--- In kicad-devel@xxxxxxxxxxxxxxx, jean-pierre charras - INPG <jean-pierre.charras@...> wrote:
Unfortunately, grid size cannot be handled in decimils.
In pcbnew grid is handled as double.
This is because for many grids values in mm, using a value rounded in
decimils creates a cummulative error that can be important
(some mm for a 300 mm board)
So grid sizes must be given
- always in inches or decimils but using a floating point value, for non
integer values as needed for metric grids
- or as an integer in decimils or um units and give the used unit (um ou
decimil) as parameter
--
Jean-Pierre CHARRAS
Maître de conférences
Directeur d'études 2ieme année.
Génie Electrique et Informatique Industrielle 2
Institut Universitaire de Technologie 1 de Grenoble
BP 67, 38402 St Martin d'Heres Cedex
Recherche :
Grenoble Image Parole Signal Automatique (GIPSA - INPG)
Grenoble France
Actually, the best internal unit to use in pcbnew would be nanometers (1nm = 0.001um = 0.000001mm), because then both grid size and all coordinates, both metric and imperial, could be expressed with 100% accuracy with integers.
And we don't even need 64bit integers either - a 32bit signed integer would be able to hold coords (and grids) ranging up to +/-2.15 meters or +/-84.5 inches, which I think is enough for kicad :-)
If 4.6 m^2 (49.6 foot^2) of board space isn't enough, use 10nm (centimicrons or decananometers :-) to get a 21.5m/845" coord range (460 m^2 or 273 yard^2), which surely must be enough for everyone ?-)
1 mm = 393.700787401... dmil (ugly and inaccurate)
1 inch = 25.4 mm = 25400 um = 25400000 nm
1 mm = 1000 um = 1000000 nm
1 mil = 0.0254 mm = 25.4 um = 25400 nm
1 dmil = 0.00254 mm = 2.54 um = 2540 nm
The first conversion is the one that requires floats, and it will never ever be 100% accurate.
All the others are simple and 100% accurate integer conversions:
mm2nm(x) = 1000000*x
dmil2nm(x) = 2540*x
So, any metric value, down to 1nm (or 10nm) of precision, and any imperial value with dmil precision can be converted to nm (or 10nm == 1cu == 1danm (*1)) without _any_ loss of accuracy.
(*1) The use of multiple SI prefixes isn't actually allowed, but let's just bend the rules a little :-)
Øyvind.
