Keep around Stas's narrative for archival purposes.

[super-star-trek.git] / doc / HACKING

This is the hackers' guide to SST2K.  Read it before messing with the code.

It consists of an introduction, a history, suggestions for regression testing,
and some notes on the Python translation.  For a to-do list, see TODO in the
top-level directory.

INTRODUCTION:

SST2K is a Python translation of a C translation of a FORTRAN
original dating back to 1973.  Beautiful Python it is not, but it
works. 

The intention of SST2K is to be able to replicate precisely the 
experience of the original game, while allowing new features to be
added under option control.  Therefore, be very conservative about
what changes are visible under the 'plain' option.

HISTORY:

Dave Matuszek, one of the two original authors, says:

SRSCAN, MOVE, PHASERS, CALL, STATUS, IMPULSE, PHOTONS, ABANDON,
LRSCAN, WARP, SHIELDS, DESTRUCT, CHART, REST, DOCK, QUIT, and DAMAGE
were in the original non-"super" version of UT FORTRAN Star Trek.

Tholians were not in the original. Dave is dubious about their merits.
(They are now controlled by OPTION_THOLIAN and turned off if the game
type is "plain".)

Planets and dilithium crystals were not in the original.  Dave is OK
with this idea. (It's now controlled by OPTION_PLANETS and turned 
off if the game type is "plain".)

Dave says the bit about the Galileo getting turned into a
McDonald's is "consistant with our original vision".  (This has been
left permanently enabled, as it can only happen if OPTION_PLANETS
is on.)

Dave also says the Space Thingy should not be preserved across saved
games, so you can't prove to others that you've seen it.  He says it
shouldn't fire back, either.  It should do nothing except scream and
disappear when hit by photon torpedos.  It's OK that it may move
when attacked, but it didn't in the original.  (Whether the Thingy
can fire back is now controlled by OPTION_THINGY and turned off if the
game type is "plain" or "almy".  The no-save behavior has been restored.)

The Faerie Queen, black holes, and time warping were in the original.

Here are Tom Almy's changes:

In early 1997, I got the bright idea to look for references to
"Super Star Trek" on the World Wide Web. There weren't many hits,
but there was one that came up with 1979 Fortran sources! This
version had a few additional features that mine didn't have,
however mine had some feature it didn't have. So I merged its
features that I liked. I also took a peek at the DECUS version (a
port, less sources, to the PDP-10), and some other variations.

1, Compared to the original UT version, I've changed the "help"
command to "call" and the "terminate" command to "quit" to better
match user expectations. The DECUS version apparently made those
changes as well as changing "freeze" to "save". However I like
"freeze".  (Both "freeze" and "save" work in SST2K.)

2. The experimental deathray originally had only a 5% chance of
success, but could be used repeatedly. I guess after a couple
years of use, it was less "experimental" because the 1979
version had a 70% success rate. However it was prone to breaking
after use. I upgraded the deathray, but kept the original set of
failure modes (great humor!).  (Now controlled by OPTION_DEATHRAY
and turned off if game type is "plain".)

3. The 1979 version also mentions srscan and lrscan working when
docked (using the starbase's scanners), so I made some changes here
to do this (and indicating that fact to the player), and then realized
the base would have a subspace radio as well -- doing a Chart when docked
updates the star chart, and all radio reports will be heard. The Dock
command will also give a report if a base is under attack.

4. Tholian Web from the 1979 version.  (Now controlled by
OPTION_THOLIAN and turned off if game type is "plain".)

5. Enemies can ram the Enterprise. (Now controlled by OPTION_RAMMING
and turned off if game type is "plain".)

6. Regular Klingons and Romulans can move in Expert and Emeritus games. 
This code could use improvement. (Now controlled by OPTION_MVBADDY
and turned off if game type is "plain".)

7. The deep-space probe feature from the DECUS version.  (Now controlled
by OPTION_PROBE and turned off if game type is "plain").

