1. 9
  1.  

  2. 2

    The article is fine as far as it goes, because it makes no prescriptions based on this data. The question is: is the data actionable, and how? If not, why is it interesting?

    1. 2

      Key observations I got from the article:

      • A tiny minority of functions have the majority of changes.
      • A vast majority of functions only have a couple of authors.
      • Functions stop being modified around 10K hours (just over 1.14 years).
      • Functions are equally likely to be modified at any given time until they stop being modified (either because they’re stabilized or removed).

      Actions this information suggest to me (trying to read the auspices):

      • For development speed, optimize for workflows that empower individual authors of functions–perhaps this includes streamlining code review and skipping maintenance documentation.
      • For longevity, either find a way of retaining the original authors (better pay? be nicer to them? avoid witchhunts for badthink?) or emphasize pair programming or similar stuff to purposefully break the trend of single-author development.
      • Any maintenance plan longer than about a year is probably optimistic.
      • There is no “quick hack” up until about a year in–expect the code to be actively modified/maintained up until that point.
      1. 2

        Functions are equally likely to be modified at any given time until they stop being modified (either because they’re stabilized or removed).

        Ooh, I missed this one. Thank you!

        The deep and well-known problem in the vicinity of OP is: 1% of the code we write will last a long time, but we don’t know which 1% it is.

    2. 1

      I’m having some difficulty identifying what the source of the data the author uses is — my guess is Github, which would make this interesting but not very practical or representative of “real life”. Anyone have any ideas?

      1. 1

        I checked out the source data (he links a tarball of code and data), it looks like it’s an analysis of just two repositories. Probably evolution and apache by googling a file name.

        1. 1

          Interesting! I need to dig into that a little bit later - seems like too small of a sample size to really draw conclusions.