Skip to content

Build and install cx_Oracle on MacOS X Leopard Intel

I recently decided to write a few tools to automate some boring database management tasks. Since I am not attached to any particular technology, I decided to do take the opportunity to get to know the Django web framework. Getting a simple server up and running is as easy as doing the same with the Rails framework. In order to do something useful, one needs to integrate a database system, which in my case is Oracle 10g.

In Django’s, one can setup the database engine, which, luckily includes oracle. Unless you already have a working Python-Oracle setup, you will have to install the cx_Oracle Python database adapter first. Prebuilt binaries for Windows and Linux can be downloaded from the Sourceforge archive directly, since I mainly work on a MacBook Pro, I had to build the module by myself.

There are a few little changes I had to make in order to build and install it successfully, which I will describe here. I used the advice on Pedro’s Blogas a starting point and adapted it so that it worked for me.

I. Downloading the required software:

System Requirements:

To build the module, you need at least the Oracle Instant Client software for MacOS X, which can be downloaded here. The minimum packages that need to be downloaded are the “Basic” and the “SDK” packages.

From the cx_Oracle Sourceforge site, download cx_Oracle-4.3.3.tar.gz (that's the version I used)

II. Unpacking files

I assume that I uncompress my files in /Users/foo/oracle. That’s just my personal setup of course, because I tend to install stuff that I just want to try out under my home folder. Many people would probably put their ORACLE_HOME under /usr/local/oracle.

cd ~/oracle
tar xvfz cx_Oracle-4.3.3.tar.gz

III. Setup of the environment

To my ~/.profile, I added the following lines:

# Setting Oracle development environment
export ORACLE_HOME=/Users/foo/oracle/instantclient_10_2

and in the /Users/foo/oracle/instantclient_10_2 directory, I entered

ln -s libclntsh.dylib.10.1 libclntsh.dylib

(otherwise the build script does not find the dynamic library).

IV. Edit

In my case, I could not run the build script out of the box, because it was complaining that the Oracle version could not be determined. That’s because is looking for a $ORACLE_HOME/lib directory (or$ORACLE_HOME/bin on a Windows system) and trying to determine the Oracle version by looking for the client libraries it contains. In my case, I don’t have a lib directory, so I patched the and set the Oracle version by hand.

In your favorite text editor, open /Users/foo/oracle/cx_Oracle-4.3.3/ and replace:

        if self.oracleVersion is None:
            raise DistutilsSetupError, "Oracle home does not refer to an " \
                    "8i, 9i, 10g or 11g installation"


        if self.oracleVersion is None:
          self.oracleVersion = "10g"

or whatever your Oracle version is.

V. Build and install

This is just as the README.txt describes:

sudo python build
sudo python install

(I got a warning "ld: warning in /Users/foo/Development/oracle/instantclient_10_2/libclntsh.dylib, file is not of required architecture", but I simply ignored it)

I also created a $ORACLE_HOME/network/admin directory and copied the service file our DBA defined (tnsnames.ora) into it in order to make the services known to the Oracle client installation on my Mac. I did a few test queries to verify that it worked as expected.

I have written down my personal setup process for cx_Oracle, because it was somewhat different from Pablo’s approach and I have different OS and software versions. Hopefully, this can be of some help for someone else (or myself, should I ever have to do it again in the future).

Challenges In Updating ZMPP’s Screen Model

So now ZMPP has a dramatically simplified setup and execution model, those changes were actually much quicker to do than I thought and each took about a day. On the other hand I did not really make changes to the core, but simply wrapped it in a different way so it can be easier reused in different contexts.

The real challenge is the rework of the screen model – in fact there are three:

  1. “classic” screen model (V1-3)
  2. “modern” screen model (V4-8 except 6)
  3. V6 screen model
1. The “classic” screen model

Classic screen model in ZMPP 1.02

What I call the “classic” screen model is the screen model used by the majority of Infocom text adventures, such as the Zork trilogy, “Leather Goddesses of Phobos” etc. It is characterized by a main text window with a standardized status bar at the top. This model is very simple to implement with standard GUI components and even transfers well to an HTML interface. Unfortunately, aside from the Infocom adventures, this screen model is not frequently used by modern Interactive Fiction.

2. The “modern” screen model

Modern screen model in ZMPP 1.02

This is the screen model which is by far the most popular of the current Z-machine screen models. A top window rendered in a fixed pitch font, without scrolling and word wrapping and a bottom window for main text display and input which can have variable pitch font, scrolling and word wrapping. The top window’s number of rows can be specified by the game and the bottom window will shrink accordingly so that the sum of the heights of both window add up to the total height of the window.

The popularity of this particular screen model can be explained by the availability of one of the best authoring systems for Interactive Fiction: Inform. Inform’s output formats are usually V5 or V8 (or Glulx, which is a different virtual machine for Interactive Fiction with larger available memory).

This screen model does not translate itself easily to standard user interface components or web interfaces, which is mainly due to the fact that a few games use the behavior of the screen model that the content of the windows are not erased after the top window was resized. So, in “Curses” by Graham Nelson, the top window is used to render a book citation in box (because of the fixed pitch font) and shrinks so that the former content of the top window now belongs to the bottom window (and will be cleared away). This behavior is difficult to implement with standard UI components or in a web interface, because rendering into other component’s regions is usually not desired.

Example for the modern screen model issue in ZMPP 1.02

3. V6 screen model

This is the most complicated screen model and neither widely used by Interactive Fiction authors nor commonly implemented by many Z-machine interpreters. It uses 8 windows which can contain graphics or text and can use an interrupt mechanism to implement word wrapping. ZMPP implements it, but currently has a number of issues due to lack of testing. I will probably describe it in more detail when I get to work on that.

For the near future, reimplementing the classic and modern screen models is the main focus. The requirements I have distilled from user requests and my own ideas are:

  • use JTextPane for the main text window if possible
  • support selection, cut, copy, paste
  • support resizing
Since I do not have enough in-depth experience using the styled text components of Swing in the way described (that’s why I ended up with custom rendering for the current versions – it was quicker to implement), I will create a number of prototypes in the next days to see what is possible.

Back To The Z-Machine

In the last two years, I have only made smaller fixes and changes to ZMPP. The reasons were various: learning to be a father, tinkering with TinyUML and playing with Z-machine implementations in Ruby and Erlang. While working on the other pet projects, I noticed that ZMPP is still the most successful of them, simply because it is the one with an existing user base besides myself.

ZMPP is ready for a major overhaul. There are several pain points I want to tackle:

  • the public interface between the core and the user interface is too complicated and makes implementing different front ends a big pain.
  • all currently released ZMPP versions follow a game-loop style scheme, which means that the Z-machine runs in its own thread, until it stops, timed input is realized by threading as well. Besides making the execution and input logic too complicated, there are reportedly deadlocks which are almost definitely a result of this design decision. However so far, ZMPP is probably the Java Z-machine where timed input works best
  • top priority and the largest challenge is the rework of the screen model. On Java, ZMPP has the most complete screen model implementation, but it is still far from what I want it to be. I will elaborate on that later.

So, with that said, it means that I am waving good-bye to TinyUML, Smeagol and Schmalz in order to focus on ZMPP entirely. As a single, it is easy to find time to do a lot of different projects, but once you get married and have children, programming in the free time while still important, should not exactly be the top priority.