Efcore: NetTopologySuite SqlServer provider incorrectly translates Within query

[DanielStout5]

I have this query:

var points = new[]
{
    new Coordinate(neLng.Value, neLat.Value), // TR
    new Coordinate(neLng.Value, swLat.Value), // BR
    new Coordinate(swLng.Value, swLat.Value), // BL
    new Coordinate(swLng.Value, neLat.Value), // TL
    new Coordinate(neLng.Value, neLat.Value), // TR
};

var fact = NtsGeometryServices.Instance.CreateGeometryFactory(srid: 4326);
var poly = fact.CreatePolygon(points).Normalized().Reverse();

// This line is only added to demonstrate that local evaluation works correctly
var list = query.ToList().Where(x => x.Location != null && x.Location.Within(poly)).ToList();

// This does not filter to the correct results; it excludes values that should be included, and includes values that should be excluded
query = query.Where(x => x.Location.Within(poly));

The server-evaluated query is incorrect - it omits points within the polygon poly, and includes ones that are outside it. The locally-evaluated version ("bug demo") does return the expected results. They should be returning the same results since they are doing the same thing. query is IQueryable.

The values received for the bounds will be like this:

neLat   60.11131083608817
neLng   -74.22879847513246
swLat   47.2231665957533
swLng   -157.85672816263252

Further technical details

EF Core version: 2.2.3
Database Provider: Microsoft.EntityFrameworkCore.SqlServer

[ajcvickers]

Note for triage: looks like this could be SQL Server polygon direction again. /cc @bricelam

[DanielStout5]

Not sure this is relevant, but I did try:

• Specifying the points in the opposite order (counter clockwise)
• Not calling Normalize
• Not calling Reverse

As well as the different combinations of the clockwise order/normalize/reverse, and it would either immediately throw an exception (saying they were declared in the wrong order) or work as it does now.

[bricelam]

What is query? Can you attach a standalone repro with data that demonstrates the issue?

I also remember MSSQL having some unorthodox behaviors for things landing exactly on a boundary. Could it be that?

[bricelam]

Could be related to #15668, but it seems unlikely

[DanielStout5]

Query is IQueryable<Business> where Business has a NetTopologySuite.Geometries.Point property called Location.

I'll see if I can get the time to setup a repo that replicates the issue - I've already spent quite a few hours trying to figure out what was going on here, so that may have to wait a bit.

The locations are not exactly on the boundary - they are many kilometers away.

15668 doesn't look related to me but I'm no expert!

[JediLeeSkywalker]

Hello, I am one of colleague whose the reporter #15668:

Some experience of mine would help :

1. Type declare of the Location column in the SQLContext directs to be a "geometry"or "geography", if there is none ,the default will be a "geography". it looks like :
   entity.Property(e => e.Location).HasColumnType("geometry");
2. Detect counter clockwise then we can decide to reverse the polygon or not.
3. Or you could try to use Location.intersects(poly)?
4. Post some translated SQL or error message that are from [ASP.NET Core Web Server] window at bottom of Visual Studio when running the web app, you might get more help.

[DanielStout5]

Type declare of the Location column in the SQLContext directs to be a "geometry"or "geography", if there is none ,the default will be a "geography"

The Location is of type geography. I thought that was correct, since I'm dealing with latitude/longitude and not arbitrary shapes?

Detect counter clockwise then we can decide to reverse the polygon or not.

I'm not sure that's the issue, since I tried both clockwise and counter-clockwise. One way would always cause an exception, and I have it the other way. It doesn't cause an exception how I have it now, but it also doesn't return the results.

Or you could try to use Location.intersects(poly)?

I've tried Intersects as well. It works very similarly to Within - maybe slightly better, but it's still not returning the correct results.

Just like Within, it works perfectly if I execute Location.Within(poly) locally instead of on SQL Server.

I can move the map to include most of the correct pins:

But if I move the map up slightly, most of the pins disappear - even though they are still obviously within the polygon formed by the bounds of the map (as evidenced by the fact that Location.Within(poly) works when executed locally).

Post some translated SQL or error message that are from [ASP.NET Core Web Server] window at bottom of Visual Studio when running the web app, you might get more help.