8. 'emexit' command from the 1979 version.

9. Bugfix: Klingon commander movements are no longer reported if long-range 
sensors are damaged.

10. Bugfix: Better base positioning at startup (more spread out).
That made sense to add because most people abort games with 
bad base placement.

In June 2002, I fixed two known bugs and a documentation typo.
In June 2004 I fixed a number of bugs involving: 1) parsing invalid
numbers, 2) manual phasers when SR scan is damaged and commander is
present, 3) time warping into the future, 4) hang when moving
klingons in crowded quadrants.  (These fixes are in SST2K.)

Here are Stas Sergeev's changes:

1. The Space Thingy can be shoved, if you ram it, and can fire back if 
fired upon. (Now controlled by OPTION_THINGY and turned off if game 
type is "plain" or "almy".)

2. When you are docked, base covers you with an almost invincible shield. 
(A commander can still ram you, or a Romulan can destroy the base,
or a SCom can even succeed with direct attack IIRC, but this rarely 
happens.)  (Now controlled by OPTION_BASE and turned off if game 
type is "plain" or "almy".)

3. Ramming a black hole is no longer instant death.  There is a
chance you might get timewarped instead. (Now controlled by 
OPTION_BLKHOLE and turned off if game type is "plain" or "almy".)

4. The Tholian can be hit with phasers.

5. SCom can't escape from you if no more enemies remain 
(without this, chasing SCom can take an eternity).

6. Probe target you enter is now the destination quadrant. Before I don't 
remember what it was, but it was something I had difficulty using.

7. Secret password is now autogenerated.

8. "Plaque" is adjusted for A4 paper :-)

9. Phasers now tells you how much energy needed, but only if the computer 
is alive.

10. Planets are auto-scanned when you enter the quadrant.

11. Mining or using crystals in presense of enemy now yields an attack.
There are other minor adjustments to what yields an attack
and what does not.

12. "freeze" command reverts to "save", most people will understand this
better anyway. (SST2K recognizes both.)

13. Screen-oriented interface, with sensor scans always up.  (SST2K
supports both screen-oriented and TTY modes.)

Eric Raymond's changes:

Mainly, I translated this C code out of FORTRAN into C -- created #defines
for a lot of magic numbers and refactored the heck out of it.

1. "sos" and "call" becomes "mayday", "freeze" and "save" are both good.

2. Status report now indicates when dilithium crystals are on board.

3. Per Dave Matuszek's remarks, Thingy state is never saved across games.

4. Added game option selection so you can play a close (but not bug-for-
bug identical) approximation of older versions.

5. Half the quadrants now have inhabited planets, from which one 
cannot mine dilithium (there will still be the same additional number
of dilithium-bearing planets).  Torpedoing an inhabited world is *bad*.
There is BSD-Trek-like logic for Klingons to attack and enslave 
inhabited worlds, producing more ships (only is skill is 'good' or 
better). (Controlled by OPTION_WORLDS and turned off if game 
type is "plain" or "almy".)

6. User input is now logged so we can do regression testing.

7. More BSD-Trek features: You can now lose if your entire crew
dies in battle.  When abandoning ship in a game with inhabited
worlds enabled, they must have one in the quadrant to beam down
to; otherwise they die in space and this counts heavily against
your score.  Docking at a starbase replenishes your crew.

8. Still more BSD-Trek: we now have a weighted damage table.  Quoth
Eric Allman in the code of BSD-Trek: "Under certain conditions you can
get a critical hit.  This sort of hit damages devices.  The
probability that a given device is damaged depends on the device.
Well protected devices (such as the computer, which is in the core of
the ship and has considerable redundancy) almost never get damaged,
whereas devices which are exposed (such as the warp engines) or which
are particularly delicate (such as the transporter) have a much higher
probability of being damaged."

This is one place where OPTION_PLAIN does not restore the original
behavior, which was equiprobable damage across all devices.  If we
wanted that, we'd return randrange(NDEVICES) and have done with it.
Also, in the original game, DNAVYS and DCOMPTR were the same device.

