On 9/17/19 10:14 AM, Martin Buchholz wrote:
On Tue, Sep 17, 2019 at 9:45 AM <naoto.s...@oracle.com
<mailto:naoto.s...@oracle.com>> wrote:
+1
On 9/17/19 8:29 AM, Martin Buchholz wrote:
> Looks good to me.
> At Google we also integrated tzdata2019c, and it was uneventful
(good!).
> But we're still using rearguard format.
> The vanguard/rearguard distinction is a source of errors, so it
should be
> made clear what format is being used to import the files.
> If you have a script to update the jdk sources, perhaps it should be
> checked in to openjdk?
> If these files are in vanguard format, is there a trap for those
of us
> doing backports to jdks that only support rearguard?
Can't think of any off the top of my head, Martin. The build tool
Which build tool? TzdbZoneRulesCompiler?
Yes.
internally converts vanguard data into rearguard compatible, by
adjusting the standard offsets, and build into tzdb.dat (which should
even be compatible to prior JDK runtimes).
So ... a change like this one is created by copying selected vanguard
format files from the tzdata source distribution into openjdk, and
then TzdbZoneRulesCompiler converts to rearguard format internally? At
Google, we chose to run tzdata's own tool to convert to rearguard format
and then check in those data files into the jdk source tree.
The tool will not convert the file format to rearguard. Read the
vanguard format file, then converts the data into tzdb.dat that can be
recognized by prior runtimes.
I worry that when other engineers backport to older releases, if only
the data files are copied, then conversion to rearguard format will not
happen, with bad results.
If an engineer only backports the vanguard files into older releases
(before the vanguard support), I don't think the JDK build will succeed.
Build will fail on creating tzdb.dat file.
Naoto
