[Kazan]

yes i am

the only diff between rc2c and rc2d at the moment is
A) RC2c caches BSP occasionally when it shouldn't, fixed in RC2d
B) RC2d has the version of the compiler implemented - which will force it to recompile all models from earlier versions of PCS2
C) RC2d has a minor cosmetic fix for SOBJ bbox display

once there are more significant fixes i'll pop it out..

whoever was having collision detection problems with rc2b pls try rc2c

[chief1983]

It seems to have fixed the problems with the Protector and Cylon Assault Frigate from the BSG Alpha pack.  I'll probably re-export all the models now just to make sure it's all fixed, or maybe just wait and hear what everyone else says about it.

[Whitelight]

WoooHooo..

Collision detection fixored.   
Big thanks, to all the people working on PCS2.

[Koth]

I wanted to reconvert this redone Colossus as its collision detection was nonexistant but the RC2c build always crashes upon converting. The funny thing is it crashes at a different subopbject nearly every time. I don't know what to do. Purging the BSP cache didn't help either. Does anybody have an idea how I might fix this one?

[attachment deleted by ninja]

[Water]

I wanted to reconvert this redone Colossus as its collision detection was nonexistant but the RC2c build always crashes upon converting. The funny thing is it crashes at a different subopbject nearly every time. I don't know what to do. Purging the BSP cache didn't help either. Does anybody have an idea how I might fix this one?

I deleted Debris 6,7,8 and 10 - it converted fine.

One (or all of them) have some modeling errors

Edit: Deleting just Debris8 allows RC2c to save without crashing.

[Nuke]

in my expirience i find most model bugs are due to bad modeling, and over conversion. i figure every format you pass through costs you between 5% and 50% of your model integrity.

[Koth]

I wanted to reconvert this redone Colossus as its collision detection was nonexistant but the RC2c build always crashes upon converting. The funny thing is it crashes at a different subopbject nearly every time. I don't know what to do. Purging the BSP cache didn't help either. Does anybody have an idea how I might fix this one?

I deleted Debris 6,7,8 and 10 - it converted fine.

One (or all of them) have some modeling errors

Edit: Deleting just Debris8 allows RC2c to save without crashing.

Thank you very much!

[Kazan]

give me the version of that model that is causing crashing

[Koth]

Look at the attachment above.

[Kazan]

lol didn't notice the attachment

i'ts crashing because PCS2 is for some reason miscalculating the tree size so it's writing off the end of the allocated bsp buffer

[Koth]

Could you fix that, please?

[Kazan]

[sarcasm]no.. i'm just going to leave it that way![/sarcasm]

of course i'm going to try and ****ing fix it

[Koth]

Good, I already thought I would have to convert it again from the .cob file and that would be a pain for a model with so many turrets.

[Kazan]

possible you will want to.. this problem could indicate some corrupt geometry

[Koth]

If it is only some debris I can just delete it and prevent unnecessery work.

[Talon 1024]

Small feature request: Automatically purge the BSP cache upon startup, and possibly, whenever you load a new model.

[Kazan]

Small feature request: Automatically purge the BSP cache upon startup, and possibly, whenever you load a new model.

*faceplam*

did you read the PCS2 wiki article entry on the BSP cache?

it's a memory cache on a per-model basis, so when you startup PCS2 there is no cache, when you unload a model it's cache (if it exists) is unloaded

[edit]
the geometry for debris8 is signficiantly corrupt, i can confirm this in truespace and the fact that every senitnel variable that i attached that should have caught this didn't.

[Koth]

Thanks, then I will just delete it. I don't think that anyone will notice a missing piece of debris as the Colossus isn't destroyed very often.

[Kazan]

yeah.. what i don't get is how it's managing to crash PCS2 when i put in the following checks

if offset >= buffer size, terminate
added two flags to tree nodes: counted and used, counted = true after been counted by the size calculate, used=true after being written to buffer.  during writing check if it's already been written, or not counted, skip if either

and it still managed to walk off the end of the buffer.

[edit]
added more thorough detection and it found the overflow

[Kazan]

got it!

it's not corrupt geometry, it was an error in calculating the size of the required buffer.

fixed

[edit]
RC2d posted

Note: the buffer size calculation bug would give you an overly sized buffer normally, but occasionally as in the case of the big C could give an undersized buffer apparently.

Size of coli86.pof from july beta build: 950kb
size of coli86.pof from rc2d: 882kb