What Went Wrong
sucks! While it's true that GOAL gave us many advantages, GOAL caused
us a lot of grief. A single programmer (who could easily be one of the
top ten Lisp programmers in the world) wrote GOAL. While he called his
Lisp techniques and programming practices "revolutionary," others
referred to them as "code encryption," since only he could understand
them. Because of this, all of the support, bug fixes, feature enhancements,
and optimizations had to come from one person, creating quite a bottleneck.
Also, it took over a year to develop the compiler, during which time the
other programmers had to make do with missing features, odd quirks, and
GOAL became much more robust, but even now C++ has some advantages over
GOAL, such as destructors, better constructors, and the ease of declaring
A major difficulty was that we worked in such isolation from the rest
of the world. We gave up third-party development tools such as profilers
and debuggers, and we gave up existing libraries, including code previously
developed internally. Compared to the thousands of programmers with many
years of C++ experience, there are relatively few programmers with Lisp
experience, and no programmers (outside of Naughty Dog) with GOAL experience,
making hiring more difficult.
GOAL's ability both to execute code at the listener and to replace existing
code in the game at run time introduced the problem of memory usage, and
more specifically, garbage collection. As new code was compiled, older
code (and other memory used by the compiler) was orphaned, eventually
causing the PC to run low on free memory. A slow garbage collection process
would automatically occur when available memory became sufficiently low,
and the compiler would be unresponsive until the process had completed,
sometimes taking as long as 15 minutes.
programming. Because we were so busy creating the technology for our
seamless world, we didn't have time to work on gameplay code until fairly
late in the project. The situation caused no end of frustration to the
designers, who were forced to design levels and creatures without being
able to test whether what they were doing was going to be fun and play
well. Eventually programmers were moved off of technology tasks and onto
gameplay tasks, allowing the designers to play the game and make changes
as appropriate. But without our designers' experience, diligence, and
forethought, the results could have been a disaster.
We were plagued with audio-related problems from the start. Our first
indication that things might not be going quite right was when our sound
programmer quit and moved to Australia. Quickly hiring another sound programmer
would have been the correct decision. We tried several other schemes,
however, made some poor choices, and had quite a bit of bad luck. We didn't
recognize until fairly late in development what a monumental task audio
was going to be for this project. Not only did Jak & Daxter
contain original music scores, creature and gadget noises, ambient sounds,
and animated elements, but there are also over 45 minutes of story sequences,
each containing Foley effects and speech recorded in six different languages.
Our audio issues could be broken up into four categories: sound effects,
spooled Foley, music, and localized dialogue. Due to the large number
of sound effects in the game, implementing sound effects became a maintenance
nightmare. No single sound effect was particularly difficult or time-consuming;
however, creating all of the sound effects and keeping them all balanced
and working was a constant struggle. We needed to have more time dedicated
to this problem, and we needed better tool support.
We used spooled Foley for lengthy sound effects, which wouldn't fit well
in sound RAM. Spooling the audio had many advantages, but we developed
the technology too late in the project and had difficulty using it due
to synchronization issues.
Our music, although expertly composed, lacked the direction and attention
to detail that we had achieved with the Crash Bandicoot games.
In previous games, we had a person who was responsible for the direction
of the music. Un-fortunately, no one performed that same role during Jak
Dialogue is a difficult problem in general due to the complexity of writing,
recording, editing, creating Foley, and managing all of the audio files,
but our localization issues made it especially challenging. Many problems
were difficult to discover because of our lack of knowledge of the various
languages, and we should have had more redundant testing of the audio
files by people who were fluent in the specific languages.
processing times. One of our greatest frustrations and loss of productivity
came from our slow turnaround time in making a change to a level or animation
and seeing that change in the actual game.
Since all of our tools (including the GOAL compiler) ran across our network,
we ran into severe network bandwidth issues. Making better use of local
hard drives would have been a smarter approach. In addition, we found
extreme network slowdown issues related to reading file time/date stamps,
and some tools took several minutes just to determine that nothing needed
to be rebuilt. When we compiled some of our tools under Linux, we noticed
dramatic improvements in network performance, and we are planning on using
Linux more extensively in our next project.
We implemented the processing of the lengthy story-sequence animations
as a hack of the system used to process the far simpler creature animations.
Unfor-tunately, this bad system caused lengthy processing times, time-consuming
debugging, and a lot of confusion. If we had initially hidden the processing
complexity behind better tools, we would have saved quite a bit of time.
We used level-configuration scripts to set actor parameters and other
level-specific data. The script processing was done at an early stage
in our tools pipeline, however, so minor data changes took several minutes
to process. We learned that tunable data should instead be processed as
close as possible to the end of the tools pipeline.
tools. We created many tools while developing Jak & Daxter,
but many of our tools were difficult to use, and many tools were needed
but never written. We often didn't know exactly what we needed until after
several revisions of our technology. In addition, we didn't spend a lot
of time polishing our tools, since that time would have been wasted if
the underlying technology changed.
Regrettably, we did not have time to program tools that were badly needed
by the artists, which resulted in a difficult and confusing environment
for the artists and caused many productivity issues. Since programming
created a bottleneck during game production, the added burden given to
the artists was considered necessary, though no less distasteful.
We lacked many visualization tools that would have greatly improved the
artists' ability to find and fix problems. For example, the main method
artists used to examine collision was a debugging mode that colorized
a small section of collision geometry immediately surrounding Jak. A far
better solution would have been to create a renderer to display the entire
collision of a level.
We created plug-ins that were used within the 3D modeling package; however,
for flexibility's sake most of the plug-ins operated by taking command
parameters and outputting results as text: not a good interface for artists.
Eventually, one of our multi-talented artists created menus and other
visualization aids that significantly improved productivity.
Many of our tools were script based, which made the tools extremely flexible
and adaptable; however, the scripts were often difficult for the artists
to understand and use. We are replacing many of these scripts with easier-to-use
GUIs for our next project.
Jak & Daxter was a monumental effort by many hardworking, talented
people. In retrospect, we were probably pretty foolish to take on as many
challenges as we did, and we learned some painful lessons along the way.
But we also did many things right, and we successfully achieved our main
goals. At Naughty Dog, there is a strong devotion to quality, which at
times can border on the chaotic, but we try to learn from both our success
and our failures in order to improve our processes and create a better
game. The things that we learned from Jak & Daxter have made us a
stronger company, and we look forward to learning from our past as we
prepare for the new challenges ahead.
& Daxter: the Precursor Legacy
Sony Computer Entertainment
Number of full-time developers: 35
Length of development: 1 year of initial development, plus
2 years of full production
Release date: December, 2001
Platform: Playstation 2
Development software used: Allegro, Common Lisp, Visual C++,
Maya, Photoshop, X Emacs, Visual Slick Edit, tcsh, Exceed, CVS