Closed GoogleCodeExporter closed 8 years ago
no objections, actually
Original comment by diegogu...@gmail.com
on 2 Nov 2011 at 7:46
+1 to extract interfaces for readers/writers, not sure if they should be in
GeoAPI.
--- some other thoughts, not sure yet:
Do we need to distinguish between type of the reader (text / binary) on
interface level or should it be IGeometryReader, IGeometryWriter?
Should they be readers / writers or something like TypeConverter?
... or even:
interface IGeometry
{
void FromWellKnownText(string text)
void FromWellKnownBinary(byte[] byteArray)
string ToWellKnownText();
byte[] ToWellKnownBinary();
}
Original comment by Gennadiy.Donchyts
on 2 Nov 2011 at 1:37
Using the reader/writer approach we can parse not just only WKB/WKT
representations of geometries, but also setup reader and writer for e.g.
- SqlServerGeometry/-Geography,
- PostGis geometry,
- SpatiaLite geometry blob (GaiaGeo),
- Oracle SDO
- GeoJSON
- ...
That way we can skip the conversion to and from WKB.
Please have a look at NetTopologySuite.IO solution folder, which contains
reader/writer for the above.
The interfaces are currently set up like this:
#region reader interfaces
interface IGeometryReader<TSource, TStream>
{
IGeometryFactory { get; set; }
IGeometry Read(TSource source);
IGeometry Read(TStream stream);
}
interface ITextGeometryReader : IGeometryReader<string, System.IO.TextReader> {}
interface IBinaryGeometryReader : IGeometryReader<byte[], System.IO.Stream> {}
#endregion
The interfaces for the writers are analogous.
On the IGeometry interface there are already
byte[] AsBinary() and
string AsText()
defined.
I think extending IGeometryFactory interface with
IGeometry FromBinary(IBinaryReader reader, byte[] bytes) and
IGeometry FromText(ITextGeometryReader reader string text)
would make more sense than adding to the IGeometry interface, since we cannot
change the concrete geometry object (e.g. make Polygon a Point) and we cannot
implement interfaces with static functions.
Original comment by felix.ob...@netcologne.de
on 2 Nov 2011 at 2:25
... right, Parse() in interface does not make much sense, in .NET it is part of
concrete class like in .NET types.
... convertion to WKB/WKT is in IGeometry and from WKT/WKB somewhere else, a
bit confusing.
Sorry, I've missed IGeometryReader<TSource, TStream>, I use old version
Is there an example somewhere demonstrating how they are supposed to be used? I
find generic arguments <TSource, TStream> too implicit. It can be anything.
Also if we already have stream, why do we need to specify some TSource?
Implementation already knows how it is stored.
... should example be *always* written demonstrating how API is supposed to be
used before changes? TDD philosophy.
// for example, assume that we want to create geometries via factory:
void CreateGeometriesFromWellKnownFormats()
{
var geometryFactory = ... how do we normally get factory without referencing NTS?
// from WKT
var point1 = geometryFactory.CreateGeometry<IPoint>("POINT(1.0 1.0)");
// from WKB, BE
var point2 = geometryFactory.CreateGeometry<IPoint>(byteArray, true);
}
interface IGeometryFactory
{
...
public TGeometry CreateGeometry<TGeometry>(string wkt) : where TGeometry is IGeometry
public TGeometry CreateGeometry<TGeometry>(byte[] wkb, bool bigEndian) : where TGeometry is IGeometry
...
}
Original comment by Gennadiy.Donchyts
on 2 Nov 2011 at 9:30
... I studied recent code in NTS.IO a bit more, now it makes more sense :).
The only concern is about TSource, TStream. Will someone use these base
interfaces? Then the client code will look a bit messy:
if(reader is IGeometryReader<string, TextReader>) { ... }.
Original comment by Gennadiy.Donchyts
on 2 Nov 2011 at 10:14
so +1 for extracting interfaces.
We should check that working with interfaces is as simple as with concrete
implementations. If we have code in WktReaderTest:
var reader = new WktReader(...);
reader.Read("POINT(1.0 1.0)");
... then it should be as simple to do the same with interface:
ITextGeometryReader reader = ... get it somewhere
reader.Read("POINT(1.0 1.0)");
-----
the same is true for writers:
IGeometryWriter oracleWriter = ...
IGeometry geometry = ...
oracleWriter.Write(geometry, stream);
Original comment by Gennadiy.Donchyts
on 2 Nov 2011 at 10:23
I tend to get my geometry factory implementation using
http://commonservicelocator.codeplex.com
var factory =
Microsoft.Practices.ServiceLocator.GetInstance<IGeometryFactory>();
I will add a sample asap.
Regarding the generic base interface:
- TStream is there because the WKT reader/writer implementation take TextReader
and TextWriter while WKB reader/writer simply uses System.IO.Stream on both
interfaces.
