What changed in EDuke32, build by build

The only operation for which proper code is generated is reading a single value from a system gamearray.

Fix 'setsprite' and 'music'. A good timing test for actors performance is the starfield of WGR2's E1L1 ("Nexus").

This is the only gamevar of its kind that is handled for Lunatic, neither PLR_MORALE nor LOGO_FLAGS are.

Introduced with the refactoring in r3490.

Also, fix con._setgamepalette.

The option is -fgendir=<directory>.

All non-option positional arguments after an @ sign are taken as list files: hashes at the beginning of a line denote comments, truly empty lines are ignored, all other lines should contain one root CON file name.

Fixing movement when not loading a.m32.

- add LuaJIT's 'v' module printing trace info - translator: fix game function name definitions - revert math.fmod -> math.modf, they are different! - disable JIT compilation for a function we're getting strange crashes with - Make some of DukePlayer_t's members 'bool' on the Lua side. It's way too easy to write something like "ps.jetpack_on" where "ps.jetpack_on~=0" was meant. [Background: Kyle873 observed that Duke was always floating.] - Error out if looping in our_module(). I find this behavior more logical than returning true. - fix a couple of missed FORBID variables

Note: usually, languages call this "continue". Jeez, CON...

Notes: - MAXJOYBUTTONS is out of sync between the two definitions - In menus.c, NUMMOUSEFUNCTIONS must equal MAXMOUSEBUTTONS, fail to compile if this becomes false.

This fixes an update issue: start EDuke32 without warping to a map in Polymost, change to classic, and change to Polymost again. The OSD background will be displaced.

Also, fix quote commands there.

Since the size of defs.ilua's bytecode is used in game.c.

We must not call these functions using the FFI, since the Lua state is considered locked across such calls.

Since it's too easy to hit the 200 locals limit.

Also, for Lunatic, handle killit from events.

If m32script gamevar "move_by_one" is nonzero, the some keys move the "player arrow" by increments of 1: - Without SHIFT: LEFT/RIGHT absolute x, DOWN/UP absolute y, A/Z absolute z. - With SHIFT: DOWN/UP (unbounded!) horiz, LEFT/RIGHT angle. This can be useful to debug renderer bugs that show a high sensitivity to the exact location ("are transient in space").

See comments in the source for details.

Also, slightly better, though still inadequate checking.

Face- and wall-aligned sprites are drawn using the wall routines in BUILD. However, the per-x-screen-coordinate distance (swall[]) is calculated in a way that potentially incurs great precision loss (for example 5 bits for xdimen=1280, yxaspect=65536). This leads to the starting (top) vertical texture coordinate possibly wrapping to large values, leaving an unsightly "stray line" on top of the sprite from certain viewing angles/horiz values. The approach to fix it has two parts: first, the distance is calculated using float values, preventing the precision loss. Because this doesn't fully prevent the unwanted lines, the texture coords are clamped to the mininum and maximum (0 and UINT32_MAX respectively) when calculating them for sprites. Note that stray lines may still appear at the *bottom* of sprites under certain circumstances, for example when viewing at a y-flipped sprite from above. These should be less noticable in real-world usage though. The feature is guarded by a macro HIGH_PRECISION_SPRITE in case using floating point or 64-bit integers is undesirable/impossible on some platforms.

Bump BYTEVERSION.

That is, first check if the sprite is in the game world and then if it already has the desired sector or status number. This doesn't change anything in our codebase, since the return values of these functions are never examined.

Also taking care that the USE_OPENGL=0 build builds.

Previously, actor[].shootzvel (implementation detail, not available to CON) was checked, and if it was !=0, that was the overridden velocity. The value 0 meant "hardcoded, projectile-dependent velocity". But that neccesiated a hack where if zvel 0 was passed and really meant, it needed to be set to 1 instead. Now we have A_ShootWithZvel() taking an additional last argument plus a macro SHOOT_HARDCODED_ZVEL permissible for that argument.

Also, simplify a couple of other lines.

actors.h: remove 'packed' from projectile_t, tiledata_t. In tiledata_t, make .cacherange member an int32_t so that the following member "projectile_t defproj" is aligned on a 4-byte boundary. player.h: remove 'packed' from playerspawn_t, DukeStatus_t, input_t. In Lunatic, correct packing attribute of the base type of the unrestricted actor_t and DukePlayer_t pointer types (the declaration was used without Bump BYTEVERSION.

Most of them are already aligned to their natural boundaries, so lowering the alignment to 1 byte can only worsen things by making the C compiler generate poorer (unaligned access) code for some platforms. The layout of structures is not specified by the C Standard, but is rather given by a particular platform + toolchain's ABI (application binary interface). Most ABIs follow the expected pattern "alignment of scalars is their size, alignment of arrays is that of its element type, alignment of structs is the maximum alignment of its members". A couple of links to particular ABIs are given in build.h. Problems are expected with archs that care about unaligned access when a pointer to a non-packed struct is taken that resides in a packed aggregate, but these uses should be weeded out (I'm not sure if there are any in our codebase). The following types are affected, only hitdata_t changes its size: sectortype, walltype, spritetype, spriteext_t, spritesmooth_t, struct validmode_t, picanm_t, palette_t, vec2_t, vec3_t, hitdata_t.

Introduced in r3358, where I missed a condition inside an 'if'.

And const them appropriately. Also remove #if 0'ed code related to an MSVC pragma.

Meaning that only *ettsprite[THISACTOR] makes sense from CON. It did before too, because .tspr was set before each event run (and not before all runs), only that it was never nulled, which was kind of untidy.

Its only use is to have a actor -> tsprite mapping for the EVENT_ANIMATESPRITE event and .tspr will be set before it is run.

Pass types via ffi.typeof() instead of declaring them in the global namespace when possible.