close#141
The problem was that the shadow matrices were setup in BeginFrame, but
GraphicsView calls SetCamera after BeginFrame. The solution was to move
the shadow matrix setup into CRenderer::SetCamera.
This caused a second problem because RenderReflections/RenderRefractions
used SetCamera to change the OpenGL matrices. Solved this problem by
distinguishing explicitly between the camera used for the scene as a
whole and the camera used to configure OpenGL. The latter can be a
virtual camera for shadow or reflection rendering and similar
render-to-texture effects.
This was SVN commit r4330.
note: previous fix (clamp to 0 in screen_to_client) causes spurious
messages if mouse moves outside of the window (which is bad). "client"
coords are allowed to be negative, but "idealized client coords" (which
are defended by assert >= 0 or is_in_window) aren't.
fixes#146
This was SVN commit r4324.
- Removed a debug assertion that could occur in windowed mode if you
moved your mouse above or to the left of the window (client x and y >=
0); instead, negative client coordinates are now clipped to 0. (The
coordinates can't actually stay negative because we are casting them to
uints, so that would give very large values.)
This was SVN commit r4323.
Also split Entity.cpp further, moving the rendering code to
EntityRendering.cpp and the JS interface code to
EntityScriptInterface.cpp. This makes each individual file smaller
(about 1000 lines still!) and lets them be compiled independently.
This was SVN commit r4314.
hopefully found last confusion between screen/client coords (now prefix
the former with screen_*)
SDL passes around client coords in message as Uint16, int in API calls,
Window gives/takes screen space POINT (LONG x,y)
This was SVN commit r4311.
This avoids stuff like
/usr/include/c++/4.0.0/cmath:85: error: template with C linkage
/usr/include/c++/4.0.0/cmath: In function 'float std::abs(float)':
/usr/include/c++/4.0.0/cmath:92: error: declaration of C function 'float
std::abs(float)' conflicts with
/usr/include/c++/4.0.0/cmath:88: error: previous declaration 'double
std::abs(double)' here
This was SVN commit r4291.
Changed player-id code a bit so the entity and actor and unit should
stay in sync more often. (The entity/actor/unit mixing still looks a bit
dodgy and unreliable, though.)
Simplified console help code.
Allowed init/shutdown to be done with the simulation/world/etc parts
disabled (so the actor viewer can load faster).
This was SVN commit r4289.
- Territory boundaries are now offset slightly and drawn in player
colour, so you can see whose territory you're on the border of.
- Removed some duplication in the code by modifying a loop slightly.
This was SVN commit r4264.
Atlas: Added ActorViewer. Moved GL canvas into separate class for shared
use. Disabled message-handling callback while blocked on the game, and
stopped creating dialog boxes inside the game thread in order to avoid
deadlocks (hopefully). Support multiple Views (for independent sets of
camera/update/render code). Recalculate territory boundaries when
necessary. Changed default list of animations to match those currently
used by actors.
# Tidied up more code.
Moved some more #includes out of .h files, to minimise unnecessary
compilation.
MathUtil: Deleted unused/unuseful macros (M_PI (use PI instead), M_PI_2
(use PI/2), MAX3, ABS (use abs)).
ObjectManager: Removed some ScEd-specific things.
Unit: Moved creation out of UnitManager, so units can be created without
adding to the manager. Changed CStr8 to the more conventional CStr.
app_hooks: Removed warning for setting multiple times.
win: Restored SEH catcher.
GameSetup, GameView: Removed RenderNoCull, because it doesn't seem to do
what it says it does ("force renderer to load everything") since we're
loading-on-demand most stuff and it doesn't seem especially useful since
we'd prefer to minimise loading times (but feel free to correct me if
I'm wrong). (And because it crashes when things need to be initialised
in a different order, so it's easier to remove than to understand and
fix it.)
PatchRData, Renderer: Work sensibly when there's no game (hence no LOS
manager, water, etc).
LOSManager: Use entity position instead of actor position when possible.
TerritoryManager: Allow delayed recalculations (so Atlas can issue lots
of move+recalculate commands per frame).
Cinematic: Non-pointer wxTimer, so it doesn't leak and doesn't have to
be deleted manually.
This was SVN commit r4261.
Initially saved a little bit of time by changing it to use vertex
arrays, and then realised that disabling point-smoothing saved much more
time (from 70ms per frame on cantabrian_generated down to 2ms, which is
worth any quality loss)
This was SVN commit r4256.
.. by removing support for WAV. see rationale in source; in short, we
don't need it and it's not worth the minimal trouble of adding freealut
dependency.
entityhandles: remove old comment
This was SVN commit r4247.
CBound operator+= updated the minimum value or the maximum value, but
never both. (I think the operator+=(CVector3D) has always been buggy
(because it assumes min<max, which isn't true of empty bounds), and I
propagated the error to operator+=(CBound) ages ago.)
Added test cases, and fixed.
This was SVN commit r4244.
-Auras now take additional parameters from XML, containing a tag for r,
g, b, and a. No line thickness parameter yet.
-For territories, added function to frustum for determining if line
segment passes through the frustum.
This was SVN commit r4242.
Premake: Changed to use PCH when compiling tests.
CxxTest: Got rid of /Wp64 warnings. Changed TS_ASSERT_STR_EQUALS to
print both strings on failure.
win_internal: Added the bits that CxxTest::Win32Gui needs (because
Win32Gui is now compiled with PCH and uses our sysdep headers).
Disabled test_cache_policies, because it took long enough for me to read
its name.
This was SVN commit r4237.
LightEnv - has byte order marks
Texture - conv from int to GLint wasn't allowed
JSConversion - to use isfinite you need the std:: namespace
(this is temporary... atm I don't know why you can't call isfinite)
ScriptableComplex - when implementing methods it did not like default
values (they were removed from the cpp file)
timer.h - simple change
This was SVN commit r4231.
Added code to requestLowLevelPath that modifies the final waypoint to
use exactly the coordinates where the player clicked, instead of the
rounded-to-nearest-tile coordinates from the A* pathfinder. A more
complete fix required eventually might be to make the A* engine itself
aware of fractional locations within a tile but for now this seemed like
a simple thing to do.
Also fixed a bug that was causing the final step of the path not to be
added: the code was adding an ORDER_PATH_END_MARKER but no
ORDER_GOTO_NOPATHING for the final point. The ORDER_PATH_END_MARKER
doesn't seem to do anything useful here so the unit never went that last
step.
Also fixed a bug with "pass-through-allies" that was causing units to
avoid allies anyway: although the pathfinder assumed you could pass
through allies, when you actually bumped into one while executing a goto
instruction, the collision test used did not check for
pass-through-allies so you ended up avoiding allies anyway. This might
not be too horrible if it works in practice (since the avoidance was
done using simple local work, not a full repath) but I've removed it for
now since it might cause problems with formations, when avoidance will
not be trivial.
This was SVN commit r4226.
tricky issue with isfinite: <cmath> also stomps on that (C99) macro,
instead of only C89 libc. fix: include math.h instead. rationale is
documented.
This was SVN commit r4224.
- Modified Xeromyces to no longer automatically convert element and
attribute names to lowercase, so that we can have camelCase names. We
should now be able to convert all the multi-word entity properties, like
pass_through_allies, to camelCase, like passThroughAllies, which is more
consistent with the rest of our JavaScript naming conventions. To
support the existing code that assumes lowercase element names, I made
the getElementID and getAttributeID methods (used in the EL and AT
macros) ignore case, and I changed any code that directly accessed
element names to use the right case. CEntityTemplate now converts
Names_LikeThis to names_likeThis (changing each separate "word" in the
name to camelCase). Changed the version letter in XMB filenames from A
to B to support this without requiring people to delete old XMBs.
- Enhanced the pathfinder's handling of contact paths, resulting in a
very large speedup for actions like attacking, construction, etc. The
problem was that the pathfinder used to not count a given state as the
goal unless it was exactly coincident with the target location. This is
fine when you order a unit to go exactly to a certain spot, but if
you're ordering a unit to build, gather or attack something, then the
target tile is impassable (because your target is there) and therefore
the pathfinder never declares a state final. As a result, the pathfinder
tries hundreds of extra tiles in case there is a long path that gets to
the goal, and after failing to find any path that reaches the goal, it
gives you one to the closest point it got to. To fix it, I made the
pathfinder take into account a radius around the goal in which it's OK
to be, which depends on the size of the target unit and the range of
your action.
This was SVN commit r4186.
ScriptableComplex.inl is now only #included in CPP files that need it
(files that use functions from it directly), which is a total of 5
files, so recompiling after modifying it should be a lot faster.
This was SVN commit r4176.
- Jan and I cleaned up CJSComplex and put all the function bodies and
helper classes in ScriptableComplex.inl. This also required minor
changes to other files.
- Removed a debug message from Technology.cpp.
This was SVN commit r4173.
- u_anim_name disambiguation
- fix implementation of finite (use our fpclassify instead of compiler's
routine)
- ä -> ae
- workaround for MAP_ANONYMOUS
- fix GLint in ogl.cpp
- add include for SIZE_MAX in string_s
- avoid PIC clobbered error and speed up rdtsc a bit
- add include for stat
This was SVN commit r4170.
add new JSUtil, automatically included via SpiderMonkey.h.
move JS_ASSERT there and rename JSU_ASSERT.
move scriptglue's REQUIRE_PARAMS there and rename to JSU_*
replace if(argc < x) bail() code with those macros.
This was SVN commit r4169.
Jan and I implemented a number of changes:
- Converted SClassSet to vector.
- Refactored get/set ClassSet from string into methods in CClassSet.
- Turned pass-by-value string params in getCollisionObject, CEntity
constructor and CEntityManager into pointers.
- Simplified processChooseMovement.
- Merged CalculateRun and CalculateHealth into a single, simpler
CalculateRegen + helper functions.
- Changed the way regen works so the rates read in the XML act as rates
(before, they were the number of seconds until the entity will be fully
regenned, which is a bit counterintuitive).
This was SVN commit r4167.
- Added the ability to specify modifiers as percentages, by typing for
example "<value>10%</value>".
- Changed the way required techs work so you need any of them rather
than all (which is how Andrew originally had it, I think, except I
changed it to all because techs with 0 prereqs counted as unresearchable
and I thought there was a logic error).
- Techs will now correctly declare themselves unavailable if a pair tech
is in progress of being researched.
This was SVN commit r4162.
entity.cpp: fix indenting (was using space instead of tab)
replace all instance of & operator=( with NO_COPY_CTOR
trace: fix bug reported by michael: if trace file is huge, indices would
keep on being recorded but data was cut off after a certain limit,
leading to invalid pointers (into garbage data)
This was SVN commit r4158.
Entities will now leave all their auras and cancel all their productions
when they change ownership, which will be useful when we have
convertible buildings.
This was SVN commit r4156.
- Added name and player JS properties for use in scripts.
- Remoed IsJSFirst since it doesn't make much sense for techs that also
apply to newly created units.
This was SVN commit r4154.
replaced all (*) CStr / CStrW by-value params with const reference. hoo
boy.
please always perform this optimization (actually standard idiom) when
writing the code - it takes little work, tells the next guy that the
string won't be modified, and makes a large performance difference.
(* where possible.. a few require other changes and will follow later)
This was SVN commit r4151.
Techs: Made a separate list of technologies for every player, rather
than array fields in CTechnology. This also involved changing the
tech-related JS functions.
Bugs:
- PS_MAX_PLAYERS should be 8 (the way it's used for array sizes, etc
indicates that it's the maximum possible number of players ever, but it
used to be 6 while the game had 8 players).
- When you changed a CJSSharedProperty that was inherited, m_Inherited
was set to false, so it was no longer inherited by subsequent entities
you created. They got initialized to garbage values as a result.
This was SVN commit r4138.
* ScriptableComplex.h: big bang (several hundred ms) for little buck:
made all CStrW params const CStrW& instead. this should *always* be done
as a matter of principle, unless the string actually needs to be
modified (in which case compiler will tell you)
* adts: slight improvement to hash table perf: change expansion
threshold to 50% full, and implement quadratic probing (as suggested by
matei)
This was SVN commit r4122.
Bars and rank icons now scale with zoom level. I've left in the old
render functions for now, but they can be removed later.
This was SVN commit r4120.
entitymanager: remove duplication, move refcnt && !destroyed check into
helper function in preparation for caching result of that calculation
scriptglue: improved timer calibration
This was SVN commit r4113.
if( !unified[(*itKey)-UNIFIED_SHIFT] ) accept = false;
This was causing an index out of bounds when the key at *itKey was
AND'ed with the HOTKEY_NEGATION_MASK and thus became greater than 65536
and way out of bounds of the array. For now I added a check that *itKey
< HOTKEY_NEGATION_MASK, but maybe this is not the intended usage.
This was SVN commit r4111.
cheezy but low-overhead. call startXTimer(i) and stopXTimer(i), where i
can be 0..9 for now. results are displayed via debug_out at end of
program run.
This was SVN commit r4095.
replace it->member with variable=it->member for easier debugging. also
skip the AddProperty redundancy check in release mode instead of only
final builds
This was SVN commit r4093.
- Modified CJSSharedProperty so it can be inherited. It now acts as our
"static property" class with a pointer-to-member field. It did much of
this before but it wasn't used much because of the way XML properties
were loaded.
- Modified CEntity and CBaseEntity so that virtually all properties are
now CJSSharedProperty's added with AddClassProperty in ScriptingInit.
- Removed m_run SEntityAction field, replacing its fields with separate
variables, so we can have proper pointer-to-members for them.
- Modified CTechnology to use CJSComplex::Get/SetProperty so that
attributes no longer need to be "registered" for use in techs.
This was SVN commit r4086.
made many (*)of the bools in CEntity into a bitfield (flags). this saves
a lot of space inside CEntity.
* only those that are not specified in XML and/or accessed via JS
This was SVN commit r4082.
Territories: Fixed memory leaks.
Atlas: New section-toolbar. Fixed colour of brush grid. Moved open/save
into File menu. Added 3D-ish lighting control. Added 'busy'
notifications.
This was SVN commit r4079.
moved some fields (e.g. size of health bar) out of CEntity and into
BaseEntity so they can be shared between entities (instead of duplicated
in each).
TODO: is it ok to remove AddProperty for those properties in CEntity?
(BaseEntity is still adding them)
This was SVN commit r4078.
big thanks to philip for seeing the problem! :D
cause: not popping FPU stack. didn't happen on my athlonXP presumably
due to different FPU usage (possibly in drivers): pushing 8 values could
clear the stack and mask overflow. the cause was not due to compiler
(and FLD/FSTP param passing) because Philip uses VC2005 as well, but saw
the crash (on P4 CPU)
fixes#130
This was SVN commit r4072.
Entities with traits.is_territory_centre == true will act as territory
centres, and territory areas are calculated and displayed on the
minimap. It remains to display these areas in the game view and to make
Civ Centres "socketable" so you can build them on Settlements to claim
them.
This was SVN commit r4070.
* debug_write_crashlog and debug_dump_stack are now responsible for
detecting reentrancy (reported via new ERR_REENTERED error code).
* export debug_error_message_build to allow unit test of stack dumper
* split+clean up debug_display_error to allow this.
* error_description_r now returns buf for convenience
* ia32: fix typo causing disassembly to fail
* wdbg_sym: bugfix causing incorrect debug_walk_stack return value.
prevent recursive locking, provide locked version of
debug_resolve_symbol. add skip_this_frame for convenience.
refs #130
This was SVN commit r4067.
ia32: fix overly zealous error return in ia32_get_call_target
sysdep: more correct definition of fpclassify() (takes into account
argument size)
This was SVN commit r4055.
remove old instrumentation, fix warnings, add measured timing results
Loader: bugfix: no more memory leak whenever callback returns a warning
code. this was a loophole in the callback return value docs.
This was SVN commit r4048.
