Hello all
In all Apache SIS releases (including 1.2), the Mercator projection
intentionally does not wraparound the longitude values into the
-180°…+180° range. For example a longitude value of 181° is converted to
metres as-is, without prior conversion to -179°. It was intentional, for
keeping the function as continuous as possible (wraparounds are
discontinuities in the mathematical function).
However this behavior of the Mercator projection was different than the
behavior of almost all other projections. Most projections perform an
implicit wraparound, because the longitude value is typically used in
expressions like sin(longitude) or cos(longitude). In those cases, a
wraparound is implicitly done by the trigonometric function.
On May 12, an "artificial" wraparound has been added to the Mercator
projection for getting the same effect than other projections. This
changes was not included in SIS 1.2 release, in order to give us more
time to see the consequences. But now we are seeing cases where this
workaround is causing problems. I posted a comment in the following JIRA
issue with pictures trying to illustrate the problem:
https://issues.apache.org/jira/browse/SIS-547
In a nutshell, wraparound is a problem which can often not be solved at
the points level. It often needs to be solved at the geometry or raster
level, in which case applying wraparound on the transformed points hide
information that the geometry or the raster needs for solving the problem.
A counter-argument would be that wraparounds are implicit in almost all
other projections (because of trigonometry), and if the application of
wraparounds in Mercator breaks some code, then that code will be broken
with other projections as well. But a counter-counter argument is that
the Mercator projection is often used for world-wide views, which is not
the case of most other projections.
So I wonder if we should revert this commit (which was adding wraparound
in Mercator projection):
https://github.com/apache/sis/commit/3fa01381f14ddac330c2b5e6cc86bd144afd90d8
The consequence would be that if users really want wraparound in
Mercator projection, they would need to apply it themselves. This is
admittedly not convenient. An alternative would be to add a parameter in
the projection for specifying whether we want wraparound or not, but
that would be SIS-specific and difficult to specify when building a CRS
from EPSG code.
Any thoughts?
Martin
