[postgis-users] Failing ST_Transform with Ross Ice Shelf polygon

[Marco Boeringa]

Sounds interesting. I think many users of PostGIS would be really glad 
to see something like this implemented if it could reasonably be done. 
Haven't tried the double cast via geography yet, but seems fun thing to 
check and see the result.

Op 7-11-2023 om 20:31 schreef Paul Ramsey:
> All that said…
>
> It would be possible to “fix” this, but it’s a scary black box.
> We already nudge geodetics back into place when casting from geometry 
> to geography (interesting workaround, take your reprojected result and 
> do a ::geography::geometry on it)
>
> https://github.com/postgis/postgis/blob/42f04a29effdd9e8280c7aba17420ba306fc73f4/liblwgeom/lwgeodetic.c#L3351
>
> For systems that we know are geodetics (and with modern proj we 
> generally know that) we could apply the nudge to the outputs. It would 
> make things slower (more logic) but it would only change those cases 
> where the coordinates are in fact out of bounds by a very small amount.
>
> P.
>
>> On Nov 7, 2023, at 11:22 AM, Paul Ramsey <pramsey at cleverelephant.ca> 
>> wrote:
>>
>> Nope.
>>
>> It can be quite reasonably argued that the answer is correct, and the 
>> problem is treating EPSG:4326 (a geodetic coordinate system with 
>> angular units) as if it was a planar system with cartesian units 
>> (spoiler: it is not that). In angular units, -180.0000000004 is 
>> ridiculously close to 180.0. You aren’t complaining about the other 
>> coordinates, like where 175.123456789 is coming through as 
>> 175.123456788. Why not? It’s the same error! :)
>>
>> I don’t know what it is about the math going through that fun CRS 
>> that is causing roundoff or even if it’s particularly large (I don’t 
>> think it is), but it is not at all unique to that system. You can 
>> generate data that is progressively offset from the original data 
>> doing nothing more exotic than going back and forth from WGS83 to UTM 
>> over and over and over.
>>
>> ATB,
>>
>> P
>>
>>> On Nov 7, 2023, at 11:16 AM, Marco Boeringa 
>>> <marco at boeringa.demon.nl> wrote:
>>>
>>> Thanks Paul,
>>>
>>> But is there a more definitive solution in PostGIS / PROJ on the 
>>> horizon in terms of future development? No one expects a perfectly 
>>> valid geometry that just happens to hit the projection boundary of 
>>> WGS1984 to come out garbled by doing a transform and back-transform 
>>> to the original CRS. I realize there may be technical challenges 
>>> here, but this will undoubtedly keep coming up many times in the 
>>> future, and likely has in the past, by other confused non-expert 
>>> users of PostGIS if nothing changes. It is really counter-intuitive 
>>> to need to use stuff like ST_SnapToGrid, ST_ReducePrecision or 
>>> ST_WrapX to "fix" something that goes right for 99.999% of all other 
>>> data. It also makes any needed code more convoluted.
>>>
>>> Yes, well, I know, storing data in WGS 1984 geometry may not be best 
>>> practice with this kind of globe spanning data, but it works for 
>>> most cases and I already cast to geography a lot to do stuff where 
>>> geography is really needed.
>>>
>>> Marco
>>>
>>> Op 7-11-2023 om 19:02 schreef Paul Ramsey:
>>>>
>>>>
>>>>> On Nov 6, 2023, at 3:39 PM, Paul Ramsey 
>>>>> <pramsey at cleverelephant.ca> wrote:
>>>>>
>>>>>
>>>>>
>>>>>> On Nov 6, 2023, at 3:33 PM, Marco Boeringa 
>>>>>> <marco at boeringa.demon.nl> wrote:
>>>>>>
>>>>>> Well, yes indeed that is what is happening, 180 came out of the 
>>>>>> reprojection steps as -180. Full output geometry below. Is there 
>>>>>> any way to prevent this behavior?
>>>>>>
>>>>>> Marco
>>>>>>
>>>>>
>>>>> Not really… Either snap to grid or reduce precision
>>>>>
>>>>> https://postgis.net/docs/ST_ReducePrecision.html
>>>>> https://postgis.net/docs/ST_SnapToGrid.html
>>>>>
>>>>> will get you back onto the dividing line (note that it is 
>>>>> at -180.00000000000014), but that won’t help in flipping -180 to 
>>>>> 180. For your particular case, applying
>>>>>
>>>>> https://postgis.net/docs/ST_ShiftLongitude.html
>>>>>
>>>>> will fix it, I think, though not in generality
>>>>
>>>> I think using
>>>>
>>>> https://postgis.net/docs/ST_WrapX.html
>>>>
>>>> would allow a more general purpose solution. At least one you have 
>>>> more control over.
>>>>
>>>> P
>>>>
>>>>>
>>>>> P
>>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20231107/555da91f/attachment.htm>