[OSM-dev] New OSM Test Data Repository

[Martin Raifer]

>> I am not sure yet what exactly the "valid" and "invalid" mean and
>> whether we need different outcomes. For the moment "valid" means more
>> something like: "The data looks good, could exist in a real database,
>> and everybody would agree on how to interpret it." While "invalid" is
>> everything else.
>
> I suppose there are many cases where it would make sense to allow an
> assessment somewhere between 'fully valid no matter how hard you look
> at it' and 'invalid beyond any consistent interpretation of the data'.
> It might make sense to say robust programs should be able to deal with
> certain data while data producing progams should avoid generating such
> data.

Yes. I'd add that any good consumer/production OSM-data-processor should  
(be able to) gracefully handle "slightly invalid" data. For example it  
should not segfault, abort or skip a linestring just because the way  
contains a node twice. The same tests could also be used against  
OSM-quality-assurance tools which of course should be able to detect the  
respective glitches.