This project mostly focuses on continuing the design of the bear installation tool (https://gitlab.com/coala/cib) and fixing dependency problems across bears. Also, there should be a way for fixing bears which have platform requirements (we could use conda).

Currently, cib does basic operations but rudimentary, by parsing strings and running "pip install" on packages.

Once this project is done, cib should be able to perform the following:

  • install any number of bears, specified as command line arguments
  • install all dependencies for required bears
  • update to the latest version any number of bears
  • check if dependencies are fulfilled for any number of bears
  • uninstall any number of bears

Each bear has a REQUIREMENTS tuple which holds instances of requirement classes, found here (https://gitlab.com/coala/package_manager), which should be reworked so that most operations in cib would be methods from these classes.

The problem right now is that some bears are not even getting installed by cib. Besides those with which we have requirement classes, the rest aren't even treated. So the plan is to create package managers for the popular managers, and treat them all in a universal DistributionRequirement.

How this will work: Each requirement class will have an "install_command" and an "install" method. We will just call requirement.install() and this will run the command on its own, not caring which platform you're on. Ideally, requirement.install() will use "install_command" to generate the command and run it with "sarge". For the DistributionRequirement class, the "install_command" method will check on which platform you are, and it will generate the respective command. If the package is named differently on each platform, it shall be specified in the constructor. Take a mock example for a random bear:

REQUIREMENTS=(DistributionRequirement( apt-get='csslint', dnf='csslinter', brew='csslt')

If you're on Ubuntu, it will run "apt-get install csslint". If you're on Fedora, it will run "dnf install csslinter", and so on. If your package manager is not on the list, we should just print a warning message so that the user tries to install it manually.

So cib should not care what kind of requirement each is, it should just iterate through requirement instances and run "install_command()" on each, with the requirement class handling the rest.

Another special case is made of bears which hold different installations for each distribution. A good example is DartLintBear: it has different installation options for Linux (a wget script), Windows and Mac. https://www.dartlang.org/install

A good idea to treat these is Conda: a conda package contains a script for each operating system (a .sh for Mac/Linux and a .bat for Windows). When you try to install the conda package, it will run the according script.

The last special case is made out of bears which have bear dependencies. We have quite a few of them (the Clang bears). For these, we should have a BearRequirement class which means that this bear is a requirement of another bear.

Milestones

PREPARATION/BONDING
CODING PHASE 1
  • All bears should have dependencies sorted out.
  • The requirement classes should be rewritten so that most of the work is done within them, cib only having to run their methods and to gather the bears.
  • Tests should be written for all the dependencies and the classes.
  • All requirement classes shall be properly documented.
CODING PHASE 2
  • The DistributionRequirement class should be fully working, and bears which have such requirements should have them fulfilled.
  • Conda packages should be created for all bears which need such deps.
CODING PHASE 3
  • cib should be fully functional such as that it can be used by users instead of having to install all the bears.
  • Bears which have bear dependencies should be sorted out.
  • Tests should be written for a full coverage of cib.
  • cib should be fully documented, while also having a nice "help" menu.

results matching ""

    No results matching ""