There are no errors, but here's the SQL generated:

```
Microsoft.EntityFrameworkCore.Database.Command:Information: Executed DbCommand (1ms) [Parameters=[@__poly_0='0xE61000000104050000002FF82028D19B45401036CB24C1D550C02FF82028D19B...' (Size = 112) (DbType = Binary), @__p_1='0', @__p_2='50'], CommandType='Text', CommandTimeout='30']
SELECT [x].[Id], [... omitted]
FROM [Businesses] AS [x]
WHERE ([x].[Type] = N'Shop') AND ([x].[Location].STIntersects(@__poly_0) = 1)
ORDER BY (SELECT 1)
OFFSET @__p_1 ROWS FETCH NEXT @__p_2 ROWS ONLY
````

[JediLeeSkywalker]

Hello:
　　Is it equal to original points array? if you try to test the binary parameter of translated SQL in SSMS Query Analyzer :
DECLARE @g geography; SET @g = geography::STGeomFromWKB(0xE61000000104050000002...<--[the complete binary　code],4326); SELECT @g.ToString();
ref STGeomFromWKB

Is yours google maps? Maybe this one Pin disappear ? closest date

By the way, I am using EF core 2.2.4 ,NetTopologySuite.core 1.15.2

[DanielStout5]

Is it equal to original points array? if you try to test the binary parameter of translated SQL in SSMS Query Analyzer

I actually did want to test this, but to be honest I'm not sure how to get the full binary representation of the polygon since the SensitiveDataLogging for EF core only includes the first bit.

I've tried BitConverter.ToString(poly.AsBinary()).Replace("-", ""); and WKBWriter.ToHex(poly.AsBinary()); but they don't match what EF Core is logging. Any idea how to get the full binary data?

EF:    0xE6100000010405000000C174FC786E494640E87E886BD7F44FC0C174FC786E49...
Other: 01030000000100000005000000E87E886BD7F44FC0C174FC786E494640E87E88AB36A14FC0C174FC786E494640E87E88AB36A14FC0EE55BEB0AD5A4640E87E886BD7F44FC0EE55BEB0AD5A4640E87E886BD7F44FC0C174FC786E494640

Is yours google maps? Maybe this one Pin disappear ? closest date

It is Google maps, but I'm quite certain that isn't the issue, or else it wouldn't work when I change from SQL server evaluation to local. The only thing that changes is whether Location.Intersects(poly) executes in C# or on SQL Server. It works when executed in C#, but not when executed by SQL Server.

By the way, I am using EF core 2.2.4 ,NetTopologySuite.core 1.15.2

I'm using Microsoft.EntityFrameworkCore.SqlServer.NetTopologySuite version 2.2.3, and my version of .Net Core is 2.2.103

[JediLeeSkywalker]

1. I don't know VS 2017 or 2019 's way but I am sure you can get full binary representation by SSMS Profilter.
2. OK, for google maps , just for a chance.
3. I suggest you post constant values of the points array to the masters of this git, if you still not get solution.
4. By the chance, get WKT from google maps , then #15668
   Wicket for Google Maps API v3
   Android API about WKT
   Lazy way

[DanielStout5]

Good idea about the Sql Server Profiler. That did get me the full parameters.

I found two zoom levels (two different bounding polygons), which should share many of the same points. The first returns nothing, the second (the same area except more zoomed in, so it should return less) returns them all.

--declare @poly varbinary(112) = 0xE610000001040500000079DA50E81999434094D172A355B75AC079DA50E8199943404046CB8D963C37C0EAC7100794B84B404046CB8D963C37C0EAC7100794B84B4094D172A355B75AC079DA50E81999434094D172A355B75AC001000000020000000001000000FFFFFFFF0000000003
declare @poly varbinary(112) = 0xE6100000010405000000BC29F036B459444094D172A3B1AB55C0BC29F036B459444028A3E546336F46C0EDE38ED110A0484028A3E546336F46C0EDE38ED110A0484094D172A3B1AB55C0BC29F036B459444094D172A3B1AB55C001000000020000000001000000FFFFFFFF0000000003

select * from businesses where location.STIntersects(@poly) = 1

--select geography::STGeomFromWKB(0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0,4326);

0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0 is a location within both of the poly values, but it only gets returned by the second one.

When I try to convert that point via STGeomFromWKB, it throws this exception:

System.FormatException: 24201: Latitude values must be between -90 and 90 degrees.

Which doesn't make sense - neither coordinate is outside of that range (so even if long/lat was reversed, it shouldn't cause that issue!)

If I try to use STGeomFromWKB on the polygon, it says:

System.FormatException: 24115: The well-known binary (WKB) input is not valid.

