@Jakob sorry what I wrote was confusing, I meant : vertices have strong
identity (same pointer), and triangles are just keys for the map so they
need to be very stable. A triangle pointer wouldn't solve the case. When I
encounter 3 vertex pointers A, B, C, then I need a "key" to determine if
triangle ABC (or CAB, etc.) is already in the map.
@Axel indeed I had not taken into account that the GC may "change the
pointers, while preserving ==" so basically yes ordering pointers looks
unreliable.
@Kulti checking equality of triangles by checking combinations of vertex
equality would be interesting but it would make the program really super
slow: instead of a O(1) map lookup, each existence check would cost O(n) to
traverse all the existing triangles, which is too slow because we have 1M+
triangles. This is because we can't use Triangle as a map key when equality
is computed by a func instead of ==.
@Egon those are nice workarounds, thanks. X,Y,Z are weak, they change all
the time.
I will try to minimize code clutter, minimize memory and cpu overhead,
maximize safety and readability. So I think the best tradeoff so far is to
add a field Vertex.ID, because I have the chance to do it, though it does
clutter a little a fundamental data structure of the whole program.
Thank you all for insights!
On Monday, August 28, 2017 at 3:20:39 PM UTC+2, Axel Wagner wrote:
>
> FTR, I don't think ordering by pointer (no matter whether using unsafe or
> not) is reliable and safe. The GC is free to move data around at any point,
> including during your sort.
> I don't think there are a lot of ways around that. You could basically
> build your own allocator and add a way for it to return a reliable id. e.g.
> it allocates a []Vertex and the id is the index of the vertex in that
> slice. You can calculate it using unsafe by (roughly. No guarantees here.
> Unsafe is subtle)
>
> func ID(v *Vertex) int {
> i := int(uintptr(unsafe.Pointer(v))) -
> int(uintptr(unsafe.Pointer(&Vertex[0])))
> if i < 0 || i > len(pool) {
> panic("not pool-allocated")
> }
> return i
> }
>
> (not that you need to take care to do both conversions in the same
> expression, to satisfy the unsafe.Pointer
> <https://godoc.org/unsafe#Pointer> rules).
>
> Apart from that, Egon gave a couple of good suggestions :)
>
> On Mon, Aug 28, 2017 at 3:02 PM, Jakob Borg <[email protected]
> <javascript:>> wrote:
>
>> In that case, maybe use *Triangle pointers instead of Triangle values,
>> and let that address be the identity for map purposes? Then the Vertex
>> order etc doesn't matter any more.
>>
>> //jb
>>
>> > On 28 Aug 2017, at 14:03, Val <[email protected] <javascript:>> wrote:
>> >
>> > That's actually a clever idea, I had not thought of that when
>> dismissing the vertex struct inspection. Thanks!
>> >
>> > But it turns out that X,Y,Z will change often during my program
>> execution: vertices move, but their identity (and the triangles identity)
>> must be strongly preserved.
>> >
>> > On Monday, August 28, 2017 at 1:42:35 PM UTC+2, Jakob Borg wrote:
>> > On 28 Aug 2017, at 13:13, Val <[email protected]> wrote:
>> > >
>> > > To achieve this, I consider normalizing the triplet so that all keys
>> of the map verify
>> > > A < B < C
>> > > for some arbitrary definition of < . This ensures that I can detect
>> any Triangle already present as key in the map, regardless the order of its
>> Vertex pointers.
>> >
>> > Can't you sort the Vertex slice by their X,Y,Z position? (You're then
>> not relying on the coordinates for identity, only for order. This breaks
>> for triangles that are actually lines or points, which may or may not be an
>> issue and can be handled separately...)
>> >
>> > //jb
>> >
>> > --
>> > You received this message because you are subscribed to the Google
>> Groups "golang-nuts" group.
>> > To unsubscribe from this group and stop receiving emails from it, send
>> an email to [email protected] <javascript:>.
>> > For more options, visit https://groups.google.com/d/optout.
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "golang-nuts" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to [email protected] <javascript:>.
>> For more options, visit https://groups.google.com/d/optout.
>>
>
>
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
