In the last few weeks I've played again through the game and kept an eye
on small oddities. And there're a lot of them. For example some GUIs and
videos getting stuck after the first frame (issue #192) or being unable
to protect the guy with the lamp in Alpha Labs 3. Some digging proved
that most - if not all - of these problems are caused by the compilers
optimization level. When build with -O2 both g++ 8.1 and clang 6.0.0 are
producing working code. g++ 8.1 with -O3 has some small, hard to notice
oddities, clang 6.0.0 with -O3 shows a lot of them. Since there's not
measurable difference between -O3 and -O2 just go down to the later:
x doom_o3.txt
+ doom_o2.txt
+----------------------------------------------------------------------+
| + |
| + * |
|x * x * |
| |_____________________|___A______M__A_________M___|_||
+----------------------------------------------------------------------+
N Min Max Median Avg Stddev
x 5 173 178 177 176.4 2.0736441
+ 5 176 178 178 177.2 1.0954451
so even with many loud sounds the overall volume isn't reduced by
OpenAL - apparently OpenAL scales down all sounds temporarily if the
mixed result would be too loud or sth like that.
Just sending all sounds to OpenAL with a lower volume prevents that from
happening (just set your system speaker volume a bit higher if needed).
This problem was especially noticable when shooting at metal walls with
the shotgun (each pellet produces an impact sound so it gets kinda loud)
Sometimes memory was allocated with new[] but freed with delete instead
of delete[], which is wrong.
And there were some small memory leaks, too.
Furtunately clang's AddressSanitizer detected all that so I could easily
fix it.
(There seem to be some more small memory leaks which are harder to fix,
though)
On FreeBSD, the game used to crash when loading the last level of RoE
(d3xp), while loading models/david/hell_h7.ma.
The problem could be reproduced on Linux whith #define USE_LIBC_MALLOC 1
and clang's AddressSanitizer.
Turns out that this file specifies a vertex transform for a non-existent
vertex (index 31, while we only have 0-30) and thus the bounds of
pMesh->vertexes[] are violated.
I added a check to ensure the index is within the bounds and a Warning
if it isn't.
It should work now. If however it turns out that more files have this
problem, maybe .ma is parsed incorrectly and we need a differently fix.
(Should) fix#138
as we do int buttonIndex = ev.button.button - SDL_BUTTON_LEFT;
it's only consistent to do if(ev.button.button < SDL_BUTTON_LEFT + 8)
it doesn't really make any difference as long as SDL_BUTTON_LEFT is 1,
but this way it's safe for SDL3 or whatever future version might break
the ABI.
__builtin_trap() causes an illegal instruction and thus the process
can't resume afterwards.
raise(SIGTRAP) only breaks into the debugger and thus allows to
"ignore" the assertion while debugging.
The assertion in idBounds::operator-(const idBounds&) was triggered
from idWeapon::Event_LaunchProjectiles() (ownerBounds - projBounds)
It only happened when using the BFG.
So I added a check to make sure calling operator- is legal.
I guess this also caused #122
The first row of AZERTY-Keyboards (used in France and Belgium) doesn't
have numbers as keys but ², &, é, ", ', (, -, è, _, ç, à, ), =
(with small differences between France and Belgium).
For some of those keys we don't have keycodes - and neither does SDL2.
See also https://bugzilla.libsdl.org/show_bug.cgi?id=3188
As a workaround, just map those keys to 1, 2, ..., 9, 0 anyway, as those
are keys Quake2 already knows (and those chars are printed on the keys
too, for typing they're reachable via shift).
This workaround only works for SDL2, as SDL1.2 doesn't have Scancodes
which we need to identify the keys.
This should obsolete one part of pull request #135
It's an SDL_TEXTEDITING event which we seem to get on Windows whenever
the Window gains focus (or is created). I think it can be safely
ignored, so that's what I do.
I also changed how those warnings are printed - as a hex number now,
because they're defined as hex numbers in the SDL source and it's easier
to find out what kind of event it is this way.
When ingame, Shift-Esc would open the menu and another Shift-Esc the
console. Now it immediately opens the console and only Esc without
Shift opens the menu.
The fullscreen guis pretend to be 640x480 internally, also for the mouse
cursor position. So adding the actually moved pixels (when playing the
game at a higher resolution) to the GUIs cursor position makes it move
too fast.
To fix that I detect (hopefully that check is reliable!) if the
idUserInterfaceLocal instance is a fullscreen GUI and if so scale the
reported mouse moved pixels with 640/actual_window_width and
480/actual_window_height.
If res_none (event with .evType == EV_NONE) is returned,
idEventLoop::RunEventLoop() will assume there are no more events this
frame => pending events will be delayed til next frame (or later if
again res_none is returned in the meantime).
So res_none shouldn't be returned just because there was an SDL event
we didn't care about or we did care about but don't generate a doom3
event for (but toggle fullscreen or something).
Instead we should just fetch and handle the next SDL event.
