Differences between revisions 31 and 32
Revision 31 as of 2006-08-31 14:26:27
Size: 14441
Comment:
Revision 32 as of 2006-11-05 23:40:15
Size: 14695
Editor: KevinMcCarty
Comment: Update to reflect new upload of ROOT to NEW
Deletions are marked like this. Additions are marked like this.
Line 8: Line 8:
   * Packages uploaded to [http://ftp-master.debian.org/new.html NEW] queue on April 27, 2006, by KevinMcCarty and ChristianHolm. These are targeted for "experimental".    * Packages uploaded to [http://ftp-master.debian.org/new.html NEW] queue on April 27, 2006, by KevinMcCarty and ChristianHolm. These are targeted for "experimental".
   * FTPmaster rejected these packages on the grounds of licensing issues. In the mean time, these issues have been worked out with the upstream authors and 3rd party developers.
   * Since then, the debian/copyright file has been greatly expanded, and the copyright holders of the non-free material were contacted and agreed to relicense it under LGPL. New ROOT packages targeted at "experimental" were uploaded to the NEW queue on November 3, 2006.
Line 14: Line 16:
   * FTPmaster rejected these packages on the grounds of licensing issues. In the mean time, these issues have ben worked out with the upstream authors and 3rd party developers. The fixed packages will be uploaded to NEW soon.
Line 16: Line 17:
   * See [http://mirror.phy.bnl.gov/debian-root/ here] for some build instructions and a repository for unstable (i386, amd64, powerpc) and stable (i386 only).    * See [http://mirror.phy.bnl.gov/debian-root/ the unofficial ROOT deb repository] for some build instructions and a repository for unstable (i386, amd64, powerpc) and stable (i386 only).

1. ROOT for DebianScience

* DebianScience

  • ?DebianSciencePhysics

1.1. Debian (unofficial) packages

1.1.1. Technical issues for packaging

The approach is that the ROOT source tree contains [http://root.cern.ch/viewcvs/build/package/lib/ scripts] and [http://root.cern.ch/viewcvs/build/package/common templates] to deal with packaging in general (that is for Debian and RPM), as well as some [http://root.cern.ch/viewcvs/build/package/debian Debian] and [http://root.cern.ch/viewcvs/build/package/redhat RPM] specific templates.

The debian packaging directory is then build by running the script [http://root.cern.ch/viewcvs/build/package/lib/makedebdir.sh makedebdir.sh], followed by fakeroot debian/rules debian/control. The packaging scripts will try to build as many packages as possible on the build host. Note, that that sometimes means that you can get packages that wouldn't be possible on a pristine build system.

Note, that this scheme does not change the way the packages are built. It merely changes how the debian directory is prepared. After the debian directory is made, the developer can still tweak it to fit the system. Of course, such tweaks should eventually be propegated back to the packaging scripts in ROOT itself, to assure consistency.

* N.B. In the eventual "official" Debian packages to come, the package maintainer would presumably run these two commands himself before dpkg-buildpackage, and the results of the commands (for instance the debian/ directory) would therefore enter into the diff.gz of the source package. Thus the source package uploaded to Debian would always be buildable with the normal steps of dpkg-buildpackage -rfakeroot.

Question: what happens if an end-user does apt-get source root and then re-runs the makedebdir script? Does it overwrite the existing debian directory with information about the dependencies installed on the end-user's system? (Seems to me that would be the desired outcome.)

Answer: Yes.

It is important that the maintainer ensure that he has all the needed "official" build-depends installed on his system when running the makedebdir.sh script, and none of the "unofficial" non-DFSG-free libraries (e.g. Pythia) installed, perhaps by building on a pristine machine or in a pristine chroot. That is, use pbuilder :-). Remember that packages are not allowed to munge debian/control in such a way that their Build-Depends change during the dpkg-buildpackage step!

1.1.2. Out-dated:

ROOT can be patched to use other fonts if needed, at least for most fonts. However, there are no real substitute for the symbol.ttf font used for Greek letters and the like. One could imagine using something like opens.ttf from OpenOffice.org, but it's not quite the same, and it requires major code changes in ROOT. Another way to deal with all this, would be to use the Computer Modern fonts (TeX fonts), which contains all the needed symbols, though scattered over many font files.

""Update to the above"": Indeed the upstream sources has been patched to be able to use [http://packages.debian.org/unstable/x11/ttf-freefont ?FreeFont] instead of the M$ Core Fonts. Also, a GPL'ed Postscript font was found to be able to substitute the M$ Core Fonts "symbol.ttf". This was converted to a TTF (using http://packages.debian.org/unstable/x11/fontforge ?FontForge]) and submitted upstream. Hence, there are now free alternatives to all fonts for ROOT.

