What changed in EDuke32, build by build

Conceptually, this amends r4428.

Background: AMC TC initializes per-player variables from this event.

It does not depend on the "swap bottom wall parts" bit being set. The fix is by matching up the related code with Polymost.

For glocal gamevars, the LunaCON semantics (session gamevar) are kept. For per-actor ones, this now prevents resetting a gamevar's value to default on spawn or loadactor. BUILD_LUNATIC.

For the other player members, where formerly negative values were allowed to be written, only -1 is now.

This fixes RPGs fired upwards at a steep angle teleporting immediately to underwater when wading in water (reported via PM by Turok 64 Jr).

The (undocumented) Lua-side API stays almost the same.

Also, on one "gamevar `xxx' is not per-*" diagnostic, print the location of the gamevar definition.

Left: projectile changes.

For this, add a setaspect_new() setup/restore pair in M32_DrawRoomsAndMasks() like for G_DrawRooms(). With this, changing viewingrange/aspect via m32script (in a.m32: [7] -- [9] on the upper row) can only be done in r_usenewaspect 0, though.

Move projectile_t to player.h.

This was formerly "fix multi-tile ceiling skies.", but TX beat me to it and Git automatically figured out how to merge while keeping a linear history. (Our fixes were identical.) Nice!

Also, make cloudx[] and cloudy[] scalars, as the per-sector values were always identical. In prelevel(), warn if some CLOUDYSKIES-ceiling sectors could not be set up due to reaching the new 256 sector limit.

Division has higher precedence than bitwise-OR. The cases concerned SE3, SE4 and NEON* sprites. Incidental info: the blinking of NEON* sprites can be controlled via their lotag, similarly to how you how you give a "random flash number" as SE4 hitag.

NONUSBACKSLASH is what is generated for the key between the LShift and 'Z' here. The aliasing with the actual ',' key is as with the SDL1.2 build. Fixes the issue reported in http://forums.duke4.net/topic/8029-cant-bind-key-anymore/

Because we get spammed otherwise, regarding initializations of 'vec3f_t' and 'vec3d_t' variables. Update: actually, it also happens for GCC builds for me.

Reported in http://forums.duke4.net/topic/1348-mapster32-problems-and-bugs/page__view__findpost__p__220722 I could not reproduce it, but it is thinkable that 'searchsector' could have been out-of-sync with the 'sectnum' local. (For example, the former one could have been -1, triggering the crash.)

Reported in http://forums.duke4.net/topic/3857-the-crash-thread/page__view__findpost__p__220954 Now, PitchTable[] in initialized with the same values as pre-r4608.

DONT_BUILD.

BUILD_LUNATIC.

The latter means that running perpendicular thin sector is handled better if there are e.g. TROR sectors in the same x/y space.

That is, resurrect the stayput condition for LIZMAN and non-flying LIZTROOP, but only if changing from a non-ST1 sector. DONT_BUILD.

This fixes engine-side sprite animation in the following scenario: CON code wants to draw a scene from EVENT_DISPLAYREST, but since it covers the whole screen, disables the drawing of the 3D scene beforehand (RETURN set to 1 from EVENT_DISPLAYROOMS). DONT_BUILD.

This fixes the level 5 corruption encountered when issuing that iteration. (Because the last sprite would be changed from e.g. statnum MAXSTATUS to 0.) It is also mandatory anyhow since the last sprite may be contained in the map.

DONT_BUILD.

- In drawvox(), make view-relative x and y high-precision on desktops. To a large extent (but not completely), this fixes "stray" fake floor shadow sprites for voxels. - In the mouse picking code for voxels, fix a possible integer overflow. A new engine.c-private function mulscale_triple30() is added. DONT_BUILD.

Also, note a place that is the real cause for "what makes the player slide off enemies", as far as I can see.

- VM_GetZRange(), which updates actor[].{floor,ceiling}z, is called from A_GetZLimits() and A_Fall(). - Style-cleanup the following: A_Dodge(), A_GetFurthestAngle(), A_FurthestVisiblePoint(). DONT_BUILD.

This reverts r4975. DONT_BUILD.

Also, in premap.c:G_CacheMapData(), comment out what was supposedly intended as a debugging message ("percentage ...").

And preventing enormous corruption due to a free() called on a garbage (malloc'd) pointer values this time. DO_BUILD_VERY_FAST_PLEASE!

Previously, that was the case only with DRONE and COMMANDER. This makes non-falling enemies pass properly through TROR bunches.

Also, some minor stylistic cleanup.

- Add declarations of C functions to defs.ilua, names to dynsymlist, fix typo in con_lang.lua, ffi.string() calls in control.lua - Assert that G_LastMapInfoIndex() is always called with ud.last_level >= 1. (A stricter requirement than necessary to prevent follow-up oob accesses, but logically the most meaningful.) - In G_PrintParTime() and G_PrintDesignerTime(), return "<invalid>" if the above does not hold. This can happen from EVENT_NEWGAME, for example. Add a test to lunatic/test/qgetsysstr.con. DONT_BUILD.

Since TILE_TILT is only allocated once, it must be done with the maximum possible size.

Exposed by Valgrind. DONT_BUILD.

This fixes the out-of-bounds read of former g_player[] in VM_EventCommon_().

By default, the following enemies have this flag set: OCTABRAIN, COMMANDER, DRONE. DONT_BUILD.

This is where things are changed at last. This fixes lizmen walking on water.

Factor out condition for whether a z offset is added or not ("fix for flying/jumping monsters getting stuck in water", etc.) into a function A_GetWaterZOffset() (and a helper A_GetVerticalVel()).

In VM_Fall(), keep a temp preliminary new z position of the actor and assign only when we return. This is to each ease watching its value in GDB. DONT_BUILD.