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

Re: [pcgen-xml] Re: XML Conversion - phased

Expand Messages
  • Eric Beaudoin
    ... I was really hoping that this would not be the case i.e. that only the parsing code would be change. The monkeys have worked hard to make PCGEN go faster,
    Message 1 of 14 , Feb 28, 2003
    • 0 Attachment
      At 11:07 2003.02.28, Keith Davies wrote:
      >To be honest, I don't know how much of the existing code -- the code
      >that would be used in #2 -- would really get reused. I think that the
      >underlying mechanism of the program would be changing, and that the
      >scope of the changes would be broad enough that, while we could use the
      >existing code as a resource, it probably wouldn't make a great place to
      >start. The internal data model would be changing, the front end code
      >would be more or less replaced... all that would be left would be a bit
      >of the code used to handle certain game constructs, and that probably
      >wouldn't even work with the generalized data.

      I was really hoping that this would not be the case i.e. that only the parsing code would be change. The monkeys have worked hard to make PCGEN go faster, rewriting it at this point is not only high risk, it will probably be a major step back.

      The goal should be that external representation of the data should have a minimal impact on the application represent them internaly. This XML representation will most probably also be used for E-Tools so it stands a good chance of becoming the defacto data representation standard for d20 products. It is very important that we design a language that is as much application independant as possible.

      Also, for the phase parts, I'm for the quick gains. In my opinion, we should go to a "transition" XML schema that mimics the .LST syntax that we have first and then build on this. No mather how hard we try, we will not be able to create the "right" schema" on the first try anyway. It will be a sery of little evolutions getting us to the right point. We might as well start with something that is already familiar and work from there.

      My opinion anyway.

      P.S. Don't count on editor to much to hide the complexity of the schemas. All the major data contributors are still using text editors to get the job done. We should assume that it will stay that way and plan a schema that will be easy to work with rather than easy for the machine to read. It's easier to optimise a parser than to find and train dedicated data monkeys.


      -----------------------------------------------------------
      √Čric "Space Monkey" Beaudoin
      >> In space, no one can hear you sleep...
      >> Camels to can climb trees (and sometime eat them)
      <mailto:beaudoer@...>
    • Scott Ellsworth
      ... As one of the monkeys who has spent upwards of a coder-month working on optimization, I must humbly disagree. I am not sure that it would take less time
      Message 2 of 14 , Mar 1, 2003
      • 0 Attachment
        On Friday, February 28, 2003, at 11:34 PM, Eric Beaudoin wrote:

        > At 11:07 2003.02.28, Keith Davies wrote:
        >> The internal data model would be changing, the front end code
        >> would be more or less replaced... all that would be left would be a
        >> bit
        >> of the code used to handle certain game constructs, and that probably
        >> wouldn't even work with the generalized data.
        >
        > I was really hoping that this would not be the case i.e. that only the
        > parsing code would be change. The monkeys have worked hard to make
        > PCGEN go faster, rewriting it at this point is not only high risk, it
        > will probably be a major step back.

        As one of the monkeys who has spent upwards of a coder-month working on
        optimization, I must humbly disagree. I am not sure that it would take
        less time to fix the current data model than to write a new one. This
        is not to say that the current code is bad, just that it was not
        written originally for optimal speed, and we have enough data that we
        really need to have indexed hashes for our data.

        For example, the current code regularly iterates over collections of
        keys, doing a caseless text comparison of each key, because the keys
        double as user editable text. Further, it regularly reparses the same
        string for information it already parsed. (Every tab switch requires
        reparsing a whole bunch of entries for Variables, splitting on |
        characters.) These two things alone eat up well over 90% of the
        execution time, according to my profiler, and they are so deep in the
        program that they are very, very hard to fix. It took me four hours to
        do the work to just replace global weapon profs, and that was one of
        the easier ones.

        Keith has proposed a core data model where keys are unique and always,
        always, always in lower case.

