[geotk] datum shift missing

[Martin Desruisseaux]

Hello Dieter

Very good catch, I own you my congratulation! (as it was deep in the
code). This bug has been introduced when we migrated from javax.vecmath
matrices to the Apache SIS matrices. Vecmath performed the
multiplication in-place (overwriting the source matrix) while SIS
returns the result in a new matrix. In the migration to SIS, we forgot
to assign the result in this particular place.

I just fixed the bug and pushed the fix to GitHub. Would you have a
chance to test again?

On the code accuracy, the testing effort has shifted to Apache SIS. Some
Geotk tests have been ignored because I was planning to review them in
depth when those parts will migrate to SIS (SIS has a more reproductive
testing environment with better diagnostics in case of failure).

My apologize for the inconvenience...

    Martin



Le 05/09/14 14:13, Dieter Stüken a écrit :

> Hi,
>  
> I participate on developing some frequency-analysis software tool
> (Java) for about ten years now:
> _http://www.irt.de/en/activities/programme-distribution/frequency-planning-software.html_
>  
> We have a lot of legacy code on board, but we also try to introduce
> advanced technologies.
> We switched to Java8 already and started using JavaFX  instead of
> Swing, since it is (still) a desktop application....
>  
> I observe the development of Geotk and SIS since a year, and the
> transition to SIS has many ideas in common with our current approach.
> So my first try was to use the referencing module to replace our
> internal coordinate transformations.
> Our current implementation depends on a very old geotoolkit version I
> want to get rid of.
>  
> Since we use only a small set of CRs (about 5-6) those are hardcoded
> and I don't like to involve the full EPSG database.
> Unfortunately I did not get the datum shift working.
>  
> I tracked it down to where the transformation matrices are build and
> ended up here:
>  
> _https://github.com/Geomatys/geotoolkit/blob/master/modules/referencing/geotk-referencing/src/main/java/org/geotoolkit/referencing/operation/DefaultCoordinateOperationFactory.java#L1080_
>  
> which performs the concatenation but finally does not use the result
> matrix returned L
>  
> I even found a unit test for this:
>  
> _https://github.com/Geomatys/geotoolkit/blob/master/modules/referencing/geotk-referencing/src/test/java/org/geotoolkit/referencing/operation/COFactoryUsingMolodenskyTest.java#L451_
>  
> but unfortunately it is @Ignored.
>  
> Do I miss something?
>  
> Dieter.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/geotoolkit/attachments/20140905/85e2a653/attachment.html>