[OSM-dev] Note to the developers of editors :)

[Matt Amos]

On Mon, Apr 13, 2009 at 7:09 PM, Stefan de Konink <stefan at konink.de> wrote:
> Frederik Ramm wrote:
>> I don't think it should. The API makes no claim that a way is some
>> geometric object; a way is just a collection of nodes.
>
> Like discussed on IRC yesterday /my/ opinion is clear; if there is or
> will be an enforcement on length because of 'client/server' interaction,
> check for actual corruption must be trivial too add, an the only thing
> useful for our data collection.

and i'll re-iterate my argument for the benefit of those not on IRC at 3am ;-)

the length restriction is a pragmatic solution - its not the ideal
solution, but it fixes some problems without introducing too many
obstacles for the clients or users. the alternatives would require
many more complex changes and lead to annoying situations where very
long ways become almost uneditable due to their massive geographic
extent.

we already do checks for data corruption, unfortunately they're
vulnerable to race conditions. 0.6 will fix that in 4 days ;-)

>> If the API would start to do geometry inspection, then you'd have to add
>> loads of additional checks as well. For example for self-intersecting
>> areas or ways with length==0 even because first and last node, while
>> being different, have the same coordinates, and whatnot.
>
> Lets check for it :) (I'm serious) I was even surprised we seem to go on
> PostgreSQL but don't go PostGIS.

i don't consider consecutive duplicate nodes to necessarily indicate
corrupt data. its up to the client to interpret the user's intent -
and if the user genuinely wanted consecutive duplicate nodes then
thats fine by me.

>>> As always OSM fixer is on rampage to filter about 6000 plus ways,
>>> leaving 4492 matches, of them 405 become one-way-ways. Relations
>>> haven't been found duplicate in this way.
>>
>> As long as you fix things that are obviously completely broken, like
>> ways referring to deleted nodes, that's fine. But if you venture into
>> the "this makes no sense to me so I'll fix it", it might be better to
>> discuss your ideas on this list *beforehand* rather than just tell us
>> afterwards what your bot is doing (or has done).
>
> I have discussed this the previous night on IRC. And 'this makes no
> sense', did not come from me, but a few days ago from someone else. But
> please speak up if you know anything useful with sequential duplication
> for members. I cannot :)

having looked at a random subsample of these ways, they did seem to
make no sense. i assume they were created when a user didn't click far
enough from an existing node to create a new one and so a duplicate
reference to the existing node was appended onto the way. i can see
how this might happen multiple times and not be noticed by a user.

i can't think of any use for consecutive duplicate nodes, *yet*. there
might be people already using this for something, or we might find a
use for it in the future. for the moment, i consider this to be a
minor client UI bug which can be most easily fixed by the clients, not
by the server.

cheers,

matt