Lunatic is in preview stage. Modifications or TCs may be started, but must not be released. The interfaces (including constants, functions, methods and structure members) described in this document may be assumed to not change significantly until release, but undocumented API may change without further notice.


1. Introduction

2. Language

The base language for writing Lunatic user code is Lua 5.1. It is extensively documented in a Reference Manual, as well as described more broadly and accessibly in the Programming in Lua books.

Because Lunatic is implemented using LuaJIT, a just-in-time compiler for the Lua language, some extensions to the core language are automatically available. They may be used if no compatibility with Rio Lua 5.1 is desired.
[Not all extensions from LuaJIT are available, since some like the FFI are targeted at C programmers rather than scripting coders.]

Note The length operator (#) for table arguments should be taken to be defined by the stricter wording of Lua 5.2.

3. Environment and Usage

Lunatic aims to provide a safe scripting environment for EDuke32. Thus, not all of Lua’s functionality is available in Lunatic, and some is slightly altered to play well with being embedded into a game, or to prevent commonly occurring mistakes.

Differences from default Lua environment
  • Creating new variables in the global environment is forbidden.

  • Referencing a non-existent variable (i.e. one that has value nil) is an error.

EDuke32 can load multiple Lunatic modules when starting up. This is always done after translating CON code to Lua and loading it. Such modules must reside in files named *.lua (case-sensitive) and should be passed directly, without any option letters, at the command line. Directory separators must be forward slashes.

The command-line option -Lopts=strict enables errors on certain conditions that may indicate problems in user Lua or CON code. Currently, strict mode checks for

  • Accesses to sprites not in the game world

  • Indexing the tables returned by require("CON.DEFS") and the like with nonexistent keys. This may indicate a missing file from a bundle belonging together.

Example 1. Invocation example

eduke32 -nologo MYTC.CON -mx addition.con test.lua weapons/nuke.lua -v1 -l1

If the OS environment variable LUNATIC_TIMING_BASEFN is defined, EDuke32 will write out aggregate timing results for actors and events in a comma-separated value format, obtained by suffixing the base name with “.actors.csv” and “.events.csv”.

4. Global Environment

When a Lua module is loaded, its global environment contains both selected functions from core Lua, as well as Lunatic’s own pre-defined variables. These allow access and manipulation of EDuke32 engine and game data.

Caution If an attempt is made to create a new variable in the global environment or to assign any value to an existing variable, the behavior is undefined.

For convenience of debugging from the command line (using the lua OSD command), the global environment contains a table d which is initially empty. It should only be used to temporarily store values of interest in a debugging session and should never be referenced from actual script code.

4.1. Lua functions

The following base Lua functions are available in Lunatic’s global environment:

assert, error, ipairs, pairs, pcall, print, module, next, require, select, tostring, tonumber, type, unpack, xpcall.

The bold ones add functionality or behave slightly differently, described below. Additionally, printf is provided for convenience.

error(message [, level])

In Lunatic, errors also print to the on-screen display (OSD) and are written to the log (unless a maximum error count is exceeded). These also have a backtrace added. Additionally, errors not caught by a pcall result a permanent message to appear on the screen, containing the source file name and line number.


Prints a single message to the OSD and the log. Color codes available in EDuke32 (e.g. ^10 for dark red) are interpreted. Overriding tostring has no effect on Lunatic’s print as it uses the initial, built-in tostring function instead of looking it up in the global environment.

printf(fmt, ...)

Calls print with the result of string.format(fmt, ...).

4.2. Writing and using modules

In Lunatic, like in Lua, a module is a conceptually a sort of package that unites related bits of functionality. Language-wise, it is simply a Lua table holding its contents, as described and motivated at length in Programming in Lua (second edition).

The “creation” and “usage” sides of the modularity concept are reflected in two functions known from Lua 5.1, module and require. The former is a convenient way of separating the import, potential gamevar, and main body sections when writing a package. It is not required to use module, but it is the only way to declare game variables, that is, variables that EDuke32 stores within savegames, recreating their values when they are loaded.

The other side of the coin, require, is also used like in Lua: “including” a named module is requested by passing its name to require. This searches for a matching module, loading and running it if the request happens for the first time.

4.2.1. The function require(modname, ...)

Attempts to find a Lua or Lunatic module named modname. The name can refer to a built-in module, of which the following ones are allowed:

  • The bit module for bitwise operations

  • math, string and table, base modules from Lua

  • Lua’s os module, containing a single function, clock. Like the timing functions, it should only be used to profile bits of code.

  • Modules provided by Lunatic, which are described in their own section.

If modname does not designate a built-in module, Lunatic first replaces every dot contained in it with a directory separator. Then, it looks for a file with that base name suffixed with .lua in the EDuke32 search path (virtual file system, GRP, ZIP). Using directory separators directly is not allowed.

The loaded module is protected so that write accesses to its table yield errors. Unlike in Lua, our require does not return true when a module is requested that has not yet finished loading (that is, the inclusion chain contains a loop). Instead, an error is raised.

Lunatic’s require allows passing additional arguments to the module to load. On the module side, they can be obtained by examining the vararg expression “...” at file scope. Given a definition of args as {...}, its first element args[1] would contain modname and the following entries the values passed in addition to require. This feature is useful for parametrizing a module: for example, the module could provide a way alter the starting tile number of an actor it defines.

Issuing require for some special names listed below has a predefined meaning that cannot be overridden by the user.

  • CON.DEFS: returns a table mapping labels defined from CON to their values. Values pre-defined by the system are not included.

  • CON.ACTION: returns a table mapping labels of actions defined from CON to immutable con.action objects.

  • CON.MOVE: returns a table mapping labels of moves defined from CON to immutable con.move objects.

  • CON.AI: returns a table mapping labels of ais defined from CON to immutable con.ai objects.

  • end_gamevars: used to mark the end of a gamevar block, described below.

4.2.2. The module() function

Initiates a module environment by creating a new empty table and setting it as the global environment of the chunk. Subsequently creating global variables will place them into this “hidden” table. Assuming no subsequent return is placed at file scope, this table will be the one obtained by a require for the module at the client side.

Example file using using module
-- Import section: cache everything the module needs into locals
local print = print
local gameactor = gameactor
local MYDEFS = require("MyEnemyDefs", 1200)  -- see next example

-- After this, the global environment will be swept clean!

-- Our enemy at last
gameactor{ MYDEFS.tilenum, strength=MYDEFS.strength,
        print("My tilenum is "..sprite[aci].picnum.." and I "
              ..(MYDEFS.canfly and "can" or "can't").." fly")

Unlike Lua 5.1, our module takes neither a name argument nor “option” varargs. A Lunatic file may have at most one call to module, which (if there is one) must be called at file scope.

If the Lua file doesn’t need to register any game variables, it is also possible to return its table directly instead of using module. However, due to the way modules are loaded, a trailing return statement must be wrapped in a doend block, like this:

Example module MyEnemyDefs.lua explicitly returning a table
local args = { ... }
do return {
   tilenum = args[1] or 1000,
   strength = 100,
   canfly = true,
Game variables

Lunatic has a special mechanism to mark variables that represent persistent state and whose values should be stored in savegames. If such variables are desired, they must be initialized between the module call in a Lua file and a closing require("end_gamevars").
[The reason that the initialization has to happen between the module and the require('end_gamevars') is that on savegame loading, gamevars are restored from the latter.]
These variables may also be local.

Game variables may take on only values of types that Lunatic knows how to serialize into savegames. These are the following:

  • Booleans, numbers, and strings, collectively called the basic types

  • Custom Lunatic types that are labeled serializable in their documentation

  • Tables, but with the following restrictions on their contents:

    • A table key may only be of basic type.

    • A table value may be (a reference to) any serializable object, but tables or Lunatic objects that are so referenced must have originated in the gamevar section of the same module. Beyond that, there are no restrictions on the table topology.

Caution If a gamevar contains a value that is not serializable at any point in the execution, the behavior is undefined. Note that in particular, gamevars are not allowed to take on the value nil.

4.3. The gv variable

Some constants, global C variables, and miscellaneous functions and structures are accessible via the global gv variable.

4.3.1. Constants


The hard engine limits on the number of sectors, walls, and sprites. These constants must be used instead of any literal numeric values because they can change depending on how EDuke32 was configured and built.


Constants permissible to the sectorsofbunch iterator.


A clipping (collision detection) mask specifying to consider only blocking walls and sprites.


A clipping (collision detection) mask specifying to consider only hitscan sensitive walls and sprites. The set bits of CLIPMASK1 and CLIPMASK0 are disjoint.


A mapping of names to values representing rendering modes: CLASSIC, POLYMOST, POLYMER.


The number of times in a second each actor executes its code and updates its position (“game tics”).




The maximum permissible inventory index plus one.




The maximum permissible weapon index plus one.


A constant indicating which program Lua is embedded into. It can be compared for (in)equality with:



    (Mapster32 supports a subset of Lunatic that is not documented and subject to change at any time.)

4.3.2. Variables

gv.numsectors (read-only)

The total number of sectors in the currently loaded map.

gv.numwalls (read-only)

The total number of walls in the currently loaded map.

gv.totalclock (read-only)

The current value of the engine timer that increments at a rate of 120 per second under default settings. (Thus, one game tic corresponds to four totalclock increments.) When staying within one “mode” such as in-menu or in-game, it is guaranteed to not decrease. However, going from one mode to another may produce discontinuities.

gv.gametic (read-only)

The number of game tics that have elapsed since starting the current level. This value is guaranteed to not decrease during a game, and is restored from savegames.

gv.screenpeek (read-only)

The player index of the player from whose position the scene is being displayed.

gv.rendmode (read-only)

The current rendering mode as a value that can be compared against those in gv.REND.


