the apr question:

  should we depend on apr/apr-util?

  we currently use:

    hash table and variable length arrays

    pools for memory allocation

    bucket brigades for our output

    file abstraction for reading from disk

    the sprintf stuff

  problems:

    arrays suck

    pools don't play nice with theoretical GC schemes

    pools suck when you're writing bindings for dynamic languages

    bucket brigades aren't really needed for what we're doing

  potential plan to replace apr:

    remove use of apr arrays, replacing with custom solution

    remote use of apr hash tables, replacing with custom solution

    switch from pools to regular old malloc/free

    replace bucket brigades with our own stream implementation

    bring in the public domain sprintf code