        Further, in Keith's data model, we would not parse a string like
        BONUS:somestuff|moreStuff|+2|otherstuff
        more than once - the data would be broken apart on read, and stored in
        appropriate hashes or lists, so it would be easy to find out if an item
        gave a strength bonus, for example.

        We could split these efforts: convert the data model, and convert the
        XML files as separate tasks. The lst files would remain essentially
        the same, but would have separate key and user visible name data, and
        the reader would break apart all bonuses/variables on reading. I have
        been noodling away at that it my spare time, but it is not a fast
        process.

        If this was a priority, we would want to get a number of code monkeys
        working on it as a major feature. We could easily get an order of
        magnitude out of this, as we make over 5000 function calls just to
        switch a tab, and we should be making in the hundreds at the outside.

        Scott
      • Keith Davies
        ... Scott, please contact me offlist. I have some questions that I think you can answer for me; it d be a big help. ICQ 8550570, or AIM keithjdavies. Thanks,
        Message 3 of 14 , Mar 1, 2003
        • 0 Attachment
          On Sat, Mar 01, 2003 at 02:51:40PM -0800, Scott Ellsworth wrote:
          >
          > On Friday, February 28, 2003, at 11:34 PM, Eric Beaudoin wrote:
          >
          > > At 11:07 2003.02.28, Keith Davies wrote:
          > >> The internal data model would be changing, the front end code
          > >> would be more or less replaced... all that would be left would be a
          > >> bit
          > >> of the code used to handle certain game constructs, and that probably
          > >> wouldn't even work with the generalized data.
          > >
          > > I was really hoping that this would not be the case i.e. that only the
          > > parsing code would be change. The monkeys have worked hard to make
          > > PCGEN go faster, rewriting it at this point is not only high risk, it
          > > will probably be a major step back.
          >
          > As one of the monkeys who has spent upwards of a coder-month working on
          > optimization, I must humbly disagree. I am not sure that it would take
          > less time to fix the current data model than to write a new one. This
          > is not to say that the current code is bad, just that it was not
          > written originally for optimal speed, and we have enough data that we
          > really need to have indexed hashes for our data.

          Scott, please contact me offlist. I have some questions that I think
          you can answer for me; it'd be a big help. ICQ 8550570, or AIM
          keithjdavies.

          Thanks,
          Keith
          --
          Keith Davies
          keith.davies@...

          PCGen: <reaper/>, smartass
          "You just can't argue with a moron. It's like handling Nuclear
          waste. It's not good, it's not evil, but for Christ's sake, don't
          get any on you!!" -- Chuck, PCGen mailing list
        • merton_monk <merton_monk@yahoo.com>
          I m logged into AIM right now as CMPMerton (might be a space between CMP and Merton, I forget) if you d like to discuss the persistence layer in PCGen or data
          Message 4 of 14 , Mar 1, 2003
          • 0 Attachment
            I'm logged into AIM right now as CMPMerton (might be a space between
            CMP and Merton, I forget) if you'd like to discuss the persistence
            layer in PCGen or data model. Anyone who wants to help with the xml
            conversion can join... I'll probably be online for a couple of hours.

