PGConnection vs. Connection

[Crosspost to psql-jdbc, postgis-devel and alex bodnaru]

Dear pgjdbc developers,

When backporting toe PostGIS jdbc extension to debian woody (which still
uses pgjdbc 7.2), Alex Bodnaru (PostGIS debian packager) stumbled over
the fact that the addDataType() methods are carried by the
org.postgresql.Connection class instead of the
org.postgresql.PGConnection interface in 7.3 and up.

An additional minor problem is that addDataType(String) is deprecated in
8.0 and this gives compiler warnings (and has the known class loader
related problems). This problem is minor because the old method works as
good as ever, and pgjdbc 8.0 provides the autoregistration via
driverconfig.properties so we do not strictly need addDataType() calls
at all in _most_ environments.

We're currently wondering what the best method is for supporting 7.2,
the following have come to our mind.

- Prodiding a patch that is to be applied manually (rsp. by the debian
package build process) when building for 7.2. (this is our "current
practice").

- Using some kind of pgjdbc autorecognition and preprocessing (e. G. ant
filters or C preprocessor). Because of the disadvantages regarding
common Java IDEs (Syntax highlighting, error checking tools etc.) and
the more complicated build process, I dislike the idea of introducing
a separate preprocessing step in a Java project as small as PostGIS
jdbc. But this is only my personal opinion, and you can convince me
with appropriate arguments.

- Manage two different source trees.

- Using reflection. This could allow us per run-time distinction to use
o.p.PGConnection.addDataType(Class) when running against pgjdbc 8.0,
o.p.PGConnection.addDataType(String) against 7.3/7.4, and
o.p.Connection.addDataType(String) against 7.2. This way we could ship
a single postgis.jar that runs against all pgjdbc versions.

- Using our own copy of PGConnection and Connection sources during our
build process. We could then use conn.getClass().getName() or such to
distinct which code to call, and encapsulate the actual code in
functor objects that get instantiated lazy, so we only load the
class that actually fits the real connection implementation (this
should avoid classloader trouble).

Which approach do you think to be the best one? Another question is
whether it is worth the effort to develop any complicated scheme, as
only a very few lines of code are actually affected.

Markus
--
markus schaber | dipl. informatiker
logi-track ag | rennweg 14-16 | ch 8001 zürich
phone +41-43-888 62 52 | fax +41-43-888 62 53
mailto:schabios(at)logi-track(dot)com | www.logi-track.com