[
https://issues.apache.org/jira/browse/IO-689?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17200727#comment-17200727
]
Uwe Schindler commented on IO-689:
----------------------------------
The general problem here: An instant is just Epoch seconds (look at source
code, there are no timezones involved!), so why do you convert it to a zoned
time at all (and back)? There is no reason to do this, you don't even have a
test that shows that this is really needed. This is overhead, because it needs
to access a synchronized global JVM setting (default zone), it also produces
many useless objects and does calculations that don't change anything - which
you verified in your test.
The roundtrip works in your example, because "atZone" includes the timezone
into the returned object. I was missing that, sorry! If you would convert it to
a LocalDateTime, then you have a problem - ZonedDatetime is OK!
With your Test you have already shown, that the Instant does not change, so
please remove the useless conversion! That's all I want to say, the risk to
change the epoch seconds is too high for nonsense. The method that's called at
the very end is just {{isFileNewer(File,long)}} and that takes a epoch milli
(that has no timezone). {{Instant}} is just a wrapper around that.
> All new FileUtils method with Java 8 time break on DST change
> -------------------------------------------------------------
>
> Key: IO-689
> URL: https://issues.apache.org/jira/browse/IO-689
> Project: Commons IO
> Issue Type: Bug
> Components: Utilities
> Affects Versions: 2.8.0
> Reporter: Uwe Schindler
> Priority: Major
>
> This commit in PR #124 breaks FileUtils with new Java 8 datetime APIs:
> https://github.com/apache/commons-io/pull/124/commits/2eb549873470844c88681e92c64631f796002020
> Because of this I had to put all of the methods to the list of blacklisted
> APIs in Apache Lucene / Solr. The reason for this change is that now all
> depend on local datetime, there's no way to pass an Instant with a UNIX
> timestamp through the API without it being corrupted due to
> forwards/backwards transformation.
> Background: During DST changes the same local date time exists two times (in
> autumn, you have in most countries two time the 2:30am time, once before and
> once after the DST change - because time is rolled back).
> With the above commit you first convert an Instant (which is by definition
> just a UNIX timestamp and can be converted as-is to a long) to a local
> Datetime and then back to an Instant. By this forward/backward transformation
> you get off by an hour during that one hour in autumn, when DST switches back
> to standard time.
> Please revert this commit and release a bugfix.
> There is no reason to convert an Instant to local and back, the arguments in
> the PR are plain wrong. Instants are timezoneless and are identical to UNIX
> timestamps.
> This commit breaks all new methods, because at the end all new methods go
> through the "Instant" path which does the faulty transformation.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