It seems that the format I'm getting from the SQL Server Profiler (which starts with 0xE61...) isn't actually WKB, and I'm not sure how to convert it into a more readable format.

If I do this in C#: var bin = WKBWriter.ToHex(poly.AsBinary());

Then I can use that result in STGeomFromWKB in SQL Server:

SQL:
select geography::STGeomFromWKB(0x0103000000010000000500000070561335C66056C096218CB798B04440E0AC266A5CD947C096218CB798B04440E0AC266A5CD947C04D2EEDD5D1EA484070561335C66056C04D2EEDD5D1EA484070561335C66056C096218CB798B04440, 4326).ToString()

Output:
POLYGON ((-89.512097615131779 41.37966055242866, -47.698132771381779 41.37966055242866, -47.698132771381779 49.834528675852745, -89.512097615131779 49.834528675852745, -89.512097615131779 41.37966055242866))

But I don't think this result is actually useful - the problem isn't on the C# side which is where that WKB representation is coming from.

[bricelam]

It seems that the format I'm getting from the SQL Server Profiler (which starts with 0xE61...) isn't actually WKB, and I'm not sure how to convert it into a more readable format.

Correct, it's the SQL Server serialization format. You can turn it into WKT like this:

SELECT geography::Deserialize(0xE61...).STAsText()

[JediLeeSkywalker]

Sorry, My wrong:
DECLARE @g geography; SET @g =0xE6100000010405000000BC29F036B459444094D172A3B1AB55C0BC29F036B459444028A3E546336F46C0EDE38ED110A0484028A3E546336F46C0EDE38ED110A0484094D172A3B1AB55C0BC29F036B459444094D172A3B1AB55C001000000020000000001000000FFFFFFFF0000000003 ; SELECT @g.ToString() is right way.

DECLARE @g geography; SET @g =0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0; SELECT @g.ToString()

[DanielStout5]

Correct, it's the SQL Server serialization format. You can turn it into WKT like this:

Oh, that's great - thanks!

So here is an example of the polygon/point that doesn't work.

SQL:

declare @poly varbinary(112) = 0xE610000001040500000079DA50E81999434094D172A355B75AC079DA50E8199943404046CB8D963C37C0EAC7100794B84B404046CB8D963C37C0EAC7100794B84B4094D172A355B75AC079DA50E81999434094D172A355B75AC001000000020000000001000000FFFFFFFF0000000003;
declare @pt varbinary(112) = 0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0;

select geography::Deserialize(@poly).STAsText();
select geography::Deserialize(@pt).STAsText()
select geography::Deserialize(@pt).STIntersects(@poly)

Output:

POLYGON ((-106.86460195744286 39.1961031336586, -23.2366722699428 39.1961031336586, -23.2366722699428 55.442017443841749, -106.86460195744286 55.442017443841749, -106.86460195744286 39.1961031336586))

POINT (-63.57745 44.64129)

0

If I pop those WKT formats into bboxfinder.com, I get this map:

I think it's pretty clear that point is within the polygon, so why is STIntersects returning 0?

[bricelam]

I think you're looking for STContains...

[bricelam]

..err, I mean STWithin.

[bricelam]

I think STIntersects means anything that crosses the polygon boundary on SQL Server

[DanielStout5]

I've tried STWithin, STContains, STIntersects, STOverlaps, as well as STDistance <= 0, and they all return 0/false.

STWithin is what I originally started with.

If I use a slightly different polygon, then both STWithin and STIntersects return true.

So this polygon says it doesn't contain the point:

POLYGON ((-106.86460195744286 39.1961031336586, -23.2366722699428 39.1961031336586, -23.2366722699428 55.442017443841749, -106.86460195744286 55.442017443841749, -106.86460195744286 39.1961031336586))

While this one does contain it:

POLYGON ((-86.682717191817858 40.700812213198759, -44.868752348067858 40.700812213198759, -44.868752348067858 49.250513262530013, -86.682717191817858 49.250513262530013, -86.682717191817858 40.700812213198759))

Here's the map of the one that does work:

Very similar to the one I posted earlier that doesn't work!

