On Mon, Jun 10, 2013 at 10:29 PM, Štěpán Turek <stepan.tu...@seznam.cz> wrote: > Hi all, > > Probably I found bug in nodes cost in vector network analysis (tested in G7, > probably it affects all branches). Problem is in lib/vector/Vlib/net.c: > > int cost; > ... > (* dglInt32_t)(dglInt32_t) & cost > > int cost is cast into dglInt32_t pointer. DGLib dglInt32_t is long. I am > relatively new to C so I do not understand the way it is casted and I can > miss something. I am reading it as cast address of cost int to dglInt32_t > and then to pointer to dglInt32_t. It does not make sense to me but it > compiles :-) > > The problem is when you have on your system different size of long and int, > it is probably also the reason why it was not noticed so far (on 32 bit > usually sizeof(int) == sizeof(long)). Because it seems that during this cast > there are added bytes to be same size as long. However these bytes have > random values and therefore results costs do not make sense.
I think the reason why this has not been noticed yet is that not many users are working with node costs. 64 bit systems with sizeof(long) > sizeof(int) are commonly used for quite a few years by now. The type name dglInt32_t is confusing because it is not necessarily a 32bit integer, it is an integer with at least 32 bits. > > If this change is made everything seems ok: > > dglInt32_t dgl_cost; > ... > dgl_cost = cost; > dglNodeSet_Attr(gr, dglGetNode(gr, (dglInt32_t) cat), > & dgl_cost); > > > Now there is: > > dglNodeSet_Attr(gr, dglGetNode(gr, (dglInt32_t) cat), > (* dglInt32_t)(dglInt32_t) & cost); > > I have attached also patch. Makes sense. I have applied the patch to all branches in r56680-2. Markus M _______________________________________________ grass-dev mailing list grass-dev@lists.osgeo.org http://lists.osgeo.org/mailman/listinfo/grass-dev