            -Bryan

            --- In pcgen-xml@yahoogroups.com, Keith Davies <keith.davies@k...>
            wrote:
            > On Sat, Mar 01, 2003 at 02:51:40PM -0800, Scott Ellsworth wrote:
            > >
            > > On Friday, February 28, 2003, at 11:34 PM, Eric Beaudoin wrote:
            > >
            > > > At 11:07 2003.02.28, Keith Davies wrote:
            > > >> The internal data model would be changing, the front end code
            > > >> would be more or less replaced... all that would be left
            would be a
            > > >> bit
            > > >> of the code used to handle certain game constructs, and that
            probably
            > > >> wouldn't even work with the generalized data.
            > > >
            > > > I was really hoping that this would not be the case i.e. that
            only the
            > > > parsing code would be change. The monkeys have worked hard to
            make
            > > > PCGEN go faster, rewriting it at this point is not only high
            risk, it
            > > > will probably be a major step back.
            > >
            > > As one of the monkeys who has spent upwards of a coder-month
            working on
            > > optimization, I must humbly disagree. I am not sure that it
            would take
            > > less time to fix the current data model than to write a new
            one. This
            > > is not to say that the current code is bad, just that it was not
            > > written originally for optimal speed, and we have enough data
            that we
            > > really need to have indexed hashes for our data.
            >
            > Scott, please contact me offlist. I have some questions that I
            think
            > you can answer for me; it'd be a big help. ICQ 8550570, or AIM
            > keithjdavies.
            >
            > Thanks,
            > Keith
            > --
            > Keith Davies
            > keith.davies@k...
            >
            > PCGen: <reaper/>, smartass
            > "You just can't argue with a moron. It's like handling Nuclear
            > waste. It's not good, it's not evil, but for Christ's sake,
            don't
            > get any on you!!" -- Chuck, PCGen mailing list
          • CC Americas 1 Carstensen James
            Keith, Eric and Scott said it, but I think the big question is how much work the Code Monkeys are willing to do. I ll sum up a couple of points from different
            Message 5 of 14 , Mar 3, 2003
            • 0 Attachment
              Keith,

              Eric and Scott said it, but I think the big question is how much work
              the Code Monkeys are willing to do. I'll sum up a couple of points from
              different posts here.

              Eric Beaudoin wrote:
              > Also, for the phase parts, I'm for the quick gains. In my opinion, we
              should go to a "transition" XML schema that mimics the .LST syntax that
              we have first and then build on this. No mather how hard we try, we will
              not be able to create the "right" schema" on the first try anyway. It
              will be a sery of little evolutions getting us to the right point. We
              might as well start with something that is already familiar and work
              from there.

              Eric, you're contributed much more then me to PCGen, take this with the
              respect it's intended. From looking at the list files, and looking at
              the "cleaner" implementation that Keith wants, it's more of a revolution
              then an evolution. Some things are just done in a different enough way
              that getting there from a direct tab-to-XML conversion is not easy, may
              not be optimal, and may end up just being "bolted on" rather then a
              clean implementation.

              I myself way hoping for the same evolution, but more from Step #2 to
              Step #3, and even there I don't know how much that's true.

              The current LST structure has evolved over time to handle more and more,
              and it has both it's strengths and weaknesses. XML has different
              strengths and weaknesses, and I think that if we just pull over the LST
              files to XML format, we'll be importing some of the weaknesses of LST
              while not mitigating some of the weaknesses of XML.

              But we grow from there - maybe that's worth that initial period to
              convert people over quickest and then grow. Is growing from a syntax
              that won't take advantage of XMLs strengths to one that will a
              longer/hander process then doing a single big change and have a good
              foundation to build on?

              Scott Ellsworth wrote:
              > We could split these efforts: convert the data model, and convert the
              XML files as separate tasks. The lst files would remain essentially
              the same, but would have separate key and user visible name data, and
              the reader would break apart all bonuses/variables on reading. I have
              been noodling away at that it my spare time, but it is not a fast
              process.

              Is this something that would be going on concurrently, before, or after
              the XML conversion? It seems like even if we stayed with LST files
              working out a way to do unique lowercase identifiers would be a benefit.

              Cheers,
              Blue

              -----Original Message-----
              From: Keith Davies [mailto:keith.davies@...]
              Sent: Saturday, March 01, 2003 1:38 AM
              To: pcgen-xml@yahoogroups.com
              Subject: Re: [pcgen-xml] Re: XML Conversion - phased


              On Fri, Feb 28, 2003 at 08:07:04AM -0800, Keith Davies wrote:
              > On Fri, Feb 28, 2003 at 09:36:54AM -0500, CC Americas 1 Carstensen
              James wrote:
              > >
              > > Let me summarize to make sure I understand your points. Your three
              > > points of XML conversion are:
              > >
              > > 1. XML that exactly mimics current tab-separated LST files
              > > 2. XML that describes all of the game elements in an XML way
              > > 3. XML that goes all they way, takes #2 but also supports
              meta-level
              > > descriptions and game rules.

