I don't think the ZIP program itself is to blame. ZIP archives list their content in the order it was added, so I can easily imagine reasons why that might differ across different systems and Java versions. Probably the best thing to do is just sort the output (and the expected result) and that way we'll be comparing content, not order (which doesn't matter). I just tested it locally and it seemed to work, so I'll commit it and see what happens.

Hugh

On Fri, Aug 31, 2018 at 4:01 AM Peter Stadler <stadler@edirom.de> wrote:
Dear all,

as I reported briefly yesterday during our conf call, we currently have sort of a software incompatibility problem. Martin was setting up a new Jenkins on an Ubuntu 18.04 which produces slightly different results on some build tests. As a consequence, if we adjust the expected results (as Raff was trying to do), it will build on Martin’s Jenkins but not mine.
Adding to this confusion is the fact, that currently the Paderborn Jenkins is the primary, but the Victoria Jenkins is sending out emails (for I had that feature turned off when it was the secondary Jenkins).

Martin and I were trying to debug the issue which surfaces as differently ordered files in a ZIP archive (in the pub and odt tests). First, we thought it might be different versions of the unzip binary (leading to the problem), but when I was comparing the zip archives themselves I found out that it’s not the decompression but the compression that’s different. My next guess was about different Java versions (because the zip archive is produced via ANT which – I'm guessing – uses the standard java zip) and indeed Ubuntu 18.04 installs Java10 by default while Debian still uses Java8. Yet, forcing Ubuntu  to use Java8 didn’t change anything for me.

While we should further investigate the issue, for the short term I propose that we adjust the expected results according to the main Jenkins. Additionally, the secondary Jenkins should turn off the email notification to avoid confusion. Emails should be sent solely from the primary Jenkins.

If Martin’s Jenkins is back and running, we should ask Ian to switch the IP addresses again which would mean no other action is needed at present. Otherwise, we’d need to revert the expected results.

Any comments welcome!
Best
Peter
_______________________________________________
Tei-council mailing list
Tei-council@lists.tei-c.org
http://lists.lists.tei-c.org/mailman/listinfo/tei-council