-
Notifications
You must be signed in to change notification settings - Fork 36
/
TODO.txt
205 lines (168 loc) · 7.83 KB
/
TODO.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
Ask
===
- dCollideCylinderTrimesh() deosn't appear to recompute the AABB (explicitly via recompiute() or implicitly via getAABB)
However, box-trimesh collision DOES recompute it: see dCollideBTL() or dCollideTrimeshPlane()
- Was there something with contact sorting? dCollideConvexTrimeshTrianglesCCD() seems to violate that
(swaps last contact with current) -> Only if length is 0 ...
- collision_libccd.cpp:768 -> "tempContact.side2 = i;" -> "tempContact.side2 = indices[i];"
- dJointAddPUTorque (...) : Can anyone point me to the implementation? I don't think there is one.
--> All reported on 2023-06-05
Before release:
===============
- Before release: Disable DEBUG
- mvn animal-sniffer:check
- mvn versions:display-dependency-updates
- mvn versions:display-plugin-updates
- Release
- After release enable DEBUG
Things left out for 0.5.0 (missing from 0.15.1 vs 0.16.2)
=========================================================
Not ported:
- resource_control.*
- simple_cooperation.*
- Updates to DxWorldProcessContext
- ThreadedEquationSolverLDLT* and *Cooperative* is missing from:
- fastldltfactor(_impl).*
- fastlsolve(_impl).*
- fastltsolve(_impl).*:
- fastvecscale(_impl).*:
- objects.h/.cpp / obstack.h/.cpp
Java Version
============
- We should aim for Oreo or later (Android 8.0, released 2017, API Level 26: https://apilevels.com/)
- We use animal-sniffer (respects desugaring by D8): https://github.com/open-toast/gummy-bears
- Android Runtime replaced Dalvik as default Runtime in Android 5: https://en.wikipedia.org/wiki/Android_version_history
- Android 8 (Oreo) is about 8% share:
- https://apilevels.com/
- https://www.composables.co/tools/distribution-chart
- https://medium.com/androiddevelopers/support-for-newer-java-language-apis-bca79fc8ef65
- Support vor never Java features depend on Android Gradle Plugin 4.0.0 -> "Desugaring"
-> Does that also work with OSGI?
-> This was release in June 2017 ->
- Android 11 support "up to" Java 13 (Question: complete 13 or partial 13??)
- 2018: https://jakewharton.com/androids-java-9-10-11-and-12-support/
March 2023:
General cleanup:
- update DVector3 safeNormalize to use max() and then compare to 0. ALso see: https://github.com/JOML-CI/JOML/issues/66
DOC:
- replace @sa with @see
- replace \li with <li>
dJointGetAMotorNumAxes: Unused @param 'num'
dJointGetAMotorAxis: Unused @param 'rel'
dJointGetAMotorMode: Unused @param 'mode'
DIFFERENCES:
- ode4j has No OPCODE, only GIMPACT. ODE defaults to OPCODE.
- ode4j LCP uses ROWPTRS=false
- No coop-multithreading, no memory arenas management
- SAP-Space uses different implementation (merge-sort i.o. radix-sort)
- DrawStuff is slow. The implementation would require significant changes to
be a fast as the C/C++ version.
- The OO-API is the main API. The static CCP-like API is only available to
simplify porting.
- A custom cpp2java library is used for porting the code.
- Much of the math has been moved to a separate package. dVector3 and their
like now provide much own functionality.
- The test harnesses now partly use JUnit 4, instead of the C/C++ test library.
- TLS Thread Local Storage has not been ported. Most likely, there is no need to
port this.
- The Java implementation is 'thread-tolerant' (CORRECT TERM???) by avoiding
any static variables.
That means it should be possible to process different dWorlds in parallel,
and possibly even multiple spaces in parallel, if the dWorld is not modified (verify!).
- Contrary to the policy in C/C++ ODE, I prefer in Java return immutable DVectors when
calling getXXX(), rather than having to pass in a DVector to be overwritten. The
reasons are partly specific to Java:
- Returning an immutable is very straight forward using an immutable interface
- I may avoid unnecessary object creation if the returned value is only used briefly
and only for reading
- It avoid possibly fatal typos, because getter and setter are distinguished by signature,
not only by one letter: dGeomGetPos(g,p) / dGeomSetPos(g,p) --> p = g.getPos() / get.setPos(p)
- Above that, I think it is normal coding practice in Java to return a value if a method has only
one result.
* On the other hand, this can cause problems, because the "immutable' Object may be modified
while it is used, either through multithreading, or simply by other calls to the Object.
* Also, it may NOT be modified, because the user does not know whether he gets an actual field
or only a calculated result back.
Cleanup uses of dVector.v, especially in e.g.
- dxBody
- dxJoint
dxPlane._p -> Change from double[4] to dVector3?
Check and remove references (and commented out occurrences) of:
- dSetZero()
- Try to avoid dPAD()
Remove commented out references to dUSE_MALLOC_FOR_ALLOCA
CPP API should not use internal classes ?!?!?!? Also avoid casting !!!!!
-> Call DJoint j; j.getXY()
TODO
- DOC: @sa -> @see
- DOC: @li -> <li>
- change DxJoint node[0]/[1] to _node0/_node1
- Use ThreadLocal DVector3 Pool to avoid synchronization?
- DEMO: remove references to CPP4J
- Implement auto-allocation of dContacts.
dContactGeom() -> auto alloc
dContactGeom(10) -> alloc w/ fixed limit
dContactGeom(10, bFixed) -> (auto-)alloc w/ bFixed limit
- Test for thread-tolerance!! No static !?!!?!!
- performance: dxGeom should use LinkedIdentityHashSet(?)! (spaceRemove())
--> Update QuadTreeSpace accordingly!
- TRIMESH et al. Make Colliders to singletons, avoid instantiation in dxHeightfield
- disable DEBUG modes
- replace concatenated dVector math with specific methods, remove set() calls
- replace methods that take x,y,z with ones that take dVector3
- PerfTest and replace 'double[] v' in dVector3 with 'double v0, v1, v2;'
- get rid of all _tome stuff and firstjoint/firstbody
- Check that 'unsigned' is always assigned with fabs(y); !!!
- rename all classes to have capital 'D'?
- Check that org.ode4j.ode.* does not reference org.ode4j.cpp !!!
- Check that org.ode4j.ode.* does not reference org.ode4j.*.internal.* ?!?
- Check that org.ode4j.demo.* does not reference org.ode4j.*.internal.* ?!!
- Get rid of Thread stuff in API_OdeInit / OdeInit (?)
- rename add/sum/setDiff to eqAdd(?)/eqSum/eqDiff
- rename xxx to retSum/retDiff
- deprecate RefDouble in APIs!!
- Think about Radix-sort in SAP SPACE
- In SAP SPACE, use sort for all three dimensions and then prune Pairs
that are not in all Dimensions?
Bigger plans:
-------------
- Migrate OPCODE TRIMESH library
- Implement GJK algorithm, see:
http://opende.sourceforge.net/wiki/index.php/How_Collision_Detection_Works
http://en.wikipedia.org/wiki/Gilbert-Johnson-Keerthi_distance_algorithm
-------------------
Migration guide -> OLD, 0.3.0 ?
-------------------
Use QuickStep. Its's faster then Step. StepFast is less stable
dMass
=====
- dMass._c is now a dVector3, but was a dVector4
ID
==
d...ID classes have been removed, use d... instead.
The d...ID.id() method, that used to return the backing object is not necessary
anymore, because the user is always dealing with the actual objects.
OdeHelper
=========
Instead of 'new', Ode objects have to be created via the static methods in OdeHelper.
Z
dContact
========
dContact contact[N];
int n = dCollide (o1,o2,N,&(contact[0].geom),sizeof(dContact));
contact[i].xxx
-->
dContactBuffer contacts = new dContactBuffer(N);
int n = dCollide (o1,o2,N,contacts.getGeomBuffer());
contacts.get(i).xxx
- CONTACT(dContactGeom* p, int skip)
deprecated, instead, please use: dContagGeomBuffer.get(skip);
nearCallback
============
Add the following:
private dNearCallback nearCallback = new dNearCallback() {
@Override
public void call(Object data, dGeom o1, dGeom o2) {
nearCallback(data, o1, o2);
}
};