yade-dev team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

> "it is believed": who?
Heh, that was my question. ;-)
Is it just me or is there a consensus?

> For the sake of consistency (Cell, O.cell &c), I would say "cell
> geometry", not "period geometry". (The same sub 2.)
Ok.
> I think it is better to say how it works than that it would "mess the
> simulation big time".
This comment was not supposed to be part of doc, of course. Is it not
clear for you how it breaks things?
>
> 1. One view is that if assigning trsf would just change the cummulated
> deformation (and refSize as well, of course, but not hSize),
No it doesn't have to change refSize, since this one is not used. See
resetTrsf function.

> without changing geometry at all; going this way will make it more
> difficult to set the geometry based on a transformation matrix (not
> very difficult, though).
How more difficult? T being the transformation matrix : Hsize*=T.
Can it be simpler?
>
> In both cases moreover, refSize must be kept in sync with the
> reference hSize column norms.
Nope, in (1) we can get rid of refSize, or move it to OGLrenderer if
really needed there.

> Nothing conclusive from my side whether 1. or 2. should be preferred.
> Perhaps others can give their opinions?
I think Chiara gave a conclusive example recently :
In (2) :
If everything is axis aligned, making current state the reference needs
to modify trsf _and_ refSize.
If the cell is sheared (be it initially axis-aligned), that is no more
possible. One has to maintain Hsize like this :
tempHsize=Hsize
trsf=identity
Hsize=tempHsize

In (1) :
In all cases :
trsf=identity
(in the c++ sense, i.e. what is currently obtained with resetTrsf)

If giving a matrix is demanding too much for setting an axis aligned
box, there can be a variant taking only a vector. More or less what
setRefSize is doing now, but without the dependance on refSize (which,
again, is useless and is here only for backward compatibility).

Bruno