[JediLeeSkywalker]

declare @poly varbinary(112) = 0xE610000001040500000079DA50E81999434094D172A355B75AC079DA50E8199943404046CB8D963C37C0EAC7100794B84B404046CB8D963C37C0EAC7100794B84B4094D172A355B75AC079DA50E81999434094D172A355B75AC001000000020000000001000000FFFFFFFF0000000003;
declare @pt varbinary(112) = 0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0;

select geography::Deserialize(@poly).STAsText();
select geography::Deserialize(@pt).STAsText()
select geometry::Deserialize(@pt).STAsText()
select geometry::Deserialize(@pt).STIntersects(@poly)

result:
POLYGON ((-106.86460195744286 39.1961031336586, -23.2366722699428 39.1961031336586, -23.2366722699428 55.442017443841749, -106.86460195744286 55.442017443841749, -106.86460195744286 39.1961031336586))

POINT (-63.57745 44.64129)

POINT (44.64129 -63.57745)

1

[DanielStout5]

select geometry::Deserialize(@pt).STIntersects(@poly)

That's interesting.. So if the point is Geometry instead of Geography, it is counted as being within the polgyon.

I could convert the stored Location from geography to geometry, but the problem is that the application uses distance queries.

select geometry::Deserialize(@pt).STDistance(@pt2)
select geography::Deserialize(@pt).STDistance(@pt2)

Result:

0.053120579816113

4393.5315056106

The latter is the correct distance in meters. The first is something else, maybe degrees?

[JediLeeSkywalker]

It is a easy way that change the poly switches x y to y x , or anyway to get poly object to real geography?

[DanielStout5]

It is a easy way that change the poly switches x y to y x , or anyway to get poly object to real geography?

I'm not entirely sure what you mean.

I create the polygon in my code based on the map bounds the frontend sends up:

var points = new[]
{
    new Coordinate(neLng.Value, neLat.Value), // TR
    new Coordinate(neLng.Value, swLat.Value), // BR
    new Coordinate(swLng.Value, swLat.Value), // BL
    new Coordinate(swLng.Value, neLat.Value), // TL
    new Coordinate(neLng.Value, neLat.Value), // TR
};

var fact = NtsGeometryServices.Instance.CreateGeometryFactory(srid: 4326);
var poly = fact.CreatePolygon(points).Normalized().Reverse();

I'm just not sure why switching the x/y would be necessary - I put the WKT result into bboxfinder.com and it matches what I expected.

I saw a comment a minute ago about setting the Z value of the points and polygon to 0 fixing the STIntersect. Did that not actually work?

[JediLeeSkywalker]

No it not work, I found the incorrect code that I put @g1.STIntersects(@g1), Z point not work.

this works:
declare @g1 geography = geography::STPolyFromText('POLYGON ((-106.86460195744286 39.1961031336586,-106.86460195744286 55.442017443841749, -23.2366722699428 55.442017443841749, -23.2366722699428 39.1961031336586,-106.86460195744286 39.1961031336586))',4326);
declare @g2 geography = geography::STGeomFromText('POINT (-63.57745 44.64129 0)',4326);

select @g1.STIntersects(@g2)

result:
1

[DanielStout5]

Hmm.. That polygon you're using is different than the original - the points are in a different order. Specifically the 2nd and 4th are switched (You're going counter-clockwise, I think?)

If I change the creation of the polygon to this:

var points = new[]
{
    new Coordinate(neLng.Value, neLat.Value), // TR
    new Coordinate(swLng.Value, neLat.Value), // TL
    new Coordinate(swLng.Value, swLat.Value), // BL
    new Coordinate(neLng.Value, swLat.Value), // BR
    new Coordinate(neLng.Value, neLat.Value), // TR
};

It still has the same issues as before. I'm not sure it's happening the exact same way, but at certain zoom levels and pan positions (i.e. certain bounding polygons) the STWithin/STIntersects still returns false where it shouldn't. Calling .Normalized().Reverse() has essentially no effect.

[JediLeeSkywalker]

I had searched what is different from STPolyFromText and STGeomFromText, and more about direction of polygon, I found diff is most closest this question.
It mentions Reverse order at last reply, or you could make the WKT reader , I wish NTS would do some work like STPolyFromTexts' doing.
Another saying ,get WKT by regular API, then you would get right polygon.

