The most obvious potential source of incompatibility when compiling with a newer JDK, targeting an older one is overlooking code referencing newer APIs. Such code will compile, but fail in runtime with ClassNotFound or similar exceptions. Luckily we are protected from that by Jenkins already.
Andrus On Feb 18, 2014, at 11:02 PM, Aristedes Maniatis <[email protected]> wrote: > Well, ideally we'd release out of a completely reproducible environment like > Jenkins. That's what I do at work and it is much more reliable. > > I'm not sure about exactly how stable the Apache Jenkins is, but it seems to > be better. Ideally we build and release from the 1.5 JDK to be absolutely > sure. I don't know if the newer javac introduced optimisations which are > backward compatible. > > Ari > > > On 18/02/2014 5:48am, Mike Kienenberger wrote: >> No, that's fine. It shouldn't block the release. >> >> I'd prefer we used the same jdk to build the release for all systems, >> but perhaps that's not practical. >> >> >> >> On Tue, Feb 18, 2014 at 9:57 AM, Andrus Adamchik <[email protected]> >> wrote: >>> So Mike, unless you'd like to follow up on that cross-JDK binary build, I >>> am going to post a release announcement on user@. To me the important piece >>> that guarantees valid binary assemblies regardless of JDK is this: >>> >>> <plugin> >>> <artifactId>maven-compiler-plugin</artifactId> >>> <version>2.3.2</version> >>> <configuration> >>> <source>1.5</source> >>> <target>1.5</target> >>> </configuration> >>> </plugin> >>> >>> >>> Andrus >>> >>> On Feb 16, 2014, at 10:37 PM, Andrus Adamchik <[email protected]> >>> wrote: >>> >>>> Sorry, I should've waited for your vote and thanks a lot for doing a >>>> thorough review. >>>> >>>>> checksums match: check (Did we change our md5 formats? The current >>>> >>>> Yeah, lately we've been using gpg for that instead of md5 command: >>>> >>>> gpg --print-md MD5 cayenne-X.X.tar.gz >>>> >>>>> Turns out we built the zip versions with >>>>> java 1.6 and the tar.gz versions with java 1.7! not sure exactly how >>>>> that happened, but I wouldn't think we should be releasing like this! >>>>> Users will potentially have different results depending on whether >>>>> they grabbed the zip or the tar.gz, and I know that I'm not always >>>>> particular about which format I use. Will the 1.7 jar files work on >>>>> a 1.6 JRE? >>>> >>>> >>>> Of course. The same set of sources is used on Mac to build .dmg and >>>> .tar.gz and then on Windows to build .zip. My two envs happened to have >>>> different JDKs. So that's causing these small difference. I'd say there >>>> are no essential differences to worry about (although I'll try to keep my >>>> JDKs in sync across platforms in the future). >>>> >>>> In fact we make a claim that Cayenne 3.1 is compatible with Java 1.5. So >>>> if there was no backwards compatibility, we would've been forced to use >>>> JDK 1.5. If we actually see a problem, we should definitely pull the >>>> binary and redo it, but I don't think we will. >>>> >>>> Andrus >>>> >>>> >>>> On Feb 16, 2014, at 9:02 PM, Mike Kienenberger <[email protected]> wrote: >>>>> I didn't realized the vote was closed, and finally finished my review >>>>> today: >>>>> >>>>> Source provided: check >>>>> checksums match: check (Did we change our md5 formats? The current >>>>> format doesn't feed back into md5sum) >>>>> signatures match: check >>>>> Source builds: check >>>>> appropriately licensed: checked by rat >>>>> >>>>> My src jar builds match the tar.gz versions (except for timestamps), >>>>> but not the zip versions. Turns out we built the zip versions with >>>>> java 1.6 and the tar.gz versions with java 1.7! not sure exactly how >>>>> that happened, but I wouldn't think we should be releasing like this! >>>>> Users will potentially have different results depending on whether >>>>> they grabbed the zip or the tar.gz, and I know that I'm not always >>>>> particular about which format I use. Will the 1.7 jar files work on >>>>> a 1.6 JRE? >>>>> >>>>> >>>>> >>>>> >>>>> >>>>> On Sat, Feb 15, 2014 at 11:17 PM, Andrus Adamchik >>>>> <[email protected]> wrote: >>>>>> I am adding my +1. And I am closing the vote. Here is the list of votes: >>>>>> >>>>>> John Huss +1 >>>>>> Aristedes Maniatis +1 >>>>>> Michael Gentry +1 >>>>>> Andrus Adamchik +1 >>>>>> >>>>>> We have 4 +1s and no other votes, so the release becomes official. I >>>>>> will post the files now and update the downloads page. >>>>>> >>>>>> Thanks everyone, and let's get ready for 3.2 vote soon :) >>>>>> >>>>>> Andrus >>>>>> >>>>>> >>>>> >>>> >>>> >>> > > -- > --------------------------> > Aristedes Maniatis > GPG fingerprint CBFB 84B4 738D 4E87 5E5C 5EFA EF6A 7D2E 3E49 102A >
