yes, it's likely to be something in the rasterizing, or above it in nr-arena-shape. Problem is, the bug doesn't appear on linux x86, so it's not endian-related, and since the win build is compiled with gcc, it's not a "different compiler, different results" problem. that leaves the system-dependent math primitives, like sin and cos, or the nasty case of an intersector bug, but i find that dubious since it works fine on OSX and linux. anyway, my knowledge of win32 is too limited to tackle this bug; sorry...
Do you know any code saying "if something, do not draw the stroke"? Since the stroke disappears completely, this is likely to be such a "draw or not draw" switch. I know of one, which you introduced lately (the check for width > 0.01 in nr-arena-shape) and I tried to disable it but this did not help. Maybe there are others checks like that somewhere?
_________________________________________________________________ MSN Premium helps eliminate e-mail viruses. Get 2 months FREE* http://join.msn.com/?pgmarket=en-ca&page=byoa/prem&xAPID=1994&DI...