- TSource/TOutput is there, to cover the cases where the result or input may be
a
concrete object (SqlGeometry, SqlGeography, some OpenMI.SpatialObject?) rather
than an array of bytes.
So MSSqlGeometryReader it would implement IGeometryReader<SqlGeometry, Stream>.
But if there are objections: I'm not married with those generic base interfaces,
we can revert that, and leave that functionality to the concrete implementation.
Original comment by felix.ob...@netcologne.de
on 3 Nov 2011 at 9:03
No, these interfaces look nice for implementation. But we probably never will
know about concrete type such as SqlGeometry on GeoAPI level. So I'm thinking
if we will need something more API-friendly in GeoAPI where no additional
knowledge of a very specific type will be required? On the other hand it sounds
useful to have from/to WKT/WKB conversion on API level since they are standards.
Maybe we should leave it as you think will be best for now and discuss again
when we can meet in person :).
If it will help to re-use code in NTS / SharpMap - (+1) to leave them in GeoAPI
and discuss later how they can be improved.
-------
... some more thoughts:
It looks like 2 tasks are beins solved here:
1) Convert IGeometry to/from another type (WKT string, GeoJSON string, WKB byte
array, specific type like SqlGeometry).
2) Actual IO, read / write goeometries using some data source (access file,
access database, web service, etc.).
... just wondering if both of these requirements fit into reader / writer
concept or is something more light-weight can be used in the first case. So
that we can have an API concept to convert geometries to / from different
representations and to read / write one or multiple geometries from different
data sources, maybe specified using some Path (file path, connection string).
The second feels more related to something like FeatureProviders where also
attributes will be read / written ...
Original comment by Gennadiy.Donchyts
on 3 Nov 2011 at 11:28
+1 to parameterize on input and output types. However, I am opposed
parameterizing based on different input types (as in TSource/TStream): I think
we should settle on streams.
I'm also positive to the addition of FromText/FromBinary and changes to
AsText/AsBinary as described above.
I'm glad to see this change as it enables a more pluggable architecture: we use
custom-made WKT-reader and -writer.
Original comment by johan.li...@astando.se
on 3 Nov 2011 at 1:07
> However, I am opposed parameterizing based on different input types (as in
> TSource/TStream): I think we should settle on streams.
You think the text interfaces should be
interface ITextGeometryReader : IGeometryWriter<string, System.IO.Stream> {};
interface ITextGeometryWriter : IGeometryWriter<string, System.IO.Stream> {};
So we can discard the System.IO.Stream part?
Original comment by felix.ob...@netcologne.de
on 3 Nov 2011 at 2:00
Personally I don't mind the <From,To> generic interfaces - though I think
Stream and string are too broad as input or output types - how do you determine
whether a string is GeoJson or Wkt without trying/ failing to parse it. same
with stream and ~everything.
Personally I prefer the concept of conversion rather than reading /writing.
Utilising the interfaces you can easily create a registry of converters which
you can add to at runtime - you can also do it without the interfaces if you
parameterize the registration of the converters.
E.g
interface IConverter<TFrom, TTo>
{
TTo Convert(TFrom obj);
}
class GeoJsonConverter : IConverter<GeoJsonStream, IGeometry> ,
IConverter<IGeometry, GeoJsonStream>
{
...
}
...
interface IConverterRegistry
{
void Register<TFrom, TTo>(IConverter<TFrom, TTo> converter);
bool CanConvert<TFrom, TTo>();
TTo Convert<TFrom, TTo>(TFrom source);
...
}
IConverterRegistry converters = GetConverters();
converters.Register<GeoJsonStream, IGeometry>(new GeoJsonConverter());
converters.Register<IGeometry, GeoJsonStream >(new GeoJsonConverter());
converters.Register<WkbStream, IGeometry>(new WkbConverter());
converters.Register<IGeometry,WkbStream>(new WkbConverter());
converters.Register<WktStream, IGeometry>(new WktConverter());
converters.Register<IGeometry, WktStream>(new WktConverter());
converters.Register<SqlGeometry, WktStream>(...
.......
IGeometry g = GetGeometry();
IConverterRegistry converters = GetConverters();
if(converters.CanConvert<IGeometry, WktStream>())
{
WktStream s = converters.Convert<IGeometry, WktStream>(g);
}
where FooStream is either an type containing a Stream property or a subclass
of stream. If stream was used directly the registry would break down unless
keyed with extra info.
hope that makes sense... got to go.. jd
Original comment by john.d...@newgrove.com
on 3 Nov 2011 at 4:03
I think it's best to keep it simple. Dealing with registrys of converters seems
well out of scope for the problem at hand. I would think the most common use
case is someone wanting to convert some GeoJSON or some WKB.