A structure containing information about the currently displayed HUD weapon. Contains the following members, which are set from C before entering EVENT_DISPLAYWEAPONS: cur, count, gunposx, gunposy, lookhalfang, lookhoriz, shade.


A structure that, prior to entering EVENT_DISPLAYROOMS, is populated with the position and orientation of the “camera” from which the scene would be drawn. Contains the following members: pos, dist, clock, ang, horiz, sect.


A special variable that is used by the game to pass specific values to game events, or to examine values passed back from events and act upon them accordingly. Refer to Appendix B for a list of how the various events interpret this variable.

4.3.3. Functions

gv.getangle(x, y)

Returns an approximation of the angle between the line segments (0,0)→(1,0) and (0,0)→(x,y) in BUILD angle units in the range [0 .. 2047].


Returns one value from the global engine-side pseudo-random number generator in the integer range [0 .. 65535].

gv.getticks(), gv.gethiticks()

Each of these functions return a number that increases at a rate of 1 per millisecond. Their only intended application is to profile bits of code; they should not be used to control the game world. The two functions differ in their precision: getticks() always returns integral values, while the result of gethiticks() also has an unspecified precision in the fractional part. (It can be expected to give a time precision of at least one microsecond.)

gv.doQuake(gametics [, snd])

Requests from the game to perform the “quake” effect that shakes the screen etc. for the next gametics game tics. If a sound index snd is passed, also start playing that sound.


Returns the one-based number of the episode currently being played.


Returns the one-based number of the level currently being played.

4.4. Lunatic structures

The primary means of effecting game state in Lunatic is via composite variables defined in the global environment. These provide direct, but restricted, access to C structure arrays of the EDuke32 engine or game.

Caution If an attempt is made to access any composite variable outside of event or actor code, the behavior is undefined.

Composite variables can be used in various ways. All of them allow indexing with an integer value from 0 to some maximum (sometimes the size of the array minus one, but occasionally less). For example, the code snippet

local sec = sector[0]

gets a reference to the first sector of the loaded map into the local sec. This reference can then be used to both read and write its members.

Various structures also provide methods in Lunatic to modify their state, usable with Lua’s v:func(args...) syntax. Building on the previous example,

local cz = sec:ceilingzat(wall[sec.wallptr])

would get into cz the ceiling z position at the first sector’s first wall-point.

Finally, some composite variables offer static data, which can contain functions or tables of constants. These are accessed using the dot notation on the composite variable, not its constituents. For instance, the following can be used to change the sector number of the sprite with index i manually:

sprite.changesect(i, sectnum)

4.4.1. Type of structure members

In the following, some structure members will be annotated with their integer type, for example i16 or u8. The letter i denotes a signed integer whereas a u designates an unsigned one. The number following that letter indicates the bit width of that integer.

Representable values
  • A member of signed integer type and bit width B can contain any whole number from –2B–1 to 2B–1–1.

  • A member of unsigned integer type and bit width B can contain any whole number from 0 to 2B–1.

  1. If an assignment to a member having signed integer type is made, the “right-hand side” value must be a number in the closed interval [–231 .. 231–1].

  2. If an assignment to a member having unsigned integer type and bit width B is made, the “right-hand side” value must be in the closed interval [–231 .. 231–1] if B is less than 32, or in [0 .. 232–1] otherwise.

  3. If the appropriate requirements hold, an assignment from a Lua number to a member having integer type begins by discarding the fractional part (“truncation”). Otherwise, the behavior is undefined.

  4. If the truncated value is outside the range of representable values for the corresponding integer type of bit width B, the final value is obtained by successively adding or subtracting 2B, until the value falls inside that range.

  1. Assignments to u8 member visibility

    • sec.visibility=3.94159 results the member to contain the integer 3

    • sec.visibility=1/0 is undefined (attempt to assign an infinity)

    • sec.visibility=-1 results the member to contain the integer 255

  2. Assignments to i16 member lotag

    • sec.lotag=32768 results the member to contain -32768

    • sec.lotag=2^32 is undefined

4.4.2. Bit fields

Some structures contain members that represent a collection of flags, for example sprite[].cstat or actor[].flags. These flags can be toggled by setting or clearing their respective bits (integer numbers that are powers of two).

For convenience, Lunatic provides alternative names for some of these members, together with methods to examine or modify any number of bits of such a member in one expression. Whenever there is such an alternative name available, it is declared as having type bitfield in the listings of the structure members.

methods of the bitfield type

Sets (toggles to an “on” state in a boolean sense) those bits of bf that are set in bits.


Clears (toggles to an “off” state in a boolean sense) those bits of bf that are set in bits.


Flips those bits of bf that are set in bits, that is, reverses their boolean state.


Returns a boolean that indicates whether bf has any of the bits set in bits set.


Returns a number containing the bits of bf bitwise ANDed with those in bits.

Example 2. Examples

After the lines setting sprite i to 33% translucent and blocking,

local CS = sprite.CSTAT
local spr = sprite[i]
spr.cstat = CS.TRANS1 + CS.BLOCK

one could proceed as follows for the sake of example:

spr.cstatbits:set(CS.TRANS2)  -- make the sprite 66% translucent now
spr.cstatbits:flip(CS.BLOCK + CS.HITSCAN)  -- make it hitscan-sensitive and non-blocking
local isflipped = spr.cstatbits:test(CS.FLIP_BITMASK)  -- is it x- or y-flipped? (no)

4.4.3. Engine-side composites

The composite variables described in this subsection provide access to engine-side structures. The first three, sector, wall, and sprite, are part of a BUILD map saved to disk. The other ones only exist when running the game.


Accessible from 0 to gv.numsectors-1. Each element has the following members:

wallptr, wallnum (read-only)

The index of the sector’s first wall and the number of walls in the sector, respectively.

u8 visibility

Determines the amount of distance fading. The sector visibility member is biased: linear visibility is determined from it by adding 16 and taking the result mod 16. This linear visibility’s interpretation is that larger values correspond to a steeper “darkening” (or fogging) attenuation with distance. Linear visibility 0 — corresponding to sector[].visibility of 240 — means no attenuation with distance.

u8 fogpal

In the OpenGL modes that don’t use per-pixel shade table lookups, setting fogpal to a non-zero value overrides the fog color of the sector to that of the corresponding pal number. These are either expected to have been defined using DEF tokens fogpal or makepalookup, or created by EDuke32 at startup.

i16 lotag, hitag, extra

General-purpose “tags” provided for game programming. They may be used by various EDuke32 sector effects, so it is not recommended to use them in scripting code.

In addition to the members described above, each sector has two sets of members for its ceiling and floor. A sector reference can be indexed with the strings ceiling or floor to get references to the respective “parts”, or one can access the consituent members by prefixing ceiling or floor to the base member names given below.

Example 3. Different ways of accessing the same member

After the code lines

local sec = sector[0]
local ceil = sec.ceiling

the following expressions all denote the same location, both if read or written to: sec.ceilingheinum, ceil.heinum, sector[0].ceiling.heinum, sector[0].ceilingheinum.

In the following, cf will stand for a ceiling or floor reference, while sec will label a sector reference.

cf.picnum (read-only)

The tile number of the ceiling or floor.

u16 cf.stat,   bitfield cf.statbits

A bit field holding various flags about how the ceiling or floor should be displayed, how collision detection should be handled, etc. The sector.STAT object should be used to obtain the values for applicable flags.

i16 cf.heinum

If cf.stat has bit sector.STAT.SLOPE set, the tangent of the slope angle, multiplied by 4096. Positive values make the ceiling or floor slope towards the floor, negative ones slope upward.

i32 cf.z

The BUILD z coordinate (scaled by 16 compared to the x and y directions) of the pivoting line of the ceiling or floor.

cf.bunch (read-only)

The “bunch” number of the ceiling or floor used for True Room over Room. One bunch comprises N ceilings and M floors (N ≥ 1, M ≥ 1) such that each set covers the same planar, connected area.

i8 cf.shade

The shade of the ceiling or floor. Larger values mean a more darker appearance.

u8 cf.pal

The “palette swap” index of the ceiling or floor.

u8 cf.xpanning, cf.ypanning

The panning values of the ceiling or floor. One full cycle is covered by values from 0 to 255.

ceiling-or-floor methods

Set the tile number of the ceiling-or-floor cf.

sector methods
sec:set_ceilingpicnum(tilenum),   sec:set_floorpicnum(tilenum)

Set the tile number of the ceiling or the floor.

sec:ceilingzat(pos),   sec:floorzat(pos)

Return the z coordinate of sector sec's ceiling or floor at position pos, which can be anything indexable with the strings x and y.

sec:zrangeat(pos, walldist, clipmask)hit

Given a starting point pos assumed to be contained in the sector, calculate the z coordinates of the objects that would be first hit by a quadratic, floor-aligned sprite pointing parallel to the grid and having side length 2*walldist, when travelling in a straight line up- and downwards.

The argument clipmask is a number specifying which objects should be checked for collision: its least significant 16 bits are bitwise-ANDed with wall[].cstat values, while the high 16 bits are ANDed with sprite[].cstat. Whenever the so masked values are non-zero, the objects are considered for collision.

The method returns an immutable structure hit, containing the sub-structures c and f (for movement upwards and downwards, respectively) with the following members:

  • spritep: a boolean signifying whether a sprite was hit

  • num: if spritep is true, the index of the hit sprite, otherwise the index of the hit ceiling/floor’s sector

  • z: the z coordinate of where the sprite would end up on collision

sector static data

Provides a mapping of symbolic names to values applicable to cf.stat. These name single bits: PARALLAX, SLOPE, SWAPXY, SMOOSH, FLIPX, FLIPY, RELATIVE, MASK, TRANS1, TRANS2, BLOCK, HITSCAN, while the following denote bit masks: FLIP_BITMASK, ORIENT_BITMASK, TRANS_BITMASK.


Contains flags permissible to updatesector and other sector updating functions. BREADTH, Z.


