Subtle bug ?

[Dimitri]

Hi Daniel,

while working on data given in cartographic coordinates (big numbers) I realized that I had an issue that I Initialy thought was simply a visualizing effect : some perpendicular banding in the data.

I think it actually has to do with a loss of resolution (even though I wasn't offered the option to translate the data).

So I did a simple test :
. Take one small good quality dataset with sub-cm resolution in local coordinates (i;e. all below 1 cm)
. Add with an external program a translation of 100000 m in X and Y
. Load in CC : I'm not given the option of translating the data, so I think I'm good
. Translate back in CC -100000 m in X and Y
. Make a closest point comparison between the original data and the back-and-forth translated one : I end with an average difference of 3 mm. And visualy the points are not in the exact same position.

. You can check that when you do the back and fort translation in CC you get the same result. By the way, when you do a + 100000 m, you'll see that the visualisation is "strange" (hard to explain !).

Maybe you should lower the threshold at which the user is proposed to have a translation (10^4 ?)

Cheers

Dimitri

[daniel]

Yes it's due to the 32 bits floating representation. See an older post here: viewtopic.php?f=10&t=544#p1453.

We could of course lower the warning limit. But I wonder if we should instead release the 64 bits version with double precision now that all machines have at least 4 or 8 Gb of RAM...