For those use cases it would suffice with:
IGeometryReader<TSource, TCoordinate> where TCoordinate : ... {
IGeometry<TCoordinate> Read(TSource);
}
And the writer:
IGeometryWriter<TSink, TCoordinate> where TCoordinate : ... {
void Write(IGeometry<TCoordinate> , TSink );
}
That puts no restriction on source or sink types.
On top of that it's possible to engineer some register with converters, or
whatever you want. If you want to put another level of abstraction wrt textual
vs binary formats, i.e. ITextGeometryReader, that's fine (although I think that
adds very little functionality).
Point being: keep it simple, please.
Original comment by johan.li...@astando.se
on 4 Nov 2011 at 9:38
>keep it simple, please.
I strongly agree with this
Original comment by diegogu...@gmail.com
on 4 Nov 2011 at 9:55
@gennadiy:
As a first step I'm only interested in reading and writing
GeoAPI.Geometry.IGeometry instances from and to other spatial objects. The
resulting reader/writers should be re-usable e.g. for SharpMap/(DotSpatial if
it adheres to GeoAPI) data providers or NHibernate.Spatial. So I'm focused on
just the geometry part.
@john:
Can't a converter registry can be setup using reader/writer?
@all:
I'm thinking about a way to control the ordinate output. IMHO we have two
options, either use
bool HandleOrdinate(Ordinate ordinate); // analogous to EmitZ, EmitM
or a flag like approach
Ordinates Ordinates { get; set; }
any suggestions/preferation?
Original comment by felix.ob...@netcologne.de
on 4 Nov 2011 at 10:35
@felix you can create a registry with reader/writer, it is just the semantics.
If we are allowing conversion between types in addition to serialization and
deserialization the I prefer the "conversion" concept as reading/writing
implies serialization. The "Converter" may internally utilize a FooReader/
FooWriter in order to do its work when the source or target is a file.
Thinking aloud how about:
IDictionary<Ordinate, double> GetOrdinates(Ordinate ordinates)
Cheers jd
Original comment by john.d...@newgrove.com
on 4 Nov 2011 at 11:20
> Thinking aloud how about:
> IDictionary<Ordinate, double> GetOrdinates(Ordinate ordinates)
I thought of it more as a ForceCoordinate option (like in the v2 ShapeFile
provider).
Original comment by felix.ob...@netcologne.de
on 4 Nov 2011 at 11:33
Agree with @jd on Conversion versus Writer/Reader. Actually they are also
different in terms that a single converter is responsible for two-way
conversion:
public interface IGeometryConverter
{
///
/// actual type used during conversion, string, byte[] ...
///
Type ConversionType { get; }
IGeometry ToGeometry<T>(TSource source);
T FromGeometry<T>(IGeometry geometry);
}
public class GeoJsonGeometryConverter : IGeometryConverter
{
IGeometry ToGeometry<T>(T source) { ... call ToGeometry(source as string) ... }
T FromGeometry<T>(IGeometry geometry) { ... call FromGeometry(geometry) ... }
// actual specific type implementations:
IGeometry ToGeometry(string str) { ... }
string FromGeometry(IGeometry geometry) { ... }
}
Original comment by Gennadiy.Donchyts
on 4 Nov 2011 at 12:02
Or maybe ConvesionType can be ommited and insted there should be:
public interface IGeometryConverter
{
bool CanConvertGeometryFrom<T>(T source);
bool CanConvertGeometryTo<T>(T target);
IGeometry ConvertGeometryFrom<T>(T source);
T ConvertGeometryTo<T>(IGeometry geometry);
}
Original comment by Gennadiy.Donchyts
on 4 Nov 2011 at 12:06
Ah right, I like Flags :)
Original comment by john.d...@newgrove.com
on 4 Nov 2011 at 12:08
@gennadiy, john:
Can we agree on the IGeometryConvert thing being a different task? I hesitate
to move away to far from the JTS source, since it may get harder to maintain or
update, and the JTS.IO namespace works with readers/writers.
Original comment by felix.ob...@netcologne.de
on 4 Nov 2011 at 12:23
Fair enough, I just thought that ToBinary, ToText, FromBinary, FromText could
delegate to a standardized Convert api. But in the interests of harmony that is
fine :)
Original comment by john.d...@newgrove.com
on 4 Nov 2011 at 12:29
@felix yes, sure, I feel that due to my comments this thread becomes an
off-topic :).
Original comment by Gennadiy.Donchyts
on 4 Nov 2011 at 1:50
done with NTS 1.12
Original comment by felix.ob...@netcologne.de
on 6 Mar 2012 at 2:49
Original issue reported on code.google.com by
felix.ob...@netcologne.de
on 1 Nov 2011 at 5:00