Loading ...
Sorry, an error occurred while loading the content.

Re: [json] Sorting the keys of formatted (pretty printed) JSON output

Expand Messages
  • Arthur Blake
    ... That was my first thought when thinking about how to do it, as well. That would make sense if you wanted to always keep and use the JSON sorted. In one of
    Message 1 of 7 , Apr 10, 2008
    • 0 Attachment
      On Thu, Apr 10, 2008 at 12:29 AM, Tatu Saloranta <tsaloranta@...> wrote:
      >
      > One question: wouldn't it be easier to just use TreeMap for storing
      > fields, instead of HashMap, if sorting is desired?
      >
      > -+ Tatu +-

      That was my first thought when thinking about how to do it, as well.

      That would make sense if you wanted to always keep and use the JSON
      sorted. In one of my main use cases (JSON-RPC debug output on the
      server side) I only want to sort the existing (non-sorted) JSON at
      print out time because the pretty printing is a special case (when
      debugging). I didn't want to transmit the JSON sorted (this would be
      an unneeded and unwanted side effect even though it is probably
      harmless.) I wanted to keep the JSON in the same format that it would
      have been in if I wasn't sorting. Thus I sort on the fly only when
      pretty printing (using a TreeSet view of the keys.)

      There would be a penalty for keeping a TreeMap versus a HashMap.
      HashMap obviously performs better (constant time put & get) for when
      you don't care about sorting. In practice I haven't noticed a speed
      difference for even very large JSON, but I'm sure it's there (TreeMap
      is O(log n) on gets and puts.) The same penalty exists for when
      sorting on the fly like I'm doing using a TreeSet, but it only comes
      in to play when pretty printing sorted (yes the downside is if you
      print the same object out multiple times, you re-sort each time.)

      It's a trade off doing it either way, and it would be quite easy to
      switch the HashMap used to a TreeMap in JSONObject if you desired to
      just always sort your JSON (or provide a sort mode switch when the
      JSONObject is created that makes it use a TreeMap versus a HashMap.)
    • Tatu Saloranta
      ... I understand this, and it s a valid point. But I was thinking more along the lines of deferring initialization of the Map, and switching between the two
      Message 2 of 7 , Apr 10, 2008
      • 0 Attachment
        On Thu, Apr 10, 2008 at 5:13 AM, Arthur Blake <arthur.blake@...> wrote:
        >
        > On Thu, Apr 10, 2008 at 12:29 AM, Tatu Saloranta <tsaloranta@...>
        > wrote:
        > >
        > > One question: wouldn't it be easier to just use TreeMap for storing
        > > fields, instead of HashMap, if sorting is desired?
        > >
        > > -+ Tatu +-
        >
        > That was my first thought when thinking about how to do it, as well.
        >
        > That would make sense if you wanted to always keep and use the JSON
        > sorted. In one of my main use cases (JSON-RPC debug output on the
        ...
        > There would be a penalty for keeping a TreeMap versus a HashMap.
        > HashMap obviously performs better (constant time put & get) for when

        I understand this, and it's a valid point. But I was thinking more
        along the lines of deferring initialization of the Map, and switching
        between the two if/as necessary. That is, it'd be a property of Json
        object, to minimize sorting overhead for cases where it's not needed.

        At any rate, I just wanted to mention it as an option, just to make
        sure it was considered. And that seems to have been the case.

        -+ Tatu +-
      • John Cowan
        ... Actually, it isn t. Even though the javadoc claims that get and put are constant time, a moment s reflection shows that this cannot be so. HashMap is
        Message 3 of 7 , Apr 10, 2008
        • 0 Attachment
          Tatu Saloranta scripsit:

          > > There would be a penalty for keeping a TreeMap versus a HashMap.
          > > HashMap obviously performs better (constant time put & get) for when
          >
          > I understand this, and it's a valid point.

          Actually, it isn't. Even though the javadoc claims that get and put
          are constant time, a moment's reflection shows that this cannot be so.

          HashMap is implemented (in OpenJDK, and almost certainly in earlier
          versions) using an array of specialized objects that implement Map.Entry.
          Each element of the array is one hash bucket, and the Entry objects form
          a linked list. So average get/put time is approximately N/c*2, where c
          is the capacity (initially 16 unless specified otherwise), which is O(N).
          The hope is that N/c*2 is always small enough that the performance looks
          like O(1), but as N grows, c is increased using a full rehash, which is
          again O(N) and is independent of c.

          So I'd say using TreeMap throughout, with its guaranteed O(log N)
          performance, is a perfectly reasonable thing to do; it makes debugging
          easier, it is prettier, and it guarantees getting the same output on
          the same input.

          --
          On the Semantic Web, it's too hard to prove John Cowan cowan@...
          you're not a dog. --Bill de hOra http://www.ccil.org/~cowan
        • kriszyp
          ... when ... Map.Entry. ... O(N). ... Not true. Full rehashes are indeed O(N), but they take place at a frequency proportional to 1/c because each time a
          Message 4 of 7 , Apr 10, 2008
          • 0 Attachment
            --- In json@yahoogroups.com, John Cowan <cowan@...> wrote:
            >
            > Tatu Saloranta scripsit:
            >
            > > > There would be a penalty for keeping a TreeMap versus a HashMap.
            > > > HashMap obviously performs better (constant time put & get) for
            when
            > >
            > > I understand this, and it's a valid point.
            >
            > Actually, it isn't. Even though the javadoc claims that get and put
            > are constant time, a moment's reflection shows that this cannot be so.
            >
            > HashMap is implemented (in OpenJDK, and almost certainly in earlier
            > versions) using an array of specialized objects that implement
            Map.Entry.
            > Each element of the array is one hash bucket, and the Entry objects form
            > a linked list. So average get/put time is approximately N/c*2, where c
            > is the capacity (initially 16 unless specified otherwise), which is
            O(N).
            > The hope is that N/c*2 is always small enough that the performance looks
            > like O(1), but as N grows, c is increased using a full rehash, which is
            > again O(N) and is independent of c.

            Not true. Full rehashes are indeed O(N), but they take place at a
            frequency proportional to 1/c because each time a rehash takes place
            the size is doubled, so the rehashes become farther apart the larger c
            gets. Therefore the average rehash time for a put operation is
            O(N*1/c) which averages to constant time. Both get and put only have
            to iterate through their own bucket, so as long as the hash method is
            reasonable, both get and put are indeed O(1), constant time operations.
            Kris
          • Douglas Crockford
            ... Excellent suggestion. I have modified JSON.Object.
            Message 5 of 7 , Apr 10, 2008
            • 0 Attachment
              --- In json@yahoogroups.com, "Arthur Blake" <arthur.blake@...> wrote:
              > When I have machine generated JSON data that is stored in files, I
              > like to have the JSON data be sorted alphabetically by the keys in the
              > output.
              >
              > It's much easier to find what you are looking for quickly this way
              > when viewing or editing the data (especially when the amount of data
              > is large.)
              >
              > Because there is no defined order of the keys in JSON, it doesn't
              > matter what order you put them in. So we might as well put it in an
              > order that makes it more human readable in contexts where the data is
              > meant to be viewed or edited by humans.

              Excellent suggestion. I have modified JSON.Object.
            Your message has been successfully submitted and would be delivered to recipients shortly.