Software provenance for SR - QIICR/ProjectIssuesAndWiki GitHub Wiki

Information currently recorded in TID 1004 Device observer identifying attributes is probably insufficient and not well-structured to facilitate reproducibility of computation. We should improve this based on Slicer experience and DBP requirements.

• come up with a list of missing items that we would like to capture
• formalize by extending TID 1004 (?)

• TID 1004

• what we should probably capture
• version of software, dependency libraries (for Slicer we have trunk version + version of the extension - need to capture both?)
• platform: OS, architecture, compiler, version of compiler (+compiler flags?)
• ...

• Information about the slicer source code
  • git repository and hash uniquely identify the Slicer source code
  • For factory builds downloaded from slicer.org,
  • this uniquely identifies the versions of all dependencies that are built with the superbuild script (VTK, ITK, python, etc)
  • version of Qt should be extracted from the library since it is not built by the superbuild
  • For non-factory builds (custom at a site or per-user) then the _DIR can be specified to select a different build of a library and we should extract the version information accordingly.
• For extensions
  • the repository and hash (or svn revision) of the extension source code
  • we should provide a format where extensions can declare the versions of other dependencies they use (e.g. superbuild and/or git submodules and the like)
• For the build machine we should track
  • version of the OS along with any information available about the patch level
  • processor architecture of the host machine and of the target build
  • memory and other hardware features of the machine
  • version of the build tools used (cmake, Visual studio, Xcode, gcc, etc)
  • versions of any development libraries used (not really applicable for slicer releases except Qt as noted above, but also NSIS package on windows that creates the installer)
  • user name (and real identity?) and environment variables for the account that triggered the build
  • build directory
  • date and time of the build
  • machine name (and IP address? mac address? other identifiers?)
• When a particular piece of data is generated
  • machine info (like above)
  • user info (like above)
  • OS environ info (like above)
  • versions of system libraries installed (msvcrt, libc, etc)
  • reference to experiment protocol under which the data was generated (if applicable)

• PROV Model primer
• W3C Provenance incubator wiki
• Provenance tracking in nipype based on PROV data model:
• https://github.com/nipy/nipype/blob/master/nipype/interfaces/base.py#L947-L970
• NI-DM
• Introduction to NI-DM (slides)