Coding Conventions

There is a pylint.rc file that is tweaked for most of the project style. It is not perfect, but is a good check to run immediately after cloning, and then right before a pull request (as pylint will report the difference in compliance once you have run it once). The script lint.sh in the root directory will run pylint with this file over the proper directory tree.

  • All indentation uses spaces

  • Indents are 2 spaces

  • Exceptions that exist only for specializing names should be written on one line:

    class MyCommonException(Exception):
      def __init__ (self, some_data):
        ...
      def __str__ (self):
        ...
    
    class SpecificNameOne(MyCommonException): pass
    class SpecificNameTwo(MyCommonException): pass
    

    There is an example of this use case in geni.aggregate.pgutil

  • Maximum line length is 132 characters

  • Class-level and global variables are highly discouraged

  • Never use bare except: clauses

  • Do not use print

Pattern Conventions

  • Exception hierarchies are highly encouraged, as they allow users to dispatch their own scripting on the the type of exception, rather than having a single exception with varying messages or errno-like mechanics.

  • Anything that imports cryptography should never be imported at module level. Always import crypto and network functionality inside functions, as this minimizes the dependencies for users who are only generating or parsing XML.

  • All public attribute storage types should be internally consistent for reads. If you want to support setting an underlying float storage type using a string or integer, use a property. Non-public attributes should also follow this rule unless there is a really good reason not to.

  • If you can foresee a problem in the future, but don’t have time to fix it now, at least leave a # TODO note.

  • At the moment many geni-lib instances are re-entrant. However, while we should support this where possible, it is not required nor will it be guaranteed to users.

Philosophy Notes

geni-lib is a surprisingly useful tool for end-users. However, that does not mean it is designed as a user tool. As it is a library, constraints and “convenience” can always be wrapped around it, but they can’t be removed if the library is too opinionated at a base level. Convenience is generally acceptable if the underlying functionality can be directly accessed by the user should they want to avoid whatever level of “help” is being offered.

  • Do not provide Magic(tm) to users in the base API. geni-lib should not go out of its’ way to protect the user from building a request that we believe the be impossible to satisfy, or from passing “bad” data to AM API calls.

  • AM wrappers provide some basic level of sanity checking of input values (specifically for POAs in the VTS support). This is acceptable as there is a lower-level API a user may use if they want to ignore those checks (geni.minigcf.amapi3). That being said, this level of convenience should be limited.

  • While we support IronPython and Jython, those geni-lib users should still prefer multiprocessing for parallel execution, instead of threading.

  • Providing convenience shouldn’t extend to essentially providing new tools - it’s hard to say where this line is, but providing too much tool-like functionality in geni-lib puts pressure on both the release schedule and the versioning and API stability.

Things That Don’t Belong

For legacy reasons, geni-lib includes code that either doesn’t belong entirely, or is in the wrong place. Anything enumerated here should not be referred to as an example of good practice moving forward:

  • geni.aggregate.* - This package is a mess. Frameworks need to move out somewhere else, and the AM locations should not be in the code (they should be loaded from data files which can be updated independently of the geni-lib code). context also doesn’t belong here, once framework support moves.

  • geni.rspec.pg* - The “pg” rspec is now the “GENI” rspec, and we should rename accordingly. Things that are actually part of Emulab and not the base functionality should live in extensions.