Accessible from 0 to gv.numwalls-1. Each element has the following members:

i32 x, y

The 2D coordinates or this wall point. Should not be set directly.

z (read-only)

Always yields 0. The primary purpose of this field is to make wall references permissible as arguments to xmath vector operations.

point2 (read-only)

The index of the second wall point.

nextwall, nextsector (read-only)

If the wall is “white”, these members equal -1. For “red” walls, they contain the wall and sector indices (respectively) of the wall on the other side.

upwall, dnwall (read-only)

For walls constrained by TROR extension, the upper and lower neighbor walls, respectively. Any of them may be -1, meaning that the wall is not attached to a neighbor in this direction.

u16 cstat,   bitfield cstatbits

A bit field holding various flags about how the wall should be displayed, how collision detection should be handled, etc. The wall.CSTAT object should be used to obtain the values for applicable flags.

picnum (read-only)

The tile number of the non-masked portion of the wall. If wall.CSTAT.BOTTOMSWAP is set on this wall’s .cstat, it is only displayed in the upper portion. (The lower portion takes it from this wall’s .nextwall then; this will be labeled use-other-bottom in the following.)

overpicnum (read-only)

The tile number of the masked portion of the wall, i.e. that which is drawn if this wall’s .cstat has bit wall.CSTAT.MASK or wall.CSTAT.ONEWAY set.

i8 shade (use-other-bottom)

The shade of the wall for both non-masked and masked portions. Larger values mean a more darker appearance.

u8 pal (use-other-bottom)

The “palette swap” index of the wall.

u8 xrepeat, yrepeat

Values that are proportional to the number of times that the wall’s texture repeats in each direction per given wall length/height. A value of 8 renders 64 texels across a length of 1024 x/y map units or a height of 16384 z units.

u8 xpanning, ypanning (use-other-bottom)

The panning values of both masked and non-masked portions of the wall. One full cycle is covered by values from 0 to 255.

i16 lotag, hitag, extra

General-purpose “tags” provided for game programming. They may be used by various EDuke32 effects internally, so it is advised to do some research before claiming them for oneself.

wall methods
wal:set_picnum(tilenum), wal:set_overpicnum(tilenum)

Set the tile number of the wall or its masked portion.

wall static functions
wall.dragto(i, pos)

Set the position of the point of the wall with index i to pos, which can be anything indexable with x and y. This function is the preferred way of changing wall coordinates, since it takes care to reposition dependent wall points, too.

wall static data

Provides a mapping of symbolic names to values applicable to wall[i].cstat. These name single bits: BLOCK, BOTTOMSWAP, ALIGNBOTTOM, FLIPX, MASK, ONEWAY, HITSCAN, TRANS1, FLIPY, TRANS2, while the following denote bit masks: FLIP_BITMASK, TRANS_BITMASK.


The sprite composite is accessible with indices from 0 to gv.MAXSPRITES-1, but accesses to sprites that do not exist in the game world have no meaning. Each element has the following members:

i32 x, y, z

The BUILD coordinates of the sprite. It is not advisable to set these directly.

u16 cstat,   bitfield cstatbits

A bit field holding various flags about how the sprite should be displayed, how collision detection should be handled, etc. The sprite.CSTAT object should be used to obtain the values for applicable flags.

picnum (read-only)

The tile number of the sprite, also used to determine which actor code is run if this sprite has a statnum of actor.STAT.ACTOR.

i8 shade

The shade of the sprite. This may not be the shade that this sprite is ultimately drawn with, though.

u8 pal

The “palette swap” index of the sprite. This may not be the palette swap that this sprite is ultimately drawn with, though.

u8 blend

The blending table index of the sprite. See Blending table interfaces for more details.

u8 clipdist

In the engine, this member is used in the clipmove() function: it controls the distance at which another moving object is considered to be in collision with this view-aligned and stationary sprite.
[The clipmove() function is called for the moving object. The stationary one is a candidate for collision out of potentially many.]
It does not control the inverse case: the distance at which the moving object is considered in collision with the stationary one is determined by the walldist argument to clipmove().
More precisely, it designates half the side-length of the bounding square divided by 4. Thus, a value of 255 keeps moving objects away from this one at a max-norm distance
[The max-norm distance between points p1=(x1, y1) and p2=(x2, y2) is defined as max(abs(x2 — x1), abs(y2 — y1)).]
of at least 1020 BUILD x/y units.

In the Duke3D game code, it is also used to determine said walldist argument for certain clipmove() calls, i.e. to control (1/8th of) the base side length of the moving sprite’s clipping cuboid: this is the case for

  • user actors declared as actor.FLAGS.enemy or actor.FLAGS.enemystayput in the Lua gameactor call (or which were provided the corresponsing flags for CON’s useractor) having an xrepeat less than or equal to 60, or

  • non-enemies that are either not projectiles, or projectiles that have a certain projectile bit set (TODO; these are not yet documented.)

u8 xrepeat, yrepeat

The size of the sprite in each dimension. For wall- and floor- aligned sprites, a value of 64 means a width of 16 x/y BUILD units or a height of 256 z BUILD units per texel.

sectnum (read-only)

The index of the sector that this sprite is currently contained in.

statnum (read-only)

The current status number of this sprite. Applicable values are contained in actor.STAT.

owner (read-only)

The index of the sprite from which this sprite was spawned. If this sprite is not a “child” of another one, then owner is the index of this sprite itself.

i16 xvel, zvel

For actors and other moving sprite kinds, the horizontal and vertical components of the current velocity. See the description of con.move for more details.

i16 yvel (read-only)

A member used for different purposes by Duke3D. Setting it is only allowed through the set_yvel method.

i16 lotag, hitag, extra

General-purpose “tags” provided for game programming. They may be used by hard-coded actors internally, so it is advised to do some research before claiming them for oneself.

sprite methods

Sets the tile number of sprite spr to tilenum. It is disallowed to issue set_picnum on an APLAYER sprite or change the tile number of a sprite to APLAYER.


Sets the yvel member of sprite spr to yvel. It is disallowed to issue set_yvel on an APLAYER sprite. Further restrictions are reserved.

spr:getheightofs()height, offset

Returns the height and z offset of sprite spr in BUILD z units. Adding the offset to spr.z yields the z coordinate at the bottom of the sprite. Subtracting from that the height results the z coordinate at its top. However, the per-tile z offset is not taken into account.

spr:setpos(pos [, newsect])

Unconditionally sets the position of spr to pos, which can be anything indexable with x, y and z. Thus, in effect a shorthand for

spr.x, spr.y, spr.z = pos.x, pos.y, pos.z

If newsect is passed, additionally calls spr:changesect(newsect).
Returns spr.


An alternative way to call sprite.changesect(index_of_spr, sectnum), where index_of_spr is the sprite index corresponding to spr. This method is provided for convenience, but may be slower than the static function changesect.


An alternative way to call sprite.updatesect(index_of_spr, flags), where index_of_spr is the sprite index corresponding to spr. This method is provided for convenience, but may be slower than the static function updatesect.



sprite iterators
for i in sprite.all()

Iterates over all sprites currently in the game world.

sprite static functions
sprite.changestat(i, statnum)

Allows to manually change the status number of the sprite with index i to statnum. If i is an invalid sprite index or the index of a sprite not in the game world, or statnum is an invalid status number, an error is thrown.

sprite.changesect(i, sectnum)

Allows to manually change the sector number of the sprite with index i to sectnum. If i is an invalid sprite index or the index of a sprite not in the game world, or sectnum is an invalid sector index, an error is thrown.

sprite.updatesect(i [, flags])

Updates the sector number of the sprite with index i, in effect setting sprite[i]'s sector number to the result of

updatesector(sprite[i].pos, sprite[i].sectnum, flags)

If the updatesector call returns -1, the sprite’s sector number remains unchanged.
Returns the result of the updatesector call.

sprite overridden operators

Returns an xmath.ivec3 object that contains the position of this sprite, diminished by zofs in the z direction. Because in BUILD, z coordinates increase toward the floor, the ^ can be thought of as the position of the sprite “raised by zofs units”.

sprite static data

Provides a mapping of symbolic names to values applicable to sprite[i].cstat. These name single bits: BLOCK, TRANS1, XFLIP, YFLIP, ALIGNWALL, ALIGNFLOOR, ONESIDE, CENTER, HITSCAN, TRANS2, INVISIBLE, while the following denote bit masks: ALIGN_BITMASK, TRANS_BITMASK.


Accessible with the same indices and with the same restrictions as sprite.

4.4.4. Game-side composites


The actor composite holds various run-time data about a sprite. Like sprite, it is accessible with indices from 0 to gv.MAXSPRITES-1, but accesses to actors that do not exist in the game world have no meaning. Each element has the following members:

u16 movflags,   bitfield movflagsbits

The actor’s current movement flags.

actor methods

The following methods query or set properties related to actor behavior.


Sets the action of actor a to act, which may be either an object returned by con.action or a number 0 or 1. Resets the actor’s action count and current frame offset to 0.


Returns a boolean of whether the current action of actor a is act, which can be any value permissible to a:set_action(). For composite action objects, equality is established using its hidden ID, not the public members. Refer to Actions, moves and AIs for further details.


Overrides the delay of the current action of actor a without changing the action’s ID.


Sets the actor’s execution count to count. The execution count of an actor increments after each time its callback function has run once.


Returns the actor’s execution count.


Resets the action count of the actor to 0. The action count is incremented on each frame advance of the actor’s current action. Also see the delay argument to con.action.


Returns the actor’s action count.

a:set_move(mov [, movflags])

Sets the move of actor a to mov, which may be either an object returned by con.move or a number 0 or 1. See the move argument to gameactor for further details. If movflags is omitted, it defaults to 0.

