Right click map selection

[Dirk Hohndel]

Linus Torvalds <torvalds at linux-foundation.org> writes:

> On Sat, Feb 9, 2013 at 2:42 PM, Lutz Vieweg <lvml at 5t9.de> wrote:
>>
>> And I just experienced a problem with picking (which I only tried to
>> record for the tutorial video :-): The precision of the "pick" seems
>> to be rather low. When I return to look at the picked position, it
>> has moved about 30m away. Maybe just some numerical precision issue?
>
> The OSM interface seems to use 'float', which should be just barely enough.
>
> Using 'float' means that the possible precision of the interface is
> *great* when you pick positions in the Gulf of Guinea. So if you go
> diving on the island of Sao Tome, you can probably get centimeter
> precision.
>
> But if you happen to be diving in Hawaii or Fiji, "float' is abysmal.
> A normal "float' has 24 bits of mantissa (23 stored bits, but there is
> actually 24 bits of precision due to the dropped implicit leading
> one-bit), and in Fiji or Hawaii, about 7.5 of the bits are used for
> the "integer" part of the degrees of longitude. Leaving just 16.5 bits
> for fractional degrees. And that's getting borderline.
>
> But it should still be much better than 30m. In fact, it should be
> just about 1m, which should still be enough

16.5 bits gives you almost 10udeg resolution which at the equator is
about 1m. I'd call that sufficient to locate a dive site.  I think the
30m problem is more likely caused by issues with reading the correct
position from the osm widget.

Lutz, if you try to reproduce this, is the error always the same? Always
in the same direction? Maybe the 'hotspot' of the mouse isn't where we
think it is and the coordinates are slightly off because of that?

I'll play with it as well...

> Internally inside subsurface we use "micro-degrees", which means that
> the worst-case precision is constant (6 decimals of degree - roughly
> 10cm along the equator).

Which is better than what any commercially available GPS receiver will
ever give you.

/D