Instead, we use a table of weights similar to the one from BSD Trek.
BSD doesn't have the shuttle, shield controller, death ray, or probes.
We don't have a cloaking device.  The shuttle got the allocation for
the cloaking device, then we shaved a half-percent off everything to
have some weight to give DSHCTRL/DDRAY/DDSP.

Also, the nav subsystem (enabling automatic course
setting) can be damaged separately from the main computer (which
handles weapons targeting, ETA calculation, and self-destruct).

After these features were added, I translated this program into Python
and added more:

9. A long-range scan is done silently whenever you call CHART; thus
the LRSCAN command is no longer needed.  (Controlled by OPTION_AUTOSCAN
and turned off if game type is "plain" or "almy".)

TESTING:

This code has been designed to be tested.  A simple shellscript
included in the distribution, 'replay', automatically reruns the
last game you played.

Here are some interesting seeds for debugging and regression testing
For each one, I list the seed, the options, and the last svn revision
for which it is known to have given the described behavior.

1160647745 regular short good fancy     (r769)

Starts you in a quadrant with the Super-Commander in it.  Slamming three
torps at him will kill him.  You can use this to regression-test both
torpedo tracks and the Deep Space Probe. Known bug: "probe aut 2 8"
triggers manual navigation.

1160707235 regular short good fancy     (r769)

Do sr/mov aut 5 4/sr/sensors/orbit/transport; you'll get a transporter
failure.

NOTES ON THE PYTHON TRANSLATION:

The Python translation was done with a regexp-based C-to-Python
translator I wrote for the purpose (I expect to release this as a
separate project).  I then hand-tuned and refactored the result.

The LOC count dropped by almost exactly 20% during this process, from
a bit over 8100 lines to a bit over 6500 lines.  If the code is still
shorter than that when you read this, it's because this file contains
most of what used to be a huge header comment.  

SST is not a data-structure- intensive program, so it compresses less
under translation to Python than the 50% drop in LOC I've found to be
more typical.  The gain in readability, though impossible to quantify,
is much greater than the drop in line count would suggest.

Some parts of the code, such as the finish() and score() functions, 
have barely been touched.  Code in the general category of report 
generators has tended to change little, especially since we've tried
to preserve the look and feel of the original.

On the other hand, the vector-arithmetic code around navigation and
torpedos, and deep-space-probe handling changed a lot.  All that 
stuff is now centralized in a 'course' object that hides the
trigonometric calculations.  

The course object builds on a 'coord' object, which I actually had
introduced while refactoring the C version.  Large parts of SST2K are,
perforce, an exercise in 2D vector arithmetic.  In the original
FORTRAN all the vector representation was done with parallel arrays;
in C, I introduced a struct; in Python, the class has a complete
repertoire of vector-algebra operations.

There's one weird archeological detail about the nav code that
deserves a remark.  This program originally required input in terms of
a (clock) direction and distance -- essentially, directions were real
numbers modulo 12 with zero being north.  Somewhere in history, it was
changed to Cartesian coordinates.  But the bearing method still computes
polar coordinates in clockface units -- that's the reason for the
wacky constant 1.90985, inherited straight from the ancestral FORTRAN.
Elsewhere, there were a bunch of computations, now centralized in the
course object, that looked like (15.0 - bearing)*0.5235988; this is a 
conversion from clockface units to radians with zero on the X axis.

As a previous maintainer, probably Tom Almy, observed: Probably
"manual" input should still be done this way -- it's a real pain if
the computer isn't working! Manual mode is still confusing because it
involves giving x and y motions, yet the coordinates are always
displayed y - x, where +y is downward!

Because I think he's arguably right, I haven't ripped out all the
clockface-to-radian conversions.  For this reason, and others, the 
trig code is still a bit wacky and obscure.  Modify with caution
and test thoroughly.