RFR JDK-8013386: (tz) Support tzdata2013c (original) (raw)

Masayoshi Okutsu masayoshi.okutsu at oracle.com
Fri May 10 09:57:44 UTC 2013

On 5/10/2013 3:30 PM, Xueming Shen wrote:

On 5/9/13 9:57 PM, Masayoshi Okutsu wrote:

Do we still need to keep the old javazic code in JDK8? It's a maintenance burden to maintain both, isn't it?

While it's a burden, but somehow it serves as a test case pretty well. The transitions are being built the "old" jdk way and threeten way, if the transition does not match, something might be wrong. This time, it's in the "old" code, maybe next time it's in the threeten. So it might be still worth keeping a while, to remove in jdk9? Btw, the Rule.java fix might need go into the tzdb update tool as well, I believe the transitions for 2011 Palestine are wrong in the updated tzdb updator.

Yes, the Rule.java fix needs to go to older JDKs and TZ updater (build system).

Masayoshi

I'm concerned about the 24:00 fix. Is there any way to produce the correct rules without hard coding time zone IDs? I don't know how to do it, yet. I definitely can have a RFE for it and spend some time on it later. -Sherman

Masayoshi On 5/10/2013 8:24 AM, Xueming Shen wrote: Hi Sean,

Thanks for the review. It appears I missed jdk/test/sun/util/calendar/zi/tzdata, webrev has been updated to include the test data update. http://cr.openjdk.java.net/~sherman/8013386/webrev I will update TCKZoneRulesProvider.java separately in JSR310 repo, this def is actually is not being used by anyone anymore, just need to be removed. jdk/makefiles/GendataTimeZone.gmk is no longer used, need to be removed, I will remove it separately later.

Masayoshi, The update also included the two changes needed to fix/workaround the following 2 issues found during running the regression test jdk/test/sun/util/calendar/zi/TestZoneInfo310.java, due to the changes for Rule Palestine and the corresponding Zone Asia/Gaza and Asia/Hebron [1]. (1) Now the Rule Palestine has the def of "lastThu 24:00", similar to Asia/Amman, so these two zones need to be handled specially in ZoneInfoFile as well [2] (2) Rule Palestine has 2 day-saving changes in 2011, so it has 4 transitions for 2011 when returned from Rule.getRules(int year). Unfortunately it appears the Comparator for Arrays.sorting is incorrectly implemented when comparing two longs [3]. The directly consequence of this decade-old bug is that the returned list has the wrong order for 2011/08/01/xxx and 2011/08/30/xxx Please help review. Thanks! -Sherman [1] http://cr.openjdk.java.net/~sherman/8013386/webrev/make/sun/javazic/tzdata/asia.sdiff.html [2] http://cr.openjdk.java.net/~sherman/8013386/webrev/src/share/classes/sun/util/calendar/ZoneInfoFile.java.sdiff.html [3] http://cr.openjdk.java.net/~sherman/8013386/webrev/test/sun/util/calendar/zi/Rule.java.sdiff.html On 05/09/2013 02:06 AM, Seán Coffey wrote: Thanks for taking care of this Sherman. I was wondering what sort of impact JSR 310 would make on tzdata updates. The Atlantic/Stanley display name issue mentioned is a regular one, we should log a bug against the test file generation scripts. I just had a quick grok of the jdk8 repo. The following files need updating also : jdk/test/sun/util/calendar/zi/tzdata/* jdk/test/java/time/tck/java/time/zone/TCKZoneRulesProvider.java (line 85) jdk/makefiles/GendataTimeZone.gmk (line 29) It looks like jdk/makefiles/GendataTimeZone.gmk still has a dependency on reading files from jdk/make. That'll all have to change too once the old build system is removed from jdk8. I think the new tzdata sources should be moved into a directory under makefiles rather than keeping them in make. The "GENDATATIMEZONEVERSION := tzdata2012i" line in jdk/makefiles/GendataTimeZone.gmk should be removed if we know that the version string can be read from the VERSION file stored with tzdata. Above points are not necessarily related to 2013c update and should be cleaned up separately perhaps. regards, Sean. On 08/05/2013 23:20, Xueming Shen wrote: Hi,

Please help review the proposed change to update the tz data in JDK8 from 2012i to 2013c. Other than the tzdb data file update under make/sun/javazic/tzdata, corresponding updates have also been made in TimeZoneNames***.java for the newly added zones, Asia/Khandyga and Ust-Nera, and updated zone display names (from EET to CET) for Africa/Tripoli (and its alias Libya) test/java/util/TimeZone/tools/share/Make has been run to generate the new test data at TimeZoneData. # I have to update the displaynames.txt "manually" to undo the change for Atlantic/Stanley from "FKST" (which is defined in southamerica.txt both in 2012i and 2013c, there is no change for Stanley from 2012i to 2013c) back to "FKT FKST" to keep Bug6329116.java passed. I'm not sure why the definition in TimeZoneNames.java (which has FKT as the standard name and FKST as the summer name) does not match the tz data file (which suggests that Stanley has moved to use only summer zone), but since this appears to be an existing issue that not related to this update, I keep the test data for Stanley untouched. Tests list below have been run and passed. java/util/TimeZone java/util/Calendar java/util/Formatter java/time closed/java/util/TimeZone closed/java/util/Calendar webrev: http://cr.openjdk.java.net/~sherman/8013386/webrev/ http://cr.openjdk.java.net/~sherman/8013386/test.closed/ Thanks! Sherman

More information about the core-libs-dev mailing list