SPKG Maintainers == * Mary Smith * Bill Jones * Leonhard Euler == Upstream Contact == Provide information for upstream contact. == Dependencies == Put a bulleted list of dependencies here: * python * readline == Special Update/Build Instructions == List patches that need to be applied and what they do. If the tarball was modified by hand and not via a spkg-src script, describe what was changed.
with PACKAGE_NAME replaced by the the package name. Legacy SPKG.txt files have an additional changelog section, but this information is now kept in the git repository.
Actual changes to the source code must be via patches, which should be placed in the patches directory. GNU patch is distributed with Sage, so you can rely on it being available. All patches must be documented in SPKG.txt, i.e. what they do, if they are platform specific, if they should be pushed upstream, etc.
Patches to files in src/ need to be applied in spkg-install, that is, if there are any patches then your spkg-install script should contain a section like this:
for patch in ../patches/*.patch; do [ -r "$patch" ] || continue # Skip non-existing or non-readable patches patch -p1 <"$patch" if [ $? -ne 0 ]; then echo >&2 "Error applying '$patch'" exit 1 fi done
which applies the patches to the sources.
A special case where no patch would be necessary is when an author provides an already fine SPKG on the net which includes all files needed for SAGE_ROOT/build/pkgs/foo and the source in its src/ subdirectory. Here it suffices to put the web link to the package into the ticket.
The spkg-src file is optional and only to document how the upstream tarball was changed. Ideally it is not modified, then there would be no spkg-src file present either.
However, if you really must modify the upstream tarball then it is recommended that you write a script, called spkg-src, that makes the changes. This not only serves as documentation but also makes it easier to apply the same modifications to future versions.
The checksums.ini file contains checksums of the upstream tarball. It is autogenerated, so you just have to place the upstream tarball in the SAGE_ROOT/upstream/ directory and run:
[user@localhost]$ sage -sh sage-fix-pkg-checksums
At this stage you have a new tarball that is not yet distributed with Sage (foo-1.3.tar.gz in the example of section Directory Structure). Now you need to manually place it in the SAGE_ROOT/upstream/ directory. Then you can run the installation via:
[user@localhost]$ sage -i package_name
[user@localhost]$ sage -i -f package_name
to force a reinstallation. If your package contains a spkg-check script (see Self-Tests) it can be run with:
[user@localhost]$ sage -i -c package_name
If all went fine, open a ticket, put a link to the original tarball in the ticket and upload a branch with the code under SAGE_ROOT/build/pkgs.
Packages that are not part of Sage will first become optional or experimental (the latter if they will not build on all supported systems). After they have been in optional for some time without problems they can be proposed to be included as standard packages in Sage.
To propose a package for optional/experimental inclusion please open a trac ticket with the respective Component: field set to either packages:experimental or packages:optional. The associated code requirements are described in the following sections.
After the ticket was reviewed and included, optional packages stay in that status for at least a year, after which they can be proposed to be included as standard packages in Sage. For this a trac ticket is opened with the Component: field set to packages:standard. Note that the script in SAGE_ROOT/build/deps is called when building Sage so please include the build command for your standard package there. Then make a proposal in the Google Group sage-devel.
Upgrading packages to new upstream versions or with additional patches includes opening a ticket in the respective category too, as described above.
If you are patching a standard Sage spkg, then you should make sure that the license information for that package is up-to-date, both in its SPKG.txt file and in the file SAGE_ROOT/COPYING.txt. For example, if you are producing an spkg which upgrades the vanilla source to a new version, check whether the license changed between versions.
For a package to become part of Sage’s standard distribution, it must meet the following requirements:
License. For standard packages, the license must be compatible with the GNU General Public License, version 3. The Free Software Foundation maintains a long list of licenses and comments about them.
Build Support. The code must build on all the fully supported platforms.
A standard package should also work on all the platforms where Sage is expected to work and on which Sage almost works but since we don’t fully support these platforms and often lack the resources to test on them, you are not expected to confirm your packages works on those platforms.
Quality. The code should be “better” than any other available code (that passes the two above criteria), and the authors need to justify this. The comparison should be made to both Python and other software. Criteria in passing the quality test include:
Previously an optional package. A new standard package must have spent some time as an optional package. Or have a good reason why this is not possible.
Refereeing. The code must be refereed, as discussed in The Sage Trac Server.