The set_move method resets the actor’s execution count. Also, if moveflags has bit actor.MOVFLAGS.randomangle set and the actor is not an enemy or a live enemy, its sprite[].ang value is set to a “random” value using gv.krand.


Returns a boolean of whether the current move of actor a is mov, which can be any value permissible to a:set_move(). Like with a:has_action(), equality is established using the move ID in case a composite one is passed.


Overrides the horizontal velocity of the actor’s current move to hvel without changing its ID.


Overrides the vertical velocity of the actor’s current move to vvel without changing its ID.


Sets the AI of actor a to ai, which must be an object returned by con.ai. In addition to setting the current AI ID of the actor, a:set_ai(ai) is equivalent to the sequence

a:set_move(ai.mov, ai.movflags)

Returns a boolean of whether the current AI of actor a is ai.

Various methods query whether the last movement step of the actor made it collide with another object, or allow getting this object’s index then.


Returns a boolean of whether the actor hit any object in the world, including ceilings and floors.


Returns a boolean of whether the actor bumped into another actor or a wall.


If the actor hit a wall with the last movement, returns that wall’s index. Otherwise, returns nil.


If the actor hit a sprite with the last movement, returns that sprite’s index. Otherwise, returns nil.

actor static functions

Causes the actor with index i to fall in a “hard-coded”, not further specified fashion.

actor.move(i, vec, cliptype [, clipdist])


actor static data

Provides a mapping of symbolic names to values applicable as sprite status numbers.


Contains symbolic names of values applicable to gameactor's flags input argument, most of which are described there.


Contains symbolic names of values applicable gameactor's movflags input argument, actor[]:set_move(), and the like.

  • faceplayer, geth, getv, randomangle, faceplayerslow, spin, faceplayersmart, fleeenemy, seekplayer, furthestdir, dodgebullet, all naming single bits

  • jumptoplayer_bits: the bitwise-OR of two bits, one of which is faceplayer

  • jumptoplayer_only: the same as jumptoplayer_bits, but without the faceplayer bit set

  • jumptoplayer: deprecated


Accessible with the index 0 and any nonnegative pli index passed to a game actor or event callback function.

Arrays with named elements

The structures obtained by indexing player contain arrays whose elements may be referred to either by index of by a particular name. Thus, the array can be thought to be overlaid with a structure containing as many members as the array has elements, all having the element’s type.

Currently, there are with two kinds of names: weapon names and inventory names. They will be marked using notation like this:

i16 ammo_amount[weapon]

means that the structure member ammo_amount is an array with named elements, themselves being of signed 16-bit integer type. The array can be indexed with valid weapon numbers, or weapon names. In the following (constructed) example, the first player’s current pistol ammo count is set to that of the currently selected weapon:

local ps = player[0]
ps.ammo_amount.PISTOL = ps.ammo_amount[ps.curr_weapon]

The inventory names are the same as the keys of gv.GET, those for the weapons coincide with the keys of gv.WEAPON.

Boolean members

Player members marked bool in the listing below yield Lua true or false on reading and expect one of these values when assigned to.

player members
xmath.ivec3 vel

The vector of the player’s current velocity, accounted each time a player’s input is processed. The x and y components are in BUILD x/y units per game tic, scaled by 214. The z component is given in BUILD z units per game tic. Processing a player’s input and other EDuke32 game code may change vel “behind a user’s back” in a further unspecified fashion.

i32 runspeed

The factor, scaled by 216, with which the player’s current horizontal velocity is multiplied before calculating its new position. On various occasions, the ultimate scaled factor is the difference of runspeed and at most 8192.


The index of the player’s currently selected weapon.

u16 gotweapon,   bitfield gotweaponbits

Indicates whether each weapon is in the possession of the player. If bit 2w is set for a weapon index w, the player has got this weapon. The player methods has_weapon, give_weapon and take_weapon can be used to query or modify this member.

i16 ammo_amount[weapon]

The current ammo amount for each weapon.

i16 max_ammo_amount[weapon]

The maximum ammo amount for each weapon. Because ammo_amount amount and max_ammo_amount are writable without restriction, it is the user’s responsibility to make sure that the former never exceeds the latter and that both are non-negative at all times. Otherwise, erratic behavior may ensue.

bool jetpack_on,   scuba_on,   heat_on

Whether the player currently has the jetpack, scuba gear, or night vision goggles activated, respectively.

weapondata_t weapon[weapon]

A struct containing information about the behavior each of weapon for this player. In CON, these are available as separate gamevars named WEAPONx_* for weapon indices x and members *.

weapondata_t members

In Lunatic, some members of the weapondata_t structure are checked when being assigned to and issue an error on attempting to set them to an invalid value. All these members assume zero to mean the neutral/no-op value (instead of –1, as would seem more logical), and consequently, Lunatic only allows values greater or equal zero to be assigned to them.

  • clip

  • reload

  • firedelay

  • totaltime

  • holddelay

  • flags

  • shoots (checked)

  • spawntime

  • spawn (checked)

  • shotsperburst

  • initialsound (checked)

  • firesound (checked)

  • sound2time // This is a time number, not a sound number

  • sound2sound (checked)

  • reloadsound1 (checked)

  • reloadsound2 (checked)

  • selectsound (checked)

  • flashcolor

player methods

Returns a boolean of whether player ps has got the weapon with index widx.


Adds the weapon given by index widx to player ps's possession without changing the currently held one.


Removes the weapon given by index widx from player ps's possession. If an attempt is made to remove the currently selected weapon, the behavior is undefined.

ps:fadecol(fadefrac, r, g, b [, speed [, prio]])

Initiates a tinting that linearly fades over time and is blended with the whole screen contents whenever player ps's view is displayed.
[The behavior is unspecified should more than one player’s view be displayed at one time.]
The first argument fadefrac specifies the starting blending coefficient; r, g and b specify the intensities of the red, green and blue color components, respectively.

Both fadefrac and the component intensities are first clamped to the range [0.0 .. 1.0]. The resulting values are interpreted as a proportion, 0.0 meaning no blending/no color and 1.0 meaning full blending/full color.
[Currently, for implementation reasons, a value of 1.0 results in only almost full blending or presence of the specified color component.]

The fourth, optional argument speed controls the rate at which the tinting diminishes. At a value of 1 (the default), a tint with a fadefrac of 0.5 finishes in approximately one second.

The last, optional argument prio must be an integer in the range [-128 .. 127], the default being 0. When a fadecol is issued in the presence of another tint fading in progress, and the prio given by the arriving fadecol is greater or equal than the prio of the ongoing one, the latter is canceled and the arriving fading is initiated in its place. (There is no support for tint fades that overlap in time.)

Caution If Lunatic code that uses fadecol is loaded together with CON code that writes to the player’s pals members directly at any point, the behavior is undefined.
player iterators
for i in player.all()

Iterates over the indices of all active players.


An array of size gv.MAXTILES. Currently, access with numeric indices by the user is disallowed for g_tile and its elements have no public members.

g_tile static data
g_tile.sizx,   g_tile.sizy

Arrays indexable with tile numbers [0 .. gv.MAXTILES-1] that hold the horizontal and vertical texel sizes of each tile.

4.5. Lunatic functions

4.5.1. Engine-side iterators

for w in wallsofsect(sectnum)

Iterates over the indices of all walls of the sector with index sectnum.

for s in spritesofstat(statnum [, maydelete])

Iterates over the indices of all sprites with status number statnum. If maydelete is omitted or false, there must be no deletion of any sprite while the loop is active. If maydelete is true, deleting sprites inside the loop is allowed. Inserting sprites is always allowed.

for s in spritesofsect(sectnum [, maydelete])

Iterates over the indices of all sprites contained in the sector with index sectnum with the same meaning for maydelete as with spritesofstat.

for s in sectorsofbunch(bunchnum, cf)

Iterates over the indices of the sectors whose ceiling’s or floor’s bunch equals bunchnum, selected by passing gv.CEILING or gv.FLOOR (respectively) to cf.

for s, what in sectorsofbunch(bunchnum, gv.BOTH_CF)

Iterates over the indices of the sectors whose ceiling’s and floor’s bunch equals bunchnum. On each iteration, what is one of the strings 'ceiling' or 'floor', denoting whether it’s the ceiling or floor of sector s that has the given bunch number.

4.5.2. Sector containment functions

inside(pos, sectnum)

Returns a boolean of whether the position pos (which can be anything indexable with x and y) is considered inside the sector with number sectnum, which must be valid. The z component is not taken into account.

Sector updating

A number of engine functions take a position pos and a starting sector number sectnum and try to find a sector that contains pos, assuming it started out in sectnum.
[Note that this is different from CON’s updatesector, which takes the starting sector to be the one of the current sprite.]

If a valid sector numeric is passed for sectnum, these functions first check whether that sector already contains pos (i.e. the position stays in the same sector) and then attempt to search neighboring sectors. Unless breadth-first search is requested (see below), if the passed sectnum is -1, all sectors are searched in an unspecified order.

On success, these functions return the sector number of the “updated” sector, otherwise -1.

updatesector(pos, sectnum [, flags])

Searches for a sector containing pos, which can be anything indexable with x and y. Thus, the z component is not taken into account. If sectnum is a valid sector number, first all its neighboring walls (wall[].nextwall, “red wall links”) are searched, falling back to a linear search on failure.

However, this strategy can be problematic if the position passes multiple thin sectors that overlap another, potentially unrelated sector (for example, in a TROR or sector-over-sector construction). If such a sector is passed without an updatesector call ever made when it contains the position, the next call may not find the real sector (as it is now two nextwall links away from the initial one), and wrongly suggest the overlapping one instead.

For this reason, the optional argument flags accepts sector.UPDATE_FLAGS.BREADTH, instructing updatesector to look for matching sectors in a breadth-first search starting from sectnum, and following all nextwall links. With this strategy, there is no fallback to a linear search if no matching sector is found.

