[
https://issues.apache.org/jira/browse/ARROW-5889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16882416#comment-16882416
]
TP Boudreau commented on ARROW-5889:
------------------------------------
That's correct. An Arrow timestamp field either has a timezone or doesn't so
we can always directly report that the isAdjustedToUTC is true or false
(respectively). So that field can be converted to a Parquet node with a user
supplied TimestampLogicalType (with fromConvertedType := false, as you surmise)
which can then be memