 |
A big fat warning should be displayed to mappers in any case. Another quick workaround would be to dynamically add "inline = 1" when a non-orthogonal rotation matrix has been detected. According to NBohr, this might decrease performance, though, since the entity is converted to brushes. |
 |
Inlining affects physical properties no less than rotation hacking: http://forums.thedarkmod.com/topic/18242-model-import-help-crystals/#entry392812 Also it happens during dmap (you end up with a bloated .proc if you inline a lot), so if you wanted to apply it to existing maps you'd have to recompile them and then test for unwanted behaviour. That's not a quick workaround any more. SteveL reported shadow errors from inlining, as well as a huge slowdown to dmap if it's used heavily within a visleaf: http://forums.thedarkmod.com/topic/16678-what-are-current-max-limits-for-the-following/page-3#entry359239 |
 |
IIRC the wiki already tell the mapper about the fact, that only the visual model is affected by the non-orthogonal part of the rotation matrix. However, if the visual model can be modified like that, the more important question is why shadows and collision are not affected. Using non-orthogonal transformations to adjust a models size to the geometry in your mission is extremely useful opposed to creating a new model for each usecase, and I have used it in the past, too, although I restricted on minimal size deformations to avoid the issues with that approach becoming too noticeable to the player. It might therefore be desireable to identify the code pieces responsible for that specific aspects and changing it so they accept such transformations. |
 |
Started a lengthy discussion about the problem on forums: http://forums.thedarkmod.com/index.php?/topic/20180-rotation-hack/ Also, added one more issue due to this: 0004779. It is reproduced be running Penny Dreadful 2, and teleporting: setviewpos 5300 1100 -700 (the big golden keys underground are culled away sometimes). |
|
|
Added 0004232 as related, since it explains why some models have ('\1' + skinName) appended to their name. |
|
|
Committed the change: svn rev 8513: fixing models of rotation-hacked func_static-s svn rev 8514: dmap complains about rotation-hacked entities While there is cvar g_rotationHack which controls this new fixing, it only takes effect when the game is started. Then the value is saved into savegame files and persisted for the whole gameplay. |
|
|
One more change in svn rev 8517. Now if dmap detects rotation-hacked func_statics, it forces the dmapper to set g_rotationHack to 0 (turning off the implemented fix). Without this additional annoyment, there is high chance that some future maps will contain a shitload of rotation-hacked func_statics without any testing under "g_rotationHack 0". This is very bad, because if we decide to remove the fix which I implemented here in future, such maps will become unplayable. |
|
|
One more change in svn rev 8525. I fixed R_RadiusCullLocalBox for non-orthogonal rotations. It was reponsible for clipping away rotation-hacked entities which were enlarged by rotation. Seeming no more culling problems remain (for static non-shadows objects). P.S. Ironically, there was a "FIXME" comment at this place =) |
|
|
Yet another change in svn rev 8526. Don't do anything with with func_static-s having "bind" spawnarg. Usually the "bind" master changes location of the func_static. Thus, it is not "static" and messing with its rotation is dangerous. |
|
|
Reverted most of the changes in svn rev 8657 on trunk. Namely, the following commits were reverted: r8513 Now func_static-s with bad rotation are forked and transformed. This is done under g_rotationHack cvar. Hopefully, this will remove almost all issues that rotation-hack causes. r8517 Throw error during dmapping if rotation-hacked func_static is detected, unless mapper has disabled g_rotationHack. The reason for it is to ensure that mappers either keep their maps clean, or test their maps r8526 Don't embed rotation into func_static-s having "bind" spawnarg. Usually the entity it points to dynamically changes the location of the entity. The following commits remain: r8514 Let dmap complain about rotation-hacked entities. Note that all func_static-s get only one global warning, but other entities get individual warnings. Perhaps we should refine this over time... r8525 Fixed R_RadiusCullLocalBox to work properly for non-orthogonal matrices. As the result, rotation-hacked entities are no longer culled away erratically. In particular, dmap will report all rotation-hacked stuff as warnings, but will not throw errors. There are two reasons for doing so: 1) Related issue 0005186 about rotation hack in stock prefabs. 2) A cleaner idea of proxy models was proposed: https://forums.thedarkmod.com/index.php?/topic/20180-rotation-hack/&do=findComment&comment=445031 |
|
|
Committed engine support for proxy models in svn rev 8658 on trunk. Also committed the test map with two proxy models in svn rev 15892 on trunk. |
|
|
Committed a small preliminary change in rev 8890. In rev 8891, committed a new way of converting rotation-hacked entities on-the-fly. Now all entities are convered without exceptions. Conversion results in "model_*" spawnargs pointing to automatically generated proxy models in real files. See also: https://forums.thedarkmod.com/index.php?/topic/20180-rotation-hack/&do=findComment&comment=449752 |
|
|
There is one problem with new approach, which blocks e.g. WS4 FM from loading. The problem happens when the following conditions hold: 1) entity is moveable 2) entity is rotation-hacked 3) entity has no "clipmodel" spawnarg 4) entity's "model" has .cm-file nearby, with same name as the original model 5) the original model cannot be converted to CM e.g. because it has too many vertices I would say the first two conditions together already make no sense, and is most likely mapper error. I have detected about 20 FMs which have any .cm files except for the main one for the map. Let's hope that most of them are not rotation-hacked. The plan is to fix FMs before next release. One reliable way to do it is to make automation start every FM (I have already done it), and look at FMs which don't boot. |
|
|
I tried to start every mission, and found four more non-booting FMs due to this. Here is the full list: 1) Hack-rotated moveable with .cm: ws4_warrens --- described above, .cm in the FM rightful --- same, but .cm in core files 2) Hack-rotated animated model (md5): antr elixir_1_3 siegeshop |
|
|
For the history: the loading errors have all been fixed by repacking FMs. More details here: https://forums.thedarkmod.com/index.php?/topic/20548-rotation-hack-strikes-back/ Now the small fix during beta: svn rev 9074: now proxy models are saved to tdm-base-dir/models/_roth_gen/xxx.proxy. |
- Anonymous