Passing sector.UPDATE_FLAGS.Z to flags makes the sector updating behave as if updatesectorz was called. This is mainly useful if updatesector is called via the sector updating functions of sprites.

updatesectorz(pos, sectnum)

Searches for a sector containing pos, which can be any value indexable with x, y and z. Thus, it additionally takes the z component into account by checking against the bounds that would be returned using a sector’s ceilingzat/floorzat methods.

The updatesectorz function first checks the initial sector for containment of pos, then it tries any TROR neighbors of sectnum. Finally, it proceeds like updatesector as far as the searching order is concerned.

hitscan(pos, sectnum, ray, clipmask)

Starting from the position pos (which is assumed to be contained in sectnum), the hitscan function determines the object that would be first hit by a ray emanating from pos into the direction given by ray. Both pos and ray may be any object indexable with x, y and z, but the components are converted to signed 32-bit integers prior to being passed to the actual engine function. Note that ray is interpreted in BUILD scaling: the z component has 16 times the precision for a given game-world length compared to x or y.

The clipmask argument determines what objects are considered being hittable by the ray and is expected to be an integral number. It is interpreted as two separate bit masks: the low 16 bits for walls and the high 16 bits for sprites. Each time there is a potential collision, the respective mask is ANDed with the cstat member of the object, and if the result is non-zero, the ray is considered having hit the object.

The hitscan function returns an object with information about the hit object (if any) as well as the position of its intersection with the ray. It contains the following members:

  • sector: The sector number of the hit object, or -1 if no object was hit. Thus, testing it for being greater or equal to zero is a quick way of finding out whether any object (ceiling/floor, wall, or sprite) was hit at all.
    [It is recommended to carry out this check for the sake of cautiousness: while proper hitscan invocations should always hit something, the function may come up empty in certain corner cases (such as a starting position outside of the designated sector).]

  • wall: If a wall was hit, its index. Otherwise, -1.

  • sprite: If a sprite was hit, its index. Otherwise, -1.

  • pos: The position that is the intersection of the emanated ray with the hit object, indexable with x, y and z. Undefined if no object was hit.

4.5.3. Customizing the game

In Lunatic, there are two main ways of customizing the game’s behavior. Both involve the user providing functions that are called back at certain points in the execution of EDuke32. Game actors are simply sprites that run a particular piece of code based on their tile number each game tic, unless they are in a “dormant” state. Game events are invoked at predetermined points in the program flow.

To register custom actor and event code with the game (and define additional actor information), Lunatic provides two functions in the global environment, gameactor and gameevent. As their sole argument, they take a table containing the appropriate data.

The function gameactor{tilenum [, ...], func}

Registers custom code for the actor given by tile number tilenum. For each non-sleeping actor, the function func is called every game tic with three input arguments: func(aci, pli, dist).

  • aci: the sprite number of the actor invoking func

  • pli: the index of the player that is nearest to this actor

  • dist: the 3D Manhattan distance
    [The Manhattan distance between points p1=(x1, y1, z1) and p2=(x2, y2, z2) is defined as abs(x2 — x1) + abs(y2 — y1) + abs(z2 — z1).]
    between actor aci and player pli

Additionally, gameactor accepts optional input arguments. They can be specified positionally by following tilenum, or be given as values to string keys of the argument table. Each such input argument may be provided in at most one of these two forms. Furthermore, func may be provided as value to the key 'func' as well.

[2] flags

A number that controls both certain aspects of the gameactor call as well as the run-time behavior of the actor itself. A couple of bits for the latter type are listed in actor.FLAGS, abbreviated AF in the following.

These values describe the “type” of the actor: AF.enemy, AF.enemystayput and AF.rotfixed. Except for enemystayput, they name single bits (enemystayput implies enemy).

In Lunatic, game actors can be chained, that is, a callback function can be either appended to the end of an already registered one, or prefixed at its front. In this case, a previous gameactor call must have taken place for that actor (this may have happened from a CON useractor block, which gets translated to gameactor). Moreover, this mechanism allows to add run-time flags to the actor in question.

Chaining two callback functions is achieved by creating a new one that calls the first one, followed by a tail call of the second one. This has certain implications if control is transferred non-locally, for example by using con.longjmp.

Several flags in AF are provided to control how a gameactor invocation handles chaining. In all cases, the actor tile flags are bitwise-ORed with the existing ones.

  • AF.replace: Replace the callback function. This is the way CON’s useractor behaves and is also the Lunatic default.

  • AF.replace_soft: deprecated alias for AF.replace

  • AF.replace_hard: deprecated

  • AF.chain_beg: Prepend the provided func to the existing callback function.

  • AF.chain_end: Append the provided func to the existing callback function.

    The following members all default to 0 if omitted.

[3] strength

The initial strength or health of the actor.

[4] action

The initial action of the actor. May be either an object returned by con.action, or the numbers 0 or 1. Both represent actions with that ID, but all public members set to 0.

[5] move

The initial move of the actor. May be either an object returned by con.move, or the numbers 0 or 1. Both represent moves with that ID, but all public members set to 0. A move of 0 disables the usual actor movement, even if its hvel or vvel subsequently get overridden (and the corresponding movflags set).

[6] movflags

The actor’s initial movement flags. Applicable bits are available in the actor.MOVFLAGS object.

The function gameevent{evtlabel [, flags], func}

Registers custom code to be run at specific points in the program. The first argument evtlabel should be a string naming the event. A complete list of events can be found at the EDuke32 wiki. The label may be stripped of the leading “EVENT_”, so that e.g. EVENT_JUMP and simply JUMP denote the same event.

The arguments flags and func can alternatively be passed as values to the same-named keys of the input argument table to gameevent. Like with gameactor, each may be provided in at most one of the two forms.

The callback func is invoked with the same arguments and meaning as for gameactor, but certain events are run in contexts where no meaningful aci and/or pli value can be assigned. In this case, func receives -1 for the respective input arguments.

Like with actors, game events may be chained or replaced by passing an appropriate flags value. However, it is not necessary for an event to be already defined when chaining is requested. In that case, it is simply registered initially. Permissible values for these flags are provided in actor.FLAGS as well (abbreviated AF here):

  • AF.replace: Replace any previously defined event code with the given one.

  • AF.chain_beg: Prepend the provided func to the existing callback function. This is the behavior of CON’s onevent.

  • AF.chain_end: Append the provided func to the existing callback function. This is the default.

5. Extended API (Lunatic modules)

5.1. The xmath module

5.1.1. Mathematical functions

Lunatic, being a Lua-based scripting system, provides the user with a single numeric data type that variables can contain on the Lua side — double-precision floating point.
[In LuaJIT, variables additionally can take on “boxed” 64-bit integer numeric types, but these should not be used for numeric calculations.]
However, since BUILD, and in turn, EDuke32, almost exclusively use integer types to represent quantities such as angles or carry out e.g. trigonometrical calculations, there is a need for convenient interoperability between the two “worlds”.

Two pairs of functions calculate the trigonometric sine and cosine, both accepting a BUILD angle as input argument, but differing in the scaling of their result. Using these functions is recommended over Lua’s math.sin or math.cos in cases where the argument is a BUILD angle in the first place, for example because it is read from an engine or game structure. The computation is both faster (because it is essentially only a bitwise-AND operation followed by a table lookup) and the results have the symmetry expected from the mathematical counterparts.

xmath.sinb(bang),   xmath.cosb(bang)

Returns the sine/cosine of the given BUILD angle bang, which can be any whole number in [–231 .. 231–1].
[Passing fractional values is possible, but discouraged. See the relevant subsection of the BitOp documentation for more details.]
In BUILD, one full cycle is covered by values from 0 to 2047; in other words, an angle of 2048 corresponds to 360 degrees.

The sinb and cosb functions return values in the range [–1 .. 1], just like their mathematical counterparts.

The following guarantees are made for sinb whenever its argument expression is permissible:

  • sinb(-a) == -sinb(a) (point symmetry around the origin)

  • sinb(a + i*2048) == sinb(a), where i is any whole number (periodicity)

  • sinb(1024 - a) == sinb(a) (mirror symmetry around a=512)

  • sinb(a - 1024) == -sinb(a) (point symmetry around a=1024)

  • The value for cosb(a) is derived as sinb(a + 512).

xmath.ksin(bang),   xmath.kcos(bang)

Returns the sine/cosine of the given BUILD angle bang, multiplied with 16384 and rounded towards zero. The same guarantees as for the sinb/cosb pair apply.


Returns a vec3 with the components math.cos(ang), math.sin(ang) and 0 for x, y and z, respectively.


Returns a vec3 with the components xmath.cosb(bang), xmath.sinb(bang) and 0 for x, y and z, respectively.

xmath.kangvec(bang [, z])

Returns an ivec3 with the components xmath.kcos(bang), xmath.ksin(bang) for x and y, respectively. The z component can be passed in the optional argument z, which defaults to 0 if omitted.

xmath.dist(pos1, pos2)

Returns an approximation of the 3D Euclidean distance between points pos1 and pos2, both of which can be any object indexable with x, y and z. BUILD z units are assumed.

xmath.ldist(pos1, pos2)

Returns an approximation of the 2D Euclidean distance between points pos1 and pos2, both of which can be any object indexable with x and y.

xmath.rotate(point, bang [, pivot])

Returns as an xmath.vec3 the position of point rotated around the line parallel to the z axis going through pivot by bang BUILD angle units in the mathematically negative (clockwise) direction. The arguments point and pivot can be anything indexable with x, y and z. The z component of the result is the difference between that of point and pivot. If pivot is omitted, it defaults to the origin vector containing zeros for all components.

