turns out that __builtin_alloca_with_align() might releases the
allocated memory at the end of the block it was allocated in, instead
of the end of the function (which is the behavior of regular alloca()
and __builtin_alloca()): "The lifetime of the allocated object ends at
the end of the block in which the function was called. The allocated
storage is released no later than just before the calling function
returns to its caller, but may be released at the end of the block in
which the function was called."
https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html#index-_005f_005fbuiltin_005falloca_005fwith_005falign
Clang also supports __builtin_alloca_with_align(), but always releases
the memory at the end of the function.
And it seems that newer GCC versions also tend to release it at the
end of the function, but GCC 4.7.2 (that I use for the official Linux
release binaries) didn't, and that caused weird graphical glitches.
But as I don't want to rely on newer GCC versions behaving like this
(and the documentation explicitly says that it *may* be released at
the end of the block, but will definitely be released at the end of
the function), I removed all usage of __builtin_alloca_with_align().
(Side-Note: GCC only started documenting the behavior of
__builtin_alloca and __builtin_alloca_with_align at all with GCC 6.1.0)
According to
https://cmake.org/cmake/help/v3.26/variable/CMAKE_LANG_COMPILER_ID.html
there's at least "Clang" (plain LLVM clang), "AppleClang", "ARMClang",
"FujitsuClang", "XLClang" and "IBMClang" (both of which are variations
of the clang/llvm-based IBM XL compiler).
This should detect all of them (I hope they all behave close enough
to normal clang to work..)
There's also "IntelLLVM", but I have no idea if that's based on Clang
or does its own thing based on LLVM.. I also doubt dhewm3 will ever
be built with any other clang-derivate than "Clang" and "AppleClang" :-p
fixes#510
For some reason MSVCs integrated CMake doesn't set CMAKE_GENERATOR_PLATFORM
so my CPU arch detection magic in CMakeLists.txt didn't work in that case..
Now (when using MSVC) the CPU architecture (D3_ARCH) is set in
neo/sys/platform.h instead (NOTE: the sys/platform.h part was merged
in "Sync sys/platform.h, incl. Workaround for MCST-LCC compiler").
-DASAN=ON enables the Address Sanitizer (ASan), if supported
-DFORCE_COLORED_OUTPUT=ON forces GCC or Clang to use colors in their
messages, useful when building with ninja
Both options default to OFF (though when building with make in a
suitable terminal, GCC and clang automatically use colors)
(no UBSAN, that requires hardlinking the game with the executable,
which is not possible in the SDK..)
it could happen that UIs are added to the internal list twice,
and also that the last UI wasn't removed from the list when a new one
was focused fast enough.
That should work better now, I hope I didn't break anything..
this was so broken, I think if anyone ever tried to use __DATE__ or
__TIME__ it must've crashed, either from free(curtime) (which was the
return value of ctime()) or from things like token[7] = NULL when token
was a pointer (to a single element!).
I think it could work now.
Also fixed potential memory leaks in cases that don't "return" anything
Some entities wrote the handle from gameRenderWorld->AddLightDef()
into savegames and reused it after restoring it.
That's a bad idea, because at that point the handle most likely belongs
to something else (likely some idLight). The most visible issue this
can create is that the flashlight may not work correctly after loading
a savegame with flashlight on, when it happens to alias a light that's
updated each frame to (mostly) being off..
The correct way to handle this (THAT FOR SOME REASON WAS ALREADY
IMPLEMENTED IN D3XP BUT NOT THE BASE GAME - WHY?!) is to get a fresh
handle with AddLightDef() when restoring a savegame - unless the handle
was -1, which means that the light didn't exist when saving.
fixes#495
Some level scripts in d3xp (erebus4, erebus4, phobos2) use
$entity.startSoundShader( "", SND_CHANNEL_BODY );
(or whatever channel) to stop the sound currently playing there.
With s_playDefaultSound 1 that results in a beep..
Added a special case to that event implementation to call StopSound()
instead when the soundName is "" (or NULL)
- TestHugeTranslation() prints positions (and asserts only afterwards),
from "Work around assertion in alphalabs4, fix#409"
- idCompiler::CompileFile() prints proper filename
from "Fix usage of invalid pointer in idCompiler::CompileFile()"
These are now used by idStr::(v)snPrintf(), and in the future can
be used if a (v)snprintf() that's guaranteed not to call
common->Warning() or similar is needed (e.g. used during early startup)
idStr is used in both the main thread and the async sound thread, so
it should better be thread-safe.. idDynamicBlockAlloc is not.
Use realloc() and free() instead.
For some reason this caused a lot more crashes (due to inconsistencies
in the allocator's heap) with newer Linux distros (like XUbuntu 20.04)
and when using GCC9, while they rarely reproduced with GCC7 or on
XUbuntu 18.04
fixes#391
The only change really is adding SE_MOUSE_ABS to sysEventType_t.
In theory this breaks the ABI, but in practice I don't think mods use
sysEventType_t, and certainly not the ones above SE_MOUSE
(SE_JOYSTICK_AXIS doesn't make sense as Doom3 never supported Joysticks,
SE_CONSOLE is for input from TTY or whatever, and doesn't make sense
for mods to use in any way)
The change in Stub_SDL_endian.h is just me being paranoid.
