| From | Sent On | Attachments |
|---|---|---|
| Palle Girgensohn | Feb 13, 2005 4:45 pm | |
| Panagiotis Astithas | Feb 14, 2005 1:15 am | |
| Palle Girgensohn | Feb 14, 2005 1:19 am | |
| Panagiotis Astithas | Feb 14, 2005 1:36 am |
| Subject: | postgresql-jdbc packaging | |
|---|---|---|
| From: | Panagiotis Astithas (pa...@ebs.gr) | |
| Date: | Feb 14, 2005 1:36:49 am | |
| List: | org.freebsd.freebsd-java | |
Palle Girgensohn wrote:
--On m?ndag, februari 14, 2005 11.15.36 +0200 Panagiotis Astithas <pa...@ebs.gr> wrote:
Palle Girgensohn wrote:
Hi!
I'm maintaining the postgresql-jdbc port.
One thing I've considered, but not come to any conclusion about, is whether the port should register somehow which version of JDBC it has built, JDBC1, JDBC2 or JDBC3. There's even a JDBC2 + EE variant... Which version is built depends on which JDK was used to build it. jdk1.1 => JDBC1, jdk1.2-1.3 => JDBC2, and jdk1.4+ => JDBC3. Hence, very few would want JDBC1 nowadays, I suppose. The only package built by the package cluster now is for JDBC1, which kind of sucks a bit :)
To fix this, the right way is to create a bunch of slave ports, on for each type as per above. Then, the package building cluster would build all version. The slave ports would set JAVA_VERSION=1.1 and 1.2 respectively, and the main port could install the greatest version. PKGNAMESUFFIX would be set to jdbcN.
Is this just overkill? If most of you use the port anyway, it probably is, but if ppl tend to use prebuilt packages, they will end up with a somewhat crippled JDBC1 jar even if they run jdk-1.5, so then it might be worth it.
I slimmer way is to just let the package name reflect which version has been built, but not bother to create slave ports.
Any opinions? What do you think, is it worth the effort?
/Palle
(See <http://jdbc.postgresql.org/download.html> for info on different versions of PostgreSQL's JDBC.)
As someone who was bitten by this, I believe package users should have some sort of warning sign. I don't mind what the solution will be, as long as a regular "pkg_add -r foo" can work as expected. Is this possible with the "slimmer" approach?
Cheers,
Panagiotis
With the slimmer approach, pkg_add will install postgresql-jdbc1, explicitally. With the fatter approach, there will be three packages to chose from, one each for jdbc{1,2,3}.
/Palle
So, in the former case, one would not be able to install a package other than -jdbc1, even if native binary jdk versions exist for 1.3, etc.? If I understand it correctly we are currently not building postgresql-jdbc2, even though we have a binary jdk 1.3, right? This sounds rather limiting. Can't the package cluster build at least postgresql-jdbc2, too? And when we get binary distributions for jdk 1.4/1.5, build packages for postgresql-jdbc3? If this requires the "fat" approach, then I'm all for it.
Cheers,
Panagiotis