5.1.2. The types xmath.vec3 and xmath.ivec3 [serializable]

Another purpose of the xmath module is to provide vector types that allow writing concise and clear code involving geometrical calculations. There are two types, both containing three components (x, y and z), but differing in their numeric type. For the most part, vec3 should be used, whose components are Lua numbers, i.e. floating point. The other type, ivec3, is part of some game structures, and consequently uses 32-bit integers for its components. With minor differences, the vec3 and ivec3 types share the same operations and methods.

The constructors of the vector types can be called in several ways. In the following, they are only described for vec3. The conventions for ivec3 are completely analogous, but since their creation involves a number type conversion, the rules about assignment to integer types apply.

v = xmath.vec3([x [, y [, z]]])

Create a 3-element vector v by passing the x, y and z components separately. Trailing components can be omitted, in which case they are initialized to 0.

v = xmath.vec3(t)

Create a 3-element vector v by passing t, which can be any variable indexable with the strings x, y and z (and yielding numbers for these lookups). For example, t can be another (i)vec3, a sprite or even wall reference, as each of them can be indexed with these three keys.

Since the vector types are compound objects, they are always passed around by reference. For example, consider executing

v = xmath.vec3(0, 10)
w = v
w.y = 20

After this code, the expression v.y yields 20 instead of the initial value 10.

Operations for vec3 and ivec3

In the following, v denotes a vec3 or ivec3 object reference while t denotes any object indexable with x, y and z. Note that for binary operations, Lua looks for overridden operators in the left operand first and the right one next. So, where t appears on the left hand side of an arithmetic expression, it is assumed that t's type does not overload the corresponding operation or provides the same semantics. Arithmetic operations always return a (reference to a) new vec3 object, even if any or both of the operands have ivec3 type.

v + t,   t + v

Returns a new vec3 object whose components are the sum of the respective components of v and t.

v - t,   t - v

Returns a new vec3 object whose components are the difference of the respective components of v and t (in the first case) or t and v (in the second case).


Returns a new vec3 object with the components of v negated.

a*v,   v*a

For a scalar number a, returns a new vec3 object whose components are those of v multiplied with a.


For a scalar number a, returns a new vec3 object whose components are those of v divided by a.


Returns an object of the same type as v and with the same components, except that v.z is diminished by zofs. Also see the power operation for sprite objects.


Returns a string representation of v for display purposes: “vec3” or “ivec3”, followed by the components of v in parentheses.

Methods for vec3 and ivec3

Returns the Euclidean length of v in three dimensions.


Returns the squared Euclidean length of v in three dimensions.


Returns the Euclidean length of v, taking only the x and y components into account.


Returns the squared Euclidean length of v, taking only the x and y components into account.


Returns the length of v calculated using the Manhattan distance
in three dimensions between the origin and the endpoint.


Returns a new ivec3 object with the same components as v, but converted to integers.


Returns a new vector of the same type as v which has the same x and y components as v, but the z element divided by 16 (if v is a vec3) or arithmetically right-shifted by 4 (if v is an ivec3).
[Right-shifting by 4 can be seen as a division by 16 with subsequent rounding to an integer towards negative infinity.]
Also see the description of the ceiling/floor z member.


Returns a new vector of the same type as v which has the same x and y components as v, but the z element multiplied with 16.

v:rotate(bang [, pivot])

Equivalent to xmath.rotate(v, bang [, pivot]).

5.2. The con module — game control

The con module on one hand provides functionality to control certain aspects of the game, such as defining game-side animations for actors. On the other hand, it hosts various functions that are familiar from CON, although sometimes under a different name.

5.2.1. Actions, moves and AIs — customizing actor behavior

In CON, three aspects of an actor can changed using a mechanism that is syntactically similar for each. Actions carry information about a particular animation for an actor, such as how many frames it contains. Moves define the potential motion characteristics of an actor. Finally, AIs are aggregates of an action and a move, plus movement flags that toggle certain movement aspects.

At file scope, one first defines each of these composites using directives such as action, and inside code blocks to be executed at game time, one instructs the game to use a particular composite by passing the chosen label to the same-named command (i.e. action for our example).

Lunatic provides a similar mechanism, but the main difference from CON is that the definition functions merely return the composite structures, without registering them under any particular name. Thus, it is up to the user to organize them, for example by storing those for each class in a separate table indexed by chosen names. Similar to CON, the creation functions may only be called when the game is not yet running.

Actor behavior composites contain an additional “hidden” ID. Each invocation of a creation function such as con.action returns a composite with a unique ID per composite type. This ID is used to compare for equality in actor methods such as actor[]:has_action(). All returned composites are immutable.

The function con.action{ ... }

Returns a new action composite created with the given values. The con.action function expects a single table as input argument, each of whose keys denote an action member. Each member may be provided by an index in the table or its name, but not both. Members may be omitted, in which case they default to either 0 or 1.

Refer to the Table Constructors subsection of the Lua 5.1 Reference Manual for the syntax and semantics of table initialization.

[1] startframe

The offset of the starting frame of the animation, relative to the sprite[].picnum of an actor (i.e. the tile number, selecting which actor code is run).

[2] numframes (default: 1)

The total number of frames in the animation.

[3] viewtype (default: 1)

The number of consecutive tiles used to construct one frame of the animation, as viewed from different angles. Valid values are 1, 3, 5, 7 and 8. In addition, -5 and -7 are allowed, which behave like the corresponding positive viewtypes, but effectively mirror the actor around the vertical axis. This can be useful if tile data is available that has the opposite orientation of what EDuke32 uses. See the Action entry in the EDuke32 wiki for how the views are constructed for different viewtype values.

[4] incval (default: 1)

The value to add the actor’s current frame on each frame advance. May be –1, 0, or 1.

[5] delay

Roughly, the number of gv.totalclock time units (120 per second) after which a frame is advanced, at a granularity of one game tic (30 per second, corresponding to a delay difference of 4).
[The reason for the granularity is due to the implementation: each execution of an actor’s code increments its hidden “action tics” counter by four (= 120/30).]

Example 4. Equivalent action definitions

Each of the following calls return an action with the same public members:

con.action{0, 4, delay=20}
con.action{0, 4, 1, 1, 20}
con.action{startframe=0, numframes=4, viewtype=1, incval=1, delay=20}
The function con.move{ ... }

Returns a new move composite created with the given values, expecting a single table as input argument. The same conventions as with con.action apply for the following members:

[1] hvel

The (most often) doubled horizontal velocity of the actor, in BUILD x/y units per game tic. An actor’s sprite[].xvel will approach this value in an “exponential” fashion, halving the difference between the current velocity and this goal velocity each movement update from the actor execution loop. For most objects, sprite[].xvel is then further halved to yield the ultimate velocity. Only effective when the actor’s movflags has actor.MOVFLAGS.geth is set.

[2] vvel

The (most often) doubled horizontal velocity of the actor, in BUILD x/y (not z) units per game tic. An actor’s sprite[].zvel will approach 16 times this value in an “exponential” fashion. For most objects, sprite[].zvel is then further halved to yield the ultimate velocity. Only effective when the actor’s movflags has actor.MOVFLAGS.getv is set.

The function con.ai([action [, move [, movflags]]])

The con.ai function differs from con.action and con.move in that it is of “second order”, i.e. it accepts two composites as input arguments, returning an AI object containing these. Also, it is called with three positional, optional arguments. Like for gameactor, the numbers 0 and 1 are permissible for action and move. Applicable bits for movflags are available in the actor.MOVFLAGS object.

5.2.2. Non-local control flow

Two functions in the con module make the executed function abort early, jumping directly to the end of the innermost event or actor callback “block”. They are used to implement among others CON’s killit and (confusingly named) return commands. If these functions are used when none of the mentioned callback functions are active, the behavior is undefined.


Silently transfers control to the end of the innermost actor or event callback block, to the same point an error() call would do. Note that since callback chaining is achieved by creating a new function for each function pair, calling both unprotected, issuing a con.longjmp() inside any “part” of a chain aborts the whole block — functions in the chain that are called later will not be reached. In contrast, returning from one function transfers control to the beginning of the next in the chain if it exists.

Example 5. con.longjmp with chained events

The following two chained EVENT_JUMP definitions,

gameevent{"JUMP", function(_, pli)
    if (player[pli].curr_weapon==gv.WEAPON.PISTOL) then

gameevent{"JUMP", function()  -- second function in EVENT_JUMP chain

would print “jump:first” and “jump:second” when holding a pistol. Otherwise, “jump:first” and “aborting” would be printed and the second chained callback function would not be reached.


Silently transfers control to the end of the active actor callback block, notifying the game to delete the executing actor’s sprite. If con.killit is called while no execution of actor code is active, the behavior is undefined.

5.2.3. Per-actor variables

Since in EDuke32, sprites in general exist in the world only for a limited duration, it is wasteful to allocate an array of fixed size gv.MAXSPRITES for the purpose of a variable that holds a value for each actor sprite. On the Lua side, one could use a plain table, but for actors that are created and destroyed during the course of a game, this would accumulate “garbage” — values for sprites that have been deleted. Moreover, per-actor variables tend to be “topical”, one such variable being potentially only used for a very specific actor tile. For this reason, per-actor variables are implemented in a “sparse” fashion in Lunatic, but provide to the user the illusion of having a value for every sprite index. They are also “cleaned” at unspecified intervals.

The type con.actorvar(defaultval) [serializable]

Creates and returns a new per-actor variable with default value defaultval which can be indexed for reading or assignment in the range [0 .. gv.MAXSPRITES-1], but access to it is subject to the same restrictions as to sprite and other per-sprite structures.

When a sprite is created using con.insertsprite or con.spawn, its value at the index of this new sprite is cleared (reset to defaultval). After a sprite has been deleted, the value of a per-actor variable is indeterminate — it may be cleared by Lunatic at any point.

Per-actor variables may contain values of any permitted type, which currently are boolean and number. Mixing values of different types is allowed: per-actor variables are heterogenous containers.