[DanielStout5]

I think this might be part of the issue:

Turns out SSMS has a Spatial Results visualizer:

declare @poly varbinary(112) = 0xE610000001040500000079DA50E81999434094D172A355B75AC079DA50E8199943404046CB8D963C37C0EAC7100794B84B404046CB8D963C37C0EAC7100794B84B4094D172A355B75AC079DA50E81999434094D172A355B75AC001000000020000000001000000FFFFFFFF0000000003;
declare @pt varbinary(112) = 0xE6100000010C3CA06CCA15524640A089B0E1E9C94FC0;

select geography::Deserialize(@poly).STAsText();
select geography::Deserialize(@pt).STAsText()
select geography::Deserialize(@pt).STIntersects(@poly)

select geography::Deserialize(@poly) union all select geography::Deserialize(@pt)

That little black dot is the @pt

So the point would be within the polygon if it had straight lines, but they connected by curves. Perhaps Reverse would just flip the polygon on the X plane - so it might work for the bottom points but wouldn't work for the top ones..

Do you know if there is some alternative way to give that polygon to EF Core that would change how it interprets the coordinates?

[JediLeeSkywalker]

I am not sure if this info could help. If I change compatibility_level of SQL Server Database to 100, I get an error running above SQL:

Microsoft.SqlServer.Types.GLArgumentException: 24205: The specified input does not represent a valid geography instance because it exceeds a single hemisphere. Each geography instance must fit inside a single hemisphere. A common reason for this error is that a polygon has the wrong ring orientation. To create a larger than hemisphere geography instance, upgrade the version of SQL Server and change the database compatibility level to at least 110.

[DanielStout5]

Interesting.. but I've already tried both ring orientations - assuming that's referring to the order the points are passed into IGeometryFactory.CreatePolygon. And the polygon definitely doesn't span hemispheres.

[JediLeeSkywalker]

I tried to draw a similar polygon in Wicket, then I get the result WKT put to SQL, then it works.
Get the WKT by API that before I mentioned, then use WKTReader.

POLYGON((-97.5701683636928 54.333628305112036,-29.3670433636928 54.333628305112036,-29.3670433636928 38.50235178337623,-97.5701683636928 38.50235178337623,-97.5701683636928 54.333628305112036))

[DanielStout5]

The only reason that works is because you have chosen a different polgyon that happens to include the point. That isn't a solution to this issue, because the polygon is dynamic depending on where the user has panned/zoomed the map to.

The problem is that the polygon doesn't have the correct shape for some reason when passed to SQL Server, even though it does when evaluated locally.

[ajcvickers]

Closing this as external since it seems there are no issues in the translation to SQL, just some confusion as to how SQL Server interprets these queries on a sphere.

[jasonhjohnson]

@DanielStout5 Daniel, did you ever get this figured out? I'm running into the same issue.

[DanielStout5]

Unfortunately not. I ended up just loading all of the data and doing the intersection calculation locally in C#.

I think the problem is to do with the coordinate system used by SQL Server as default. I'm not that familiar with GIS - if you end up finding a solution I'd be very eager to hear it!

[jasonhjohnson]

My issue didn't appear to be exactly the same as I wasn't getting the proper match when evaluating locally or via sql server. It's working for me now when I rearranged the coordinates as follows:

var points = new[]
            {
                new Coordinate(neLat, neLng), // TR
                new Coordinate(neLat, swLng), // BR
                new Coordinate(swLat, swLng), // BL
                new Coordinate(swLat, neLng), // TL
                new Coordinate(neLat, neLng), // TR
            };

[bricelam]

I suspect this is an orientation issue. With SQL Server geography, the shell of a polygon must be oriented counter-clockwise, and the holes must be oriented clockwise. If all the rings are oriented clockwise, it's interpreted as holes in FullGlobe. I think that's what the blue part of this image is showing:

[DanielStout5]

@bricelam That screenshot was just a demonstration. This is the one that was actually in use:

And in the code, the points are defined in counter-clockwise order:

var points = new[]
{
    new Coordinate(neLng.Value, neLat.Value), // Top Right
    new Coordinate(swLng.Value, neLat.Value), // Top Left
    new Coordinate(swLng.Value, swLat.Value), // Bottom Left
    new Coordinate(neLng.Value, swLat.Value), // Bottom Right
    new Coordinate(neLng.Value, neLat.Value), // Top Right
};

var fact = NtsGeometryServices.Instance.CreateGeometryFactory(srid: 4326);
var poly = fact.CreatePolygon(points);

var within = query.Where(x => x.Location != null)
    .ToList() // Force local evaluation
    .Where(x => x.Location.Within(poly))
    .ToList();

return Ok(PagedData<Shop>.FromList(within));

//query = query
//    .Where(x => x.Location.Within(poly));