Review and Potentially Overhaul the Dependency Framework

[raphaelcohn]

The current dependency framework works well enough, but has the reached the limitations of its design. Some of the nice-to-have enhancements are:-

• Support CentOS 5 detection.
  • Currently, CentOS 5 is identified as 'RedHat' because it is missing
• Need to duplicate yum-only packages for RedHat, CentOS and Fedora
• The alternatives system doesn't do anything useful
  • Quite a complex problem, as the base64 or random logic shows
  • As alternatives can require program combinations and the presence of files
• Yum system requires more ways to install repos
  • Currently have 3 systems (what's on the machine, repoforge-specific, EPEL *-release style)
  • Would need to be able to download from a repo file from https (eg http://copr-fe.cloud.fedoraproject.org/coprs/petersen/pandoc-el5/repo/epel-5/petersen-pandoc-el5-epel-5.repo)
  • Would need to be able to create a repo file
• Need to be able to have per distribution version variants of path files
  • eg on CentOS 5, pandoc is not in EPEL but a different repository
  • could do this with more lines in the path fie
• Specify minimum / maximum versions of a program (inclusively)
  • Allows one to avoid older versions with different option syntax or missing essential features (eg curl, xmlstarlet)
  • Would need specific version parsers
• Support more core utils without needing a package manager by writing compatibility wrappers for common tasks, eg
  • BusyBox and ToyBox
  • Mac OS X and the BSDs
    • require the 'core_which' logic to be understand built ins
    • require cleverer ability to make tr as busybox tr
  • One of the biggest headaches is date. Others include hexdump, od and tr.
  • Out-of-scope: rarer commercial Unices (eg AIX, Solaris, HP-UX). Too difficult to get access to test on in the main. AIX and Solaris have quite good support for the GNU tools; HP-UX is moribound.