1.2. Licensing

1.3. Obsolete

1.3.1. Repos, etc.

1.3.2. Differences between Ricardo Yanez and ChristianHolm work.

  • Package build infrastructure
    • Ricardo does a traditional packing.
      • That is, he maintains a patch to the ROOT source tree, with static packaging files, including the file lists, control file, and so on. This is attractive in it is completely standard. However, it has the down-side, that changes in the ROOT source tree has to be dealt with explicitly.
    • Christian does a somewhat non-traditional thing.
      • He managed to convince the upstream maintainers of ROOT, that they should include packaging stuff directly in the ROOT source tree. This was done to avoid the hurdle of updating the packaging scripts when ever the ROOT source tree changed.

        Therefore, in Christians approach, the ROOT source tree contains [http://root.cern.ch/viewcvs/build/package/lib/ scripts] and [http://root.cern.ch/viewcvs/build/package/common templates] to deal with packaging in general (that is for Debian and RPM), as well as some [http://root.cern.ch/viewcvs/build/package/debian Debian] and [http://root.cern.ch/viewcvs/build/package/redhat RPM] specific templates.

        The debian packaging directory is then build by running the script [http://root.cern.ch/viewcvs/build/package/lib/makedebdir.sh makedebdir.sh], followed by fakeroot debian/rules debian/control. The packaging scripts will try to build as many packages as possible on the build host. Note, that that sometimes means that you can get packages that wouldn't be possible on a pristine build system. Note, that this scheme does not change the way the packages are built. It merely changes how the debian directory is prepared. After the debian directory is made, the developer can still tweak it to fit the system. Of course, such tweaks should eventually be propegated back to the packaging scripts in ROOT itself, to assure consistency.

        • N.B. In the eventual "official" Debian packages to come, the package maintainer would presumably run these two commands himself before dpkg-buildpackage, and the results of the commands (for instance the debian/ directory) would therefore enter into the diff.gz of the source package. Thus the source package uploaded to Debian would always be buildable with the normal steps of dpkg-buildpackage -rfakeroot. Question: what happens if an end-user does apt-get source root and then re-runs the makedebdir script? Does it overwrite the existing debian directory with information about the dependencies installed on the end-user's system? (Seems to me that would be the desired outcome.) It is important that the maintainer ensure that he has all the needed "official" build-depends installed on his system when running the makedebdir.sh script, and none of the "unofficial" non-DFSG-free libraries (e.g. Pythia) installed, perhaps by building on a pristine machine or in a pristine chroot. Remember that packages are not allowed to munge debian/control in such a way that their Build-Depends change during the dpkg-buildpackage step!

          As a side note, it is also important to ensure that the debian/changelog file is preserved when moving to new upstream versions; without sufficient caution, it looks like it might accidentally be overwritten in such a case. --KevinMcCarty

  • Plug-ins and extensions
    • Ricardo makes separate run-time and development packages for all plug-ins and extensions. That is, the plugin mysql has packages libroot-mysql<version> and libroot-mysql-dev, and the extesion roofit has the packages libroot-roofit<version> and libroot-roofit-dev.

      • For extension packages, this is the right thing to do, because users might actually link explicitly against libRooFit.

        For plug-ins this approach is questionable. ROOT normally loads these plugins in response to implicit requests by the user. For example, if a user requests a connection to a MySQL database, she does TSQLServer* serv = TSQLServer::Connect("mysql://...."), and ROOT loads the plugin module libMySQL by it self in the background. Hence, the user needn't know about the library libMySQL or any of the header files that goes with it. Hence, the packaging stuff should not include the link library libMySQL or any of its headers.

    • For plug-ins and extensions, Christian makes only one package, e.g. root-plugin-mysql and root-plugin-roofit. For plug-ins, this is exactly the right thing to do (except that the headers are still distributed) as discussed above. For extensions this is wrong, as discussed above.

    • For extensions (things that the end-user should be able to link against explicitly in her code, as with g++ myprogram.cc -L/usr/lib/root -lRooFit), the Debian library packaging guidelines must be respected: packages of the extension with different sonames must have different package names and be installable in parallel; the -dev package of the extension contains header files and the development symlink (e.g.) /usr/lib/root/libRooFit.so. For plug-ins, it should be sufficient that they can't be dlopened by the wrong version of ROOT. I think the latter is already OK since ROOT 5.08 will read /etc/root/system.rootrc, find out that its plugins are in /usr/lib/root/5.08/ and therefore not see version 5.06 of libMinuit.so which is in /usr/lib/root/5.06/. --KevinMcCarty

  • Merging the two approaches.
    • In my (Christian) mind, we'd do best with a single approach. The current situation with two packaging schemes is not satisfactory, and is confusing at best. I think we need to find a way to unify the two approaches. In my mind that, that amounts to making the right kind of extension and plug-in packages. That is, a a run-time and a development package for each extension, and only a run-time package for the plug-ins. I think the best way to go about it, is to make changes to my approach (not because it's mine, but because it's simpler, and ROOT is willing to support it to some extent), by adding new development packages for the plug-in packages.

1.4. Thanks