Handling of orient="auto" markers

Hi all

When I was implementing my SVG renderer, the Marker section was one of the
head-scratchiest parts of the spec to try and understand and get right.  It
did not help that pretty much all SVG renderers have bugs (or "quirks") in
their marker handling so there wasn't any consensus to go on.

I notice the SVG 2 spec is a bit better, but I think the section on "auto"
oriented markers is still a bit confusing.

'auto'A value of 'auto' indicates that the marker is oriented such that its
positive x-axis is pointing in the direction of the path at the point it is
placed.
If the marker is a segment
marker<http://www.w3.org/TR/SVG2/painting.html#TermSegmentMarker>,
then the direction the marker is oriented is, if considering the incoming
and outgoing directions as unit vectors, in the direction of the sum of
these two vectors. If this sum is zero, then the marker is oriented in the
incoming direction.
Is this a typo? Seems like this should be referring to a vertex marker.
If the marker is on the first or last vertex of a closed subpath, [..snip..]
I think it is not obvious in all cases what the "first" and "last" vertex
is.  If the last point in the subpath does not coincide with the first
point, we are told to draw a line to the start point. In this case, which
point is the "last" point?  Also should we be drawing a start marker, an
end marker, or both?

My feeling that in a closed subpath, there is conceptually no real start or
end point. All vertexes should be considered mid points and only midpoint
markers should be rendered.  So my proposed replacement for this section
would be as follows: ------------------------------------------'auto' A
value of 'auto' indicates that the marker is oriented such that its
positive x-axis is pointing in the direction of the path at the point it is
placed.
If the marker is a *vertex marker*, then the direction the marker is
oriented is, if considering the incoming and outgoing directions as unit
vectors, in the direction of the sum of these two vectors. If this sum is
zero, then the marker is oriented in the incoming direction.
If there are two or more consecutive points in a path that coincide exactly
(both their x and y values are the same) then all the consecutive
coinciding points should be treated as if they were a single vertex.

If the marker is on the first vertex of a closed subpath, then the outgoing
direction is taken from the first path segment and the incoming direction
is taken from:

   - the last path segment of the subpath, if the last point coincides with
   the first point, or
   - the line drawn from the last point to the start point, if they do not
   coincide.

For closed subpaths, the marker definition to be rendered for all vertexes
is the marker specified by marker-mid. No start or end markers should be
applied. ------------------------------------------

That last paragraph might be a bit controversial, but for closed subpaths I
think it makes sense.  As I said, from my testing, renderers rarely agree
on how markers are positioned at the start/end of closed subpaths, so I
don't think making this change would affect too many people.

I also have a suggestion for an enhancement:
------------------------------------------ 'auto-start' A value of
'auto-start' behaves exactly the same as 'auto' except for the treatment of
the start marker.  If a subpath is not closed, then the start marker is
positioned so that it points in the exact opposite direction (inverse
vector) to the first path segment.
------------------------------------------

The idea is that would allow for the easy creation of double-ended arrows,
for example, without having to define two markers.
This is my first contribution to this list, so if I am an idiot, go easy on
me! :)
Paul