GitLab

Previous to this commit, the game would hijack the sprite, wall and sector
arrays to store the label data temporarily, which would get copied into
freshly allocated arrays once the scripts were fully parsed.

This lead to the maximum number of labels depending on MAXSPRITES.
This coupling is now removed, and an independent limit MAXLABELS
has been introduced, which is currently set to 16384.

Additionally, the limit is now checked before character data is loaded
into memory, rather than afterwards, preventing out of bounds accesses.

No longer backup the walock, waloff and picsiz arrays as this caused
problems elsewhere.

Fixes #66

With 'RELEASE := 0', gcc (Raspbian 8.3.0-6+rpi1) 8.3.0 said:

 savegame.cpp: In function 'int32_t G_LoadPlayer(savebrief_t&)':
 savegame.cpp:655:12: warning: 'void* memcpy(void*, const void*, size_t)' offset [297, 328]
   from the object at 'h' is out of the bounds of referenced subobject '_savehead::boardfn'
   with type 'char [224]' at offset 72 [-Warray-bounds]
      Bmemcpy(boardfilename, h.boardfn, sizeof(boardfilename));

It also emits many warnings about possible truncation with (B)snprintf(),
presumably based on computed upper bounds on the length of the input varargs.

Also, resurrect static assert for post-commit-687000e2 situation and
remove comment referencing a tag which was removed in commit bd3a933a.

source/audiolib/src/driver_adlib.cpp:499:13: runtime error: member access within null pointer of type 'AdLibVoice'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior source/audiolib/src/driver_adlib.cpp:499:13 in

Background: gcc (Raspbian 8.3.0-6+rpi1) 8.3.0
I'm getting a missing compiler-provided symbol:

 /usr/bin/ld: obj/build/enet.o: in function `enet_time_get':
 source/build/include/enet.h:4948: undefined reference to `__atomic_load_8'
 /usr/bin/ld: source/build/include/enet.h:4958: undefined reference to `__atomic_compare_exchange_8'
 collect2: error: ld returned 1 exit status

so have a means of omitting the object file when it is not actually needed.

The condition inside VM_Assert always returned false because
`ResizeArray()` ensures that `sequenceSize == aGameArrays[arrayNum].size`.

We may have gotten an error message like:

  videoSetGamma(): Failed loading udev_device_get_action:
      (...)/libSDL2-2.0.so.0: undefined symbol: _udev_device_get_action

But:
 1. The "error" in question did actually happen earlier in SDL, but is harmless
    because SDL first tries to load a symbol using 'dlsym(RTLD_DEFAULT, ...)' and
    failing that (as it happens in my case), from an explicitly loaded libudev
    (which succeeds).
 2. The message misleads a user into thinking that it may have something to do
    with gamma adjustment having no effect. (As is the case for me.)

See:
 https://github.com/zpl-c/zpl/pull/44
 https://github.com/zpl-c/zpl/issues/45

The first is something that was missed when the hardcoded World Tour stuff was added, and the second looks like it was just a typo.

Fixes !40

mapster32/build: Make AlignWalls_ not a partially broken, cryptic mess. Correctly aligns textures of any size now.

engine: Implement tileflags member of picanm, and added support for new flag: TILEFLAGS_TRUENPOT, that allows you to enable true non-power-of-two texture support per-tile.

(Fixes classic's buggy tiling and panning behaviour and bypasses the emulation of it in Polymost)

engine: Enable non-power-of-two wall support in classic renderer. (Doesn't do anything yet without either a new map format, or upcoming TILEFLAGS_TRUENPOT).

I needed to change the weird VS workaround in order for the EDUKE32_STATIC_ASSERT on line 850 to succeed.

Things seemed OK when I tested them, but I'd like more information on the suspected compiler bug in question and why this was necessary in the first place.