Re: [svg-developers] Marker differences - ASV vs Batik


Thanks Niklas for reporting that. I have looked into it and there was
indeed a bug in Batik which did the wrong math for the last two line
segments on the first path and  for the last two segments on the
polygon. I just commited a fix for that in the Batik code base.

Regarding the definition of the bisector 'vector' (i.e., the unit vector
defining the direction of the marker's x axis), I think it is implicitly
defined as the sum of the tangent vectors on the input and output
segments (at least this is how I read the spec) but it might be good to
clarify this and be more explicit. Something like: "if I is the input
segment's tangent unit vector and O the output segment's tangent unit
vector, then, the direction of the marker's x axis is the same as I+O.
If I+O is null, then the direction of the marker's x axis is given by
the vector orthogonal to I and forming a +90deg angle with I."


P.S: As Thierry pointed out, there is an issue  with the arc command in
the sample showed, but Batik seems right there.


Chris Lilley wrote:
  > On Sunday, December 1, 2002, 3:05:06 AM, Niklas wrote:
  > NG> (cross-post: be warned)
  > (wondering what is)
  > NG> Hi all
  > NG> When doing some experiments with markers I found some differences
  > NG> between Batik 1.5b4 and ASV3 and I'm unsure to which one is
  > NG> correct. Please help me out since I can't really judge based on
  > NG> the spec and I need to know where to send the bug report :-)
  > I tried it in CSIRO viewer and got a third result... seems to use the
  > only outgoing tangent, sometimes ..
  > NG> I used this file:
  > NG>
  > NG> This produces the following rendering in ASV:
  > NG>
  > NG> (note that the circled arrow should not be there at all. This is
  > NG> clearly a bug in ASV).
  > Yes. Perhaps the Q curve is being rendered internally as two curves ....
  > NG> In Batik it looks like this:
  > NG>
  > NG> As you can see, the rotation of some of the markers differs
  > NG> between the two implementations. Which one is correct?
  > I don't see why the markers in Batik (and CSIRO) flip by 180 degrees
  > for that first mid vertex on the lower left.
  > NG> The spec has the following to say about rotations on mid-markers:
"if there
  > NG> is a path segment coming into the vertex and another path segment
going out
  > NG> of the vertex, the marker's positive x-axis should point toward
the angle
  > NG> bisector for the angle at the given vertex, where that angle has
one side
  > NG> consisting of tangent vector for the path segment going into the
vertex and
  > NG> the other side the tangent vector for the path segment going out
of the
  > NG> vertex (note: if the tangent vectors are the same, the angle
bisector equals
  > NG> the two tangent vectors)".
  > NG> (
  > NG> This doesn't really help me but for someone more fluent in math
it might :-)
  > OK well the path segments part id obvious I hope, for example M L L Q
  > T a l z has two straight-line path segments generated by the two L
  > commands, then one curved path segment generated by the Q and so on.
  > For mid markers, drawn between two path segments, there is an incoming
  > path segment (the one nearest the start of the path) and an outgoing
  > path segment (the next one along, towards the end of the path).
  > Its easier to see with straight lines so consider the angle formed by
  > the two L segments. There is a tangent vector (a line at right angles
to the
  > line drawn by the first L) which, for a straight line, is the same
  > vector all along the line. There is another tangent vector for the
  > second L. The math verbiage above says that the marker lies along the
  > average of those two tangent vectors, ie a vector mid way between
  > them.
  > Once you get onto curves then the tangent varies along the curve so
  > the incoming tangent is a line drawn at 90 degrees to the curve just
  > before the join onto the next path segment, and the outgoing tangent
  > is a line drawn at 90 degrees to the curve just after the join; and
  > the marker is once again on a line half way between those tow -
  > dividing the angle between those two tangents into two.
  > Hmm it should say the bisector on the shortest arc, clearly there are
  > two possible bisectors ....  a possible explanation for the flipped
  > first marker ....

Received on Tuesday, 3 December 2002 04:46:45 UTC