              <killer snip>

              Any questions or comments? I've been asked to move things up and unless
              someone comes up with something I haven't considered before now, it'll
              be decision time.


              Keith
              --
              Keith Davies
              keith.davies@...

              PCGen: <reaper/>, smartass
              "You just can't argue with a moron. It's like handling Nuclear
              waste. It's not good, it's not evil, but for Christ's sake, don't
              get any on you!!" -- Chuck, PCGen mailing list


              To unsubscribe from this group, send an email to:
              pcgen-xml-unsubscribe@yahoogroups.com



              Your use of Yahoo! Groups is subject to
              http://docs.yahoo.com/info/terms/
            • Keith Davies
              ... Hi All, things have changed over the weekend; I will be posting more later today (I m a little busy at work right now). The schedule is being moved up and
              Message 6 of 14 , Mar 3, 2003
              • 0 Attachment
                On Mon, Mar 03, 2003 at 09:34:09AM -0500, CC Americas 1 Carstensen James wrote:
                > Keith,
                >
                > Eric and Scott said it, but I think the big question is how much work
                > the Code Monkeys are willing to do. I'll sum up a couple of points from
                > different posts here.

                Hi All,

                things have changed over the weekend; I will be posting more later today
                (I'm a little busy at work right now). The schedule is being moved up
                and I've had to make some (somewhat unilateral) decisions. A plan has
                been submitted to the BoD for ratification and I expect we'll start on
                it very soon. To summarize, however:

                1. We will use (more or less) the simplest and most direct translation
                of LST to XML for now. *Some* of my design will apply, but for the
                most part the XML will be very recognizable and easily understood by
                non-XML monkeys. This was a key point both for the sake of data
                monkey comfort and simple time pressure.
                2. The internal data model will, for the most part, remain untouched.
                We may be able to sneak some changes into the IDM, but for the most
                part we're aiming at minimal impact.
                3. XML serialization will be supported in parallel with LST I/O, at
                least during the early stages. Once XML support is in, LST use is
                deprecated and, in my schedule, slated for earliest possible removal
                (next major release will be as soon as we can) in order to reduce
                maintenance headaches.
                4. All distributed data files will be converted in one pass, rather than
                file type by file type (increase amount of pain, minimize duration).
                5. A converter will be provided for LST files in the field.
                6. We will use the time until the release of 5.0.0 for analysis of the
                code and LST structure; once 5.0.0 hits the streets we'll branch the
                source and start the changes.
                7. After 6.0.0 (full XML, LST removed) is released, we branch the source
                as described in an earlier posting and pursue the path I'd *like* to
                take.

                The plan above will get us into XML and play on the monkey's familiarity
                with LST files. It will be crap XML -- in the sense that while it will
                comply with XML standards it will be poorly-designed XML that will not
                take advantage of all the benefits that XML will give us -- but will
                cause the least distress among data monkeys and least impact on the
                code. The conversion will also be simplest, almost 1:1 congruence
                except where benefits can be gained by making a change.

                We're moving, folks, and while it's not the way I'd been hoping to go,
                it will get us headed in the right direction.


                Keith
                --
                Keith Davies
                keith.davies@...

                PCGen: <reaper/>, smartass
                "You just can't argue with a moron. It's like handling Nuclear
                waste. It's not good, it's not evil, but for Christ's sake, don't
                get any on you!!" -- Chuck, PCGen mailing list
              Your message has been successfully submitted and would be delivered to recipients shortly.