kicad-developers team mailing list archive

[jp charras <jp.charras@xxxxxxxxxx>]

Le 29/04/2013 10:39, Lorenzo Marcantonio a écrit :
> On Mon, Apr 29, 2013 at 02:56:47AM -0500, Dick Hollenbeck wrote:
> > Please state the file and line number please.
>
> OK, one example:
>
> In PCB_BASE_FRAME::GlobalChange_PadSettings (pcbnew/globaleditpad.cpp)
> on line 232 pad_orient is computed from the pad orientation and the
> module orientation. pad_orient is an int.
>
> Later in the function it's used for pad->SetOrientation (after being
> used for filtering), so it loses precision. You could say that's a bug
> in the global pad dialog, but shifting the blame doesn't fix the
> problem.

Currently, pads and footprint orientations are in 0.1 degrees.
In .brd files they are stored in O.1 degrees (for historical reasons).

In new code, orientations are (slowly) moved from int to double. to 
allow a better resolution than 0.1 degree, *in the future*.

If the decision to allow more than 0.1 degree is taken, dialogs *and 
documentation* need to be updated (not just dialogs).

Currently, there is no bug.
Just we keep in mind when writing new code the fact one day, the 
orientation could be finer than 0.1 degree.

At least it means the boards are no more saved in old .brd format.

Currently there is no truncation in pads angles.


> In DRC::doTrackDrc the angle is computed using ArcTangente which returns
> an int. Not strictly related but it helps (given that just the other day
> someone talked about numeric instabilities in the DRC). Again, probably
> nobody would notice it since tracks are mostly multiple of 45 degrees.
> However in DRC::checkClearancePadToPad the bounding pad polygon is
> computed from the truncated angle so it could possibly give wrong
> results (maybe with huge pads :D)
>
> > The rest is bullshit.
>
> Your opinion, not mine. I hope somebody else think the same of me.
>
> If angles are defined to be integers you can (usually) put them in
> doubles; if angles are 'promoted' to doubles and you actually *can* have
> fractional angles (i.e. fraction of decidegree) you should check that
> they aren't truncated on the way (when it matters; for screen drawing
> probably a truncation is not significant). Otherwise you could simply
> say that the resolution is one decidegree and use a double for
> convenience (but then why promote it to int in the first place?)
>
> Just decide what an angle is.


--
Jean-Pierre CHARRAS