Fix memory corruption and assert failures in convex decomposition

[Malcolmnixon]

This PR fixes issue 85439 by changing the index pre-increment to post-increment so index 0 is not skipped and preventing a potential overrun past the end of the vertices vector.

Testing involved constructing a GLB file consisting of a single triangle:

Before this change the code allocates vertices and indices arrays of size 3 and writes the following:

Vertices (size=3)	Indices (size=3)
0 = unused	1
1 = [-1,-1,-1]	2
2= [1,-1,-1]	3
3 = [0,1,-1] (BUFFER OVERRUN)	unused

After this change the code writes the following:

Vertices (size=3)	Indices (size=3)
0 = [-1,-1,-1]	0
1 = [1,-1,-1]	1
2= [0,1,-1]	2

The collision shape with this new data appears correct:

offset_triangle.zip

[BastiaanOlij]

Makes sense to me, would like the original author to weigh in if there was a reason for this. This is not something people tent to do by accident, most likely there was other code here before that makes this make sense..

[Malcolmnixon]

Makes sense to me, would like the original author to weigh in if there was a reason for this. This is not something people tent to do by accident, most likely there was other code here before that makes this make sense..

It looks to have been introduced with 2ca94e5 and no code before it seems to implement the vertex-deduplication map trick. The older code uses duplicated vertices, but the loop doesn't seem to suffer from N+1 overruns.

[lawnjelly]

Good spot, must have missed this when we looked at #50404 .

[lawnjelly]

Makes sense to me, would like the original author to weigh in if there was a reason for this. This is not something people tent to do by accident, most likely there was other code here before that makes this make sense..

Pouley is probably unlikely to reply as he's moved on to other work, but this was implementing something I suggested about removing unnecessary conversion between longhand vertex format (with duplicated verts) and vertex + index format, if I remember right. Nothing super complex going on here.

[akien-mga]

Thanks!

[YuriSizov]

Cherry-picked for 4.2.1.

[yythlj]

I meet the problem same in 4.2.1 windows

[YuriSizov]

@yythlj Please open an issue with more details about your case.

[YuriSizov]

Cherry-picked for 4.1.4.