(side-note: it *might* make sense to replace all that inline-asm in
Stub_SDL_endian.h with compiler intrinsics like GCC/clangs
__builtin_bswap* and MSVCs _byteswap_* - if they are supported in
all relevant compiler versions. GCC supports 32 and 64 bit swaps since
4.3, 16bit since 4.8 and 128bit since 11.x; clang supports 32 and
64 bit swaps at least since 3.0 and 16bit swaps since 3.2;
VS introduced _byteswap_ushort, .._ulong and .._uint64 in VS2003)
working around CMake (specifically CMAKE_SYSTEM_PROCESSOR) being useless
and related commits ported from dhewm3, up to including
"CMake: Fix typo in MSVC-specific CPU detection for x64"
- (Hopefully) better workaround for miscompiled cross products, #147
(-ffp-contract=off)
- GCC/Clang: Remove -fno-unsafe-math-optimizations
(redundant as long as -ffast-math isn't used)
- Fix runtime assert on PPC OS X
(-mone-byte-bool)
All pointer<->integer conversion truncation warnings I'm aware of are
now enabled for MSVC, to hopefully finally get that tool code 64bit-clean.
I had to adjust the idCVar code for this - it should've been safe enough
(highly unlikely that a valid pointer is exactly 0xFFFFFFFF on 64bit),
but triggered those warnings - now it should behave the same as before
but the warnings (which are now errors) are silenced.
Turns out that ${CMAKE_SYSTEM_PROCESSOR} is broken on Windows
(both for Visual Studio and for 32bit MinGW/msys shells which for some
reason seem to output x86_64 in uname -a)
idClipModel::axis is an idMat3 rotation matrix.
Usually it's an identity matrix, but if the player is pushed around by
an idPush entity it's modified and apparently can (wrongly) remain
modified, possibly when saving while idPush is active.
This seems to happen sometimes on the crashing elevator in game/delta1.
The fix/workaround is to reset it to mat3_identity when loading a
savegame.
like the dhewm3 version and the OS and architecture of the dhewm3
version that created the savegame.
Also added an internalSavegameVersion so be independent of BUILD_NUMBER
fixes#344
"Fix "t->c->value.argSize == func->parmTotal" Assertion in Scripts, #303"
had broken old savegames because the script checksum
(idProgram::CalculateChecksum()) changed, see #344.
This is fixed now, also the BUILD_NUMBER is increased so old savegames
can be identified for a workaround.
Don't use this commit without the next one which will further modify the
savegame format (for the new BUILD_NUMBER 1305)
.. with "no trivial copy-assignment".
All the cases I checked were harmless; as long as a class has no vptr
or owns memory (or other resources) that need to be handled properly
in the destructor, memcpy() and memset() aren't too dangerous.
If a "class" (object) in a Script has multiple member function
prototypes, and one function implementation later calls another before
that is implemented, there was an assertion when the script was parsed
(at map start), because the size of function arguments at the call-site
didn't match what the function expected - because the function hadn't
calculated that size yet, that only happened once its own
implementation was parsed.
Now it's calculated (and stored) when the prototype/declaration is
parsed to prevent this issue, which seems to be kinda common with Mods,
esp. Script-only mods, as the release game DLLs had Assertions disabled.
It corrupted the stack when called with buffers allocated on the stack
and numSamples that are not a multiple of four, apparently, by writing
4 floats too many, at least in the 22KHz Stereo case..
This caused the crash described in
https://github.com/dhewm/dhewm3/issues/303#issuecomment-678809662
Now it just uses the generic C code, like all platforms besides MSVC/x86
already do.
If you save, you get a message like "Game Saved..." which goes away
after a few seconds. This happens at the very end of idPlayer::Save():
if ( hud ) {
hud->SetStateString( "message", /* .. left out .. */ );
hud->HandleNamedEvent( "Message" );
}
And handled in hud.gui, "onNamedEvent Message { ..."
However, if you save again before it's gone, it'll be shown after
loading the savegame and not go away until you save again..
This works around that issue by setting an empty message after loading
a savegame.
The underlying problem (which is not fixed!) seems to be that the
transition GUI command (that's executed when hud.gui handles the
"Message" event that's used to show this message) is probably not
properly saved/restored so fading out the message isn't continued
after loading.
idStr::StripFileExtension() (and SetFileExtension() which uses it) and
others didn't work correctly if there was a dot in a directory name,
because they just searched from last to first char for '.', so if the
current filename didn't have an extension to cut off, they'd just cut
off at any other '.' they found.
So D:\dev\doom3.data\base\maps\bla could turn into D:\dev\doom3
(or, for SetFileExtension(), D:\dev\doom3.map)
While at it, I made most of the idStr code that explicitly checked for
'\\' and '/' (and maybe ':' for AROS) use a little
"bool isDirSeparator(int c)" function so we don't have the #ifdefs
for different platforms all over the place.
On Windows, ID_INLINE does __forceinline, which is bad if the function
calls alloca() and is called in a loop..
So use just __inline there so the compiler can choose not to inline
(if called in a loop).
This didn't cause actual stack overflows as far as I know, but it could
(and MSVC warns about it).
(This includes "Fix ID_MAYBE_INLINE on non-Windows platforms")
