Re: [Pcb2gcode-devel] autoleveller

[Mathias K. <mk...@gm...>]

Hello Nicola,

 yes of course it is a bad hack as I wrote.

So I wanted to know what is really going on and found out the size of
the error corresponds to the value of the 'dpi' option. Therefore I
believe it could just be quantisation error from the outline bitmap in
surface.cpp.
The patch below fixes the issue for me ( again it enlarges the probe
area, but only 4 dots of the outline bitmap this time ). Your
extrapolation proposal might even be better because it does not assume
so much about the max. possible error, but I'm not into the pcb2gcode
code enough to do it right now.

diff --git a/board.cpp b/board.cpp
index c8c1f9b..b9a673f 100644
--- a/board.cpp
+++ b/board.cpp
@@ -119,10 +119,10 @@ void Board::createLayers() {
             shared_ptr<Isolator> trace_mill =
boost::static_pointer_cast<Isolator>(prepared_layers.at("front").get<1>());
             ivalue_t radius = trace_mill->tool_diameter / 2;
             int passes = trace_mill->extra_passes + 1;
-            min_x -= radius * passes;
-            max_x += radius * passes;
-            min_y -= radius * passes;
-            max_y += radius * passes;
+            min_x -= radius * passes+2./dpi;
+            max_x += radius * passes+2./dpi;
+            min_y -= radius * passes+2./dpi;
+            max_y += radius * passes+2./dpi;
         }
         catch( std::logic_error& e ) {
             min_x -= margin;

Bye,

  Mathias


On 30.01.2015 15:07, Nicola Corna wrote:
> I don't like your solution, it forcefully increases the area, especially when there are only 4 probing points at each angle
> Since the problematic points are very close to the border, we can simply do a linear interpolation of the closest points
> Send me your Gerber file, I'll use it as a test
> Thanks for your bug report
>
30 Gennaio 2015 12:27, "Mathias Katzer" <mk...@gm...> wrote:
>> It is because some pad in the Gerber file sticks out by ~0.01in from the
>> bounding box returned by the importer.
>>
>> For myself I enlarged the probing area by one ProbeDist in both
>> coordinates. Of course that does not solve the root of the problem, but
>> adds robustness. I guess the actual bug is in gerbv or kicad - don't
>> know yet.
>>