X-Authentication-Warning:	delorie.com: mail set sender to geda-user-bounces using -f
X-Recipient:	geda-user AT delorie DOT com
X-Original-DKIM-Signature:	v=1; a=rsa-sha256; c=relaxed/relaxed;
	d=googlemail.com; s=20120113;
	h=mime-version:in-reply-to:references:date:message-id:subject:from:to
	:content-type;
	bh=BFZHEI8uAjZJGH1dnu7KFC+a1PSH2KdwSvcgfuHPIfs=;
	b=XeK4RHr4mn4uiGY/goOmXihYFsZ4XT8FCiXTIBmEG9NVmfDv4k8ed/0sOsLL9qcD1l
	Slk4aQT/1f7LVaGyBogH3VCZWSxmbn7u+/q3gxXRHC1Lu6kfIJjiYfeZB046H0yN60gd
	9inm0VG7IxY7YV2H/RoVQbZlDYzzAFVNwMcVMyLZfhM5iSDccjsb7VcP7GqBAgAPUiro
	MAWrwHYQ96YgYo2mpLk3E2pJ8e6dFvMFW5AjswAng4PQjm+T/KoHQwP7PNtFNqedPJE2
	UbQHckgxLkR3MA5gWwR3+/T7EIArPy46QhqQpfGx4BobnV7GB5DzKKVUTd5uIeRvU3oW
	Rtxw==
MIME-Version:	1.0
X-Received:	by 10.202.85.146 with SMTP id j140mr6795300oib.4.1450814076074;
	Tue, 22 Dec 2015 11:54:36 -0800 (PST)
In-Reply-To:	<20151222184519.GO14853@abax>
References:	<20151222144719 DOT GN14853 AT abax>
	<CAJXU7q9yKKheYd5fnQz=cax=+rMmjDGNCuojLquq6zsiYZmRxw AT mail DOT gmail DOT com>
	<CAJXU7q9ufwTHKAZf2Dp-_S5FpAs5BmJSPatrWnCi8uCXmPeyVg AT mail DOT gmail DOT com>
	<20151222184519 DOT GO14853 AT abax>
Date:	Tue, 22 Dec 2015 19:54:36 +0000
Message-ID:	<CAJXU7q_wS=HMdUPRHAyu9q2ZKawRUr-aZXvqzedR5ct5f+B-kQ@mail.gmail.com>
Subject:	Re: [geda-user] Self-intersecting polygon appears because of
	intersection coordinates rounding
From:	"Peter Clifton (petercjclifton AT googlemail DOT com) [via geda-user AT delorie DOT com]" <geda-user AT delorie DOT com>
To:	gEDA User Mailing List <geda-user AT delorie DOT com>
Reply-To:	geda-user AT delorie DOT com
Errors-To:	nobody AT delorie DOT com
X-Mailing-List:	geda-user AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

Did you spot that there was some incorrect vertices being generated
for certain cutouts? (polygon.c)? I think my branch fixed some of
those..

The way to tell is dial down the number of vertices being used to
approximate a circle.. (can't recall, but possibly polygon.h?) You
will see that things get quite badly distorted if the bug is present.
Changing to (say) 4-8 segments should make it very clear if the bug is
still there.

So - we have:

1. Some bad geometry in the circular approximations.
2. Sometimes we get rounding errors, that make polygon edges which
should end up as exactly straight, out by 1 or 2 nm over their length
- which produces tiny little polygon slivers that are prone to
problems, and rounding.
3. Polygon routines that can (under certain circumstances - I forget
_all_ the details) - walk along the wrong contour, and produce
completely incorrect results.

I'd be curious to see you assessment as to whether your recent fix in
git HEAD is similar, different, or addresses a similar bug to those
addressed in my branch.

Many apologies to everyone for sitting on these for so long... I'd
guess there was a reason (such as that I wasn't 100% confident in my
fix - looks like there was still a few "hmm... ?" type comments left
in there.

Martin - I can't promise a lot of availability, but would you be up
for some shared effort on this (perhaps chat on IRC at some point?) to
try to dig to the bottom of these issues once and for all?

I doubt we'll solve snap-rounding (with any elegant algorithm - unless
we do a mad re-write), but according to Harry Eaton's old comments- It
appears he thought the iterative intersection + edge splitting
approach taken within PCB ought to have avoided the problem.


Peter


Peter

On 22 December 2015 at 18:45, Martin Beranek <martin AT mb5 DOT cz> wrote:
> Hi Peter,
>
>> I'd be interested to know how you got on running your board against
>> this old branch of mine:
>>
>> http://www.repo.or.cz/geda-pcb/pcjc2.git/shortlog/refs/heads/for_master2
>
> OK, tested now. Compiling from this branch directly, I do not have this
> problem on my test board, but there is a lot of changes including
> size/shape of the clearout around lines/pads as it looks like (I did not
> analyzed things in details yet, but it seems that coordinates of
> polygon vertices are slightly different).
>
> So I decided to take polygon1.c only and transplant it into the main pcb
> branch. Now, it seems that the buggy intersection calculation happens,
> but without any adverse effect on polygon rendering. It seems that order
> of collecting is different because of other changes and the collect does
> not run over problematic edges.
>
> But with small modification (letting ground fill into one previously
> clear area), I was able to make polygon disappear after component move
> in almost the same way as in the main branch.
>
> I do not have enough debug info processed yet, so it is maybe another
> vertex/edge causing the bug this time.
>
>
> While I was not able to reproduce the issue with the clean pcb version from
> your branch, I believe it is not fixed there, it would just occur for
> some different layout as lot of things related to polygons is changed
> and I am not getting the 'right' vortices (coordinates) there with my
> test board.
>
>
>
> Martin
>

- Raw text -