[OSM-talk] conclusions from my postgres / postgis experiments

[David Earl]

> -----Original Message-----
> From: talk-bounces at openstreetmap.org
> [mailto:talk-bounces at openstreetmap.org]On Behalf Of raphael Jacquot
> Sent: 23 November 2006 09:23
> To: talk at openstreetmap.org
> Subject: [OSM-talk] conclusions from my postgres / postgis experiments
>
>
> here's what I found during my little^Wextensive experiment:
>
> 1) we have tons of duplicate segments
>      ((s1.from = s2.from) AND (s1.to = s2.to)) OR
>      ((s1.from = s2.to) AND (s1.to = s2.from))
> 2) we have 4000+ degenerate segments
>      (s.from = s.to)
> 3) we have ways with no segments
> 4) we have degenerate ways where the way is neither a proper linear
> feature nor a proper closed polygon, but rather an unwidedly spaghetti
> monster. most of those furthermore involve a loop of some sort...

I'm not sure (4) is necessarily wrong or undesirable. I certainly started
creating some of these quite deliberately before I understood better how the
rendering works. But I shouldn't have to subvert the actual structure to the
needsof a particular renderer. A typical example is a street which branches
many ways but has the same name throughout. I have come across two "streets"
so far in Cambridge which are basically huge roundabouts with branches
leading off at many points. Another example is a linear street with dozens
short side arms. Many, many dead end residential streets are linear with a T
at the end (indeed nearly all, though most are just short turning arms which
I haven't represented, except where they are longer than a car's length).
Many also end in a little loop with green space in the middle, though these
can in fact be represented as linear (with the last point incommn wth
another point in the way).

So the question is, how should these naturally spaghetti-like entities
actually be represented. More recently (for the above, for example) I've
been splitting them up into ways that are linear. But this has disadvantages
as well in that it is not immediately (mechanically) obvious that they are
all part of the same entity.

I think it would be better to allow the spaghetti and have the intelligence
built into the renderer to make sense of these. It is semantically more
useful, creates a better index (gazeteer).

The other case is where ways end up with segements out of order (which at
present osmarender causes to destroy the names by splitting the text). But
if a renderer understands spaghetti it can understand these too.

David