5.2.4. Sprite insertion

In Lunatic, there are two functions that insert a sprite into the game world. They mainly differ in how they are used, and to which extent they imply “hard-coded” behavior.

The function con.insertsprite{tilenum, pos, sectnum [, statnum [, owner]] [key=val...]}

Inserts a new sprite into the game with the properties given as input arguments. If the world already contains the maximum number of sprites (gv.MAXSPRITES), currently the game is aborted and EDuke32 exits.
[This is subject to change and must not be relied on.]
Otherwise, relevant per-sprite data for the newly inserted sprite is cleared. No additional “hard-wired” C code is run.

Returns the index of the inserted sprite.

The function con.insertsprite can be used in one of two forms:

  • In the table-call form specified above, a single table is passed whose values are taken as the actual arguments. The first three, tilenum, pos and sectnum, are passed positionally. All other input arguments are passed as key-value pairs, but owner and statnum may be provided either positionally or as key/value pair.

  • Passing only the three to five positional arguments is also directly possible. For example, all of the following calls are equivalent:

    local i = con.insertsprite(tile, pos, sect, stat, ow)  -- direct-call
    local i = con.insertsprite{tile, pos, sect, stat, ow}  -- table-call
    local i = con.insertsprite{tile, pos, sect, statnum=stat, owner=ow}  -- table-call with 2 k/v args
    local i = con.insertsprite{tile, pos, sect, stat, owner=ow}  -- table-call with one k/v arg

The five main arguments are as follows:

[1] (tilenum)

The tile number of the sprite to insert.

[2] (pos)

The position at which to insert the sprite (anything indexable with x, y and z).

[3] (sectnum)

The index of the sector in which to insert the sprite.

[4] statnum (default: actor.STAT.DEFAULT)

The initial status number of the inserted sprite.

[5] owner (default: see below)

The index of the sprite that is in some sense the “parent” of the newly created one. If omitted, it is set to the index of the newly spawned sprite.

These keys are permissible as optional input arguments in the table-call form, corresponding to the same-named sprite members:

  • shade, ang, xvel, zvel (default: 0)

  • xrepeat and yrepeat (default: 48)

The function con.spawn(tilenum, parentspritenum)

Spawns a new sprite with tile number tilenum from a given “parent” sprite with index parentspritenum, which must be valid. The sprite is spawned at the same position as its parent and its owner member is set to parentspritenum. Additional “hard-wired” code dependent on tilenum may be run afterwards, possibly modifying the sprite.

Returns the index of the spawned sprite on success.

The function con.shoot(tilenum, parentspritenum [, zvel])

Attempts to shoot a projectile with tile number tilenum from the sprite with index parentspritenum. The z velocity can be overridden by passing zvel.

Returns the index of the spawned sprite on success, or –1 otherwise.

5.3. The engine module

5.3.1. Base palette interfaces

The 8-bit (“classic”) Build renderer operates in indexed-color mode: the pixels of the frame buffer do not contain red/green/blue values themselves, but only indexes into a color table with 256 entries, the base palette.

$\mathrm{ColorComponent} := \mathbb{N}_{64}$   // Build’s base palettes have 6 bits of precision per color component
$\mathrm{ColorIndex} := \mathbb{N}_{256}$
$\mathrm{basepal}: \: \mathrm{ColorIndex} \rightarrow \mathrm{ColorComponent}^3$

The following functions provide a bridge between color indices and their corresponding color components.

r, g, b = engine.getrgb(i)

Returns the red, green and blue color components of the default base palette for color index i. The color components are in the range [0 .. 63].

If i is 255, r, g and b are all returned as 0, even if the actual base palette may contain different values for that index.

i = engine.nearcolor(r, g, b [, lastokcol])

Given the red, green and blue components r, g and b of a query color, returns the color index i whose color in the default base palette is closest to the query color. “Closeness” is established using an Euclidean distance with a weighting on the color components.

The optional argument lastokcol can be used to restrict the range of color indices to search: the returned color index is guaranteed to be in the range [0 .. lastokcol]. It defaults 255, so that all colors are searched.
[For example, it may be desirable to omit “fullbright” colors from being returned. The shade table loaded from Duke3D’s PALETTE.DAT makes color indices 240—254 fullbright, so passing 239 to lastokcol achieves the mentioned filtering.]

5.3.2. Shade table interfaces

To implement shading and visibility attenuation, Build maintains tables mapping pairs of a color index and a shade level (Duke3D’s table uses 32 such gradients) to a color index representing the darkness-faded color. Each such table is called shade or palookup table.

$\mathrm{ShadeLevel} := \mathbb{N}_{\mathrm{Numshades}}$
$\mathrm{palookup}: \: \mathrm{ShadeLevel} \times \mathrm{ColorIndex} \rightarrow \mathrm{ColorIndex}$

When a pixel is about to be drawn, a palookup table chosen depending on the object’s pal is consulted to determine its ultimate color index (in the absence of blending with the translucency table or see-through texels). Given a texel’s color index as $i_{\mathrm{in}}$, the resulting pixel’s one $i_{\mathrm{out}}$ is computed as

$s_1 = \mathrm{shade} + C \cdot \mathrm{visdist}$
$s_2 = \mathrm{clamp}(s_1, \: 0, \: \mathrm{Numshades}-1)$
$\mathrm{shade_index} = \mathrm{round}(s_2)$
$i_{\mathrm{out}} = \mathrm{palookup}(\mathrm{shade_index}, i_{\mathrm{in}})$   // This is only a table lookup, palookup[shade_index][$i_{\mathrm{in}}$]

Here, $C$ is a positive constant and $\mathrm{visdist}$ is the product of a. the distance of an object’s sampled pixel to the view plane with b. the object’s “visibility”.
[Visibility would be more appropriately called “anti-visibility” or “co-visibility”: greater values make objects appear more faded for the same distance. Also, the visibility that is meant here has the origin at 0, unlike sector[].visibility.]
Thus, shade and visibility are inherently confounded in the 8-bit mode and the ultimate shade_index is bounded below by (the minimum of $\mathrm{Numshades}-1$ and) the shade of the object.

Examples of effects using shade tables

While palookup tables are primarily used for shading and visibility attenuation, they can be set up in other ways to yield different effects with respect to how pixels of objects farther away are drawn. For example:

  • Distance fading with fog. For a fog color $\mathbf{c} = (c_r, c_g, c_b)$, the table is set up so that for a source color index $i$ and a shade level sh, palookup[sh][$i$] contains a color index whose color is close to that of $i$ blended with $\mathbf{c}$,

      $\frac{\mathrm{sh} + 0.5}{\mathrm{Numshades}} \cdot \mathrm{basepal}(i) + \frac{\mathrm{Numshades}-\mathrm{sh}+0.5}{\mathrm{Numshades}} \cdot \mathbf{c}$.

    Note that distance fading to black can be seen as a special case of this fogging effect. However, Duke3D’s base shade table (i.e. the table for pal 0) is not constructed in this way.

  • Color index remapping. Given a mapping $m: \: \mathrm{ColorIndex} \rightarrow \mathrm{ColorIndex}$, the table is set up so that for each shade level sh, the 256 color indices are selected or reordered in the same way: for all color indices $i$, palookup[sh][$i$] = original_palookup[sh][$m(i)$].
    For example, pal 21 remaps the fifth and sixth blocks of consecutive 16-tuples of color indices (a ramp of blue colors) to the fourth and 14th such blocks (red colors, the first one part of a 32-color ramp).

  • “Fullbright” colors — those that aren’t affected by distance — with index $i$ are achieved by setting palookup[sh][$i$] to palookup[0][$i$] for each shade sh.

Shade table functions
sht = engine.shadetab()

Creates and returns a new shade table object sht with all entries initialized to zero. This object can be indexed once with a shade index from 0 to 31, yielding a reference to an array of 256 8-bit unsigned integers. Thus, shade table objects can be used just as indicated in the notation above: sht[sh][i] is the resulting color index for shade sh and input color index i.

sht = engine.getshadetab(palnum)

Returns a new shade table object sht containing the values for the palookup table of pal palnum, provided it is not an alias for the default one (see below). Modifying the returned sht does not alter the actual engine-side shade table. An unspecified number of pal indices are reserved from the end of the hard limit 255; attempting to retrieve a shade table for a reserved pal raises as error.

At engine initialization, the shade tables for all non-zero pal numbers are aliased to the default pal 0 one. Subsequently, custom palookup tables are either loaded from LOOKUP.DAT, created by appropriate DEF directives (fogpal or makepalookup), or assigned using engine.setshadetab. If the table for pal palnum is aliased to the default one when getshadetab is called, nil is returned.

engine.setshadetab(palnum, sht)

Copies the shade table sht to the engine-side palookup table for pal palnum. An error is issued if an attempt is made to copy to a reserved palnum.

When running in EDuke32, there are additional restrictions:

  • A palnum for which a shade table has already been registered (that is, one which is not aliased to the default one) cannot be re-assigned to.

  • setshadetab may only be called at first initialization time, that is, when the value of LUNATIC_FIRST_TIME is true. (LUNATIC_FIRST_TIME is a variable in the global environment that indicates whether the Lua game state is created for the very first time and the game is not yet running.)

Shade table methods
newsht = sht:remap16(tab)

Returns a new shade table with consecutive runs of 16 values of every 256-tuple of sht remapped as specified by tab. Specifically, tab must be a table whose keys in [0 .. 15] may be set to values in [0 .. 15]. For a shade index sh and a color index i, the returned shade table newsht then has

newsht[sh][i] = sht[sh][newi],


newi = 16*tab[math.floor(i/16)] + i%16   // if tab[math.floor(i/16)] is non-nil
newi = i   // otherwise

-- Creates a shade table with the same remapping as pal 21 (blue -> red) and
-- registers it under pal 22, overwriting its previous contents.
local newsht = engine.getshadetab(0):remap16({[4]=13, [5]=8})
engine.setshadetab(22, newsht)

5.3.3. Blending table interfaces

EDuke32 supports installing multiple blending tables used by the 8-bit renderer. A blending (or translucency) table is used whenever an object with the “translucent” bit set is drawn, and maps pairs of color indices to a blended color index.

$\mathrm{transluc}: \: \mathrm{ColorIndex} \times \mathrm{ColorIndex} \rightarrow \mathrm{ColorIndex}$

Given a background color index bi and a color index of an incoming foreground fragment fi (usually obtained by looking up a shade table), when fi is different from 255, the resulting color index is


if the “reverse translucency” bit is clear, and


if reverse tranlucency is set. If the fragment sampled from the foreground object has color index 255, it is discarded: translucent rendering is always “masking”.

Currently, only sprites support rendering with custom blending tables, by setting their .blend member to the number of a blending table.

Blending table functions
tab = engine.blendtab()

Creates and returns a new blending table object tab with all entries initialized to zero. This object can be indexed once with a color index from 0 to 255, yielding a reference to an array of 256 8-bit unsigned integers. Thus, blending table objects can be used just as indicated in the notation above: tab[i][j] is the blended color index for input color indices i and j.

tab = engine.getblendtab(blendnum)

Returns a new blending table object tab containing the values for the engine-side translucency table numbered blendnum, or nil if no blending table with that number is registered. Modifying the returned tab does not alter the actual engine-side blending table.

engine.setblendtab(blendnum, tab)

Copies the blending table tab to the engine-side translucency table with number blendnum.

Similar to engine.setshadetab, there are certain restrictions when running in EDuke32:

  • A blendnum for which a table has already been registered cannot be re-assigned to.

  • setblendtab may only be called at first initialization time, that is, when the value of LUNATIC_FIRST_TIME is true.

5.3.4. Mapster32-only functions

engine.savePaletteDat(filename [, palnum [, blendnum [, moreblends]]])ok, errmsg

Writes out a full PALETTE.DAT-formatted file named filename with the base shade table numbered palnum and the base translucency table numbered blendnum, both defaulting to 0.

Passing moreblends allows to specify additional blending tables to store in EDuke32’s extended PALETTE.DAT format. These must have previously been registered with engine.setblendtab. The moreblends argument must be a sequence table with each element being either

  • a blending table number in the range [1 .. 255]

  • a table t containing a pair of such numbers, in which case it is taken to mean the inclusive range [t[1] .. t[2]]

There must be no duplicate blending table numbers.

The function returns a status ok which is true on success and nil on failure. In the latter case, errmsg is a diagnostic error message.

engine.saveLookupDat(filename, lookups)ok, errmsg

Writes out a LOOKUP.DAT-formatted file named filename with the lookup tables specified by lookups at the beginning and the five additional base palettes at the end.

The lookups argument is interpreted analogously to the moreblends argument of engine.savePaletteDat (with the numbers being palookup numbers instead of blending table numbers) and the return values ok and errmsg have the same meaning as well.

5.4. The fs module — virtual file system facilities

files = fs.listpath(path, mask)

Returns a sequence table files of file names that can be found in a directory constructed as concatenation of any directory in the search path with path, and matching the wildcard mask. Currently, neither ZIP nor GRP files registered as file containers with EDuke32 are searched.

The path argument must separate directories by forward slashes (/). If no suffix is desired (i.e. the directories in the search path themselves are to be searched), '/' should be passed.

The file name mask is applied case-insensitively for the 26 characters of the basic Latin alphabet. It may contain the following meta-characters which are interpreted in a special way:

  • a * matches any (potentially empty) sequence of characters

  • a ? matches any single character

While the match with mask proceeds case-insensitively, file names are returned named exactly like on the file system. However, names differing only in case appear exactly once in the output list.

Note For portability, it is crucial that path is specified with the same case as the actual directory on the file system.

Suppose the search path contains two directories foo and bar with the following file listing:



Then a query with

fs.listpath("/", "myhouse?.map")

will return a table with these strings:

  • myhouse1.MAP or MYHOUSE1.map, but not both

  • myhouse2.map

Appendix A: How to read Programming in Lua

On the Lua homepage, the first edition of Programming in Lua is available online. While targeting Lua 5.0, it still remains applicable to Lua 5.1 to a large extent. This section gives hints for reading it when learning to code for Lunatic.


May be interesting to get an idea of the philosophy behind Lua. A Few Typographical Conventions should be read to be familiar with them.

1 — Getting Started

Mentions the stand-alone Lua interpreter. When none is available, a LuaJIT stand-alone binary can be used for experimentation as well.

1.1 — Chunks

Introduces chunks as the basic “blocks” of Lua code and notes how whitespace is treated. Mentions dofile, which is not available in Lunatic (require is preferred).

1.2 — Global Variables

Section may be read, but usage of local variables is strongly recommended whenever possible. Also, trying to read a non-existent global or to write any value to the global environment gives an error in Lunatic (except that global writes are allowed in in module context, i.e. after module(...)).

1.3 — Some Lexical Conventions

Must read.

2 — Types and Values

Must read, also subsections. However, “2.7 — Userdata and Threads” is not relevant.

3 — Expressions

Must read, also all subsections.

4 — Statements

Must read, also all subsections.

5 — Functions

Must read, also all subsections. Subsection 5.2 mentions io.write, which is not available in Lunatic.

6 — More about Functions

May be read (subsections as well) for a more complete understanding of functions in Lua, as well as the utility of lexical scoping.

7 — Iterators and the Generic for

May be read (subsections as well), but for basic programming, the knowledge of how to merely use (as opposed to write) iterators such as spritesofsect suffices.

8.3 — Errors

May be read. Provides guidelines on how to write error handling (status code vs. error).

8.4 — Error Handling and Exceptions

May be read. Discusses protected calls, which should be used sparingly.

11 — Data Structures

May be read, also subsections. The most relevant subsections are 11.1 — Arrays, 11.5 — Sets and Bags and to some extent, 11.6 — String Buffers. The way “11.2 — Matrices and Multi-Dimensional Arrays” suggests to construct matrices is rather memory-intensive; also it and “11.3 — Linked Lists” and “11.4 — Queues and Double Queues” are not relevant to simple Lunatic coding.

15 — Packages

Lua package system received various additions in 5.1, so the PiL first edition’s section is out-of-sync. For Lunatic, the modules section of this manual should be consulted.

(The rest of Part II deals with advanced concepts not needed for simple Lunatic coding.)

19.3 — Sort

May be read if needed.

20 — The String Library

May be skimmed (also subsections), though ultimately the Lua Reference should be consulted for the exact semantics.

Appendix B: Game event RETURN usage

The following list notes how the special gv.RETURN variable (known simply as RETURN from CON) is treated for the various events. The game may either pass some value to a particular event as additional information, or it can examine RETURN after the event has finished, invoking some behavior conditional on its value. If an event is not listed here, its usage of RETURN is unspecified.

Receives zero, checks for non-zero

These events get passed a value of 0 for RETURN, and after finishing, check it for being non-zero, in which case some hard-coded behavior is suppressed:


KILLIT (deprecated from Lua)

If non-zero, the pending sprite deletion is aborted.


Don’t draw scene if RETURN is 1. Values other than 0 and 1 are reserved.


If non-zero, the “fire” or “use” shared key bits are cleared (respectively).

Game considers post-event RETURN an index

For some events, the game examines RETURN after they have finished, and potentially uses its value as some kind of index.

CHANGEMENU (deprecated from Lua)

Receives and examines RETURN as a menu index to change to.


Receives 0. If the post-event RETURN equals 1, no crosshair is drawn. If it is greater than one, RETURN is the tile index of the crosshair to be displayed. The value 0 makes it draw using the CROSSHAIR tile.


Receives and examines RETURN as a background tile for the loading screen. A negative value suppresses drawing it and running the subsequent EVENT_DISPLAYLOADINGSCREEN entirely.


Receives and examines RETURN as a background tile for the menu.


Receives and examines RETURN as an index of a sound to start playing.

Cheat events

The cheat events receive a hard-coded, item-dependent target amount in RETURN. It can be overridden to a different value to make the respective cheat give a different “full” amount. A negative value is ignored.



Triggered when a ceiling or a floor (collectively called “hplane”) is determined as being damaged. The event receives RETURN in the variable gv.RETURN as well as the third, dist argument to the event callback function.
[Passing RETURN in the dist argument serves the possibility to create chained callbacks for EVENT_DAMAGEHPLANE. Otherwise, once gv.RETURN were assigned to, there would be no way to obtain its original value in subsequent chained callbacks.]
This value can be decoded into two parts by passing it to sector.damagehplane_whatsect:

function(aci, pli, RETURN)
   local what, sectnum = sector.damagehplane_whatsect(RETURN)
   -- (...)


  • what is one of the strings 'ceiling' or 'floor' and

  • sectnum is the sector whose hplane is considered to be damaged.

When EVENT_DAMAGEHPLANE is left, gv.RETURN is examined to determine the further action. It may be one of three values given by sector.DAMAGEHPLANE (abbreviated DHP in the following):

  • DHP.SUPPRESS: the hard-wired code that would subsequently be run is suppressed entirely

  • DHP.DEFAULT: The default code for hplane damaging is run. For floors, it does nothing. For ceilings, it checks whether it has a tile number in a hard-coded set of values depicting a breakable light. In that case, the tile number is changed to the “broken” version and a “glass breaking” effect consisting of playing a sound and spawning glass sprites is started. Also, certain code related to SE3 and SE12 effects is run.

  • DHP.GLASSBREAK: The light-breaking effect described above is run unconditionally, but without changing the hplane’s tile number, which is assumed to have been done by the event.

If value last assigned to RETURN is not one in the above-mentioned set when EVENT_DAMAGEHPLANE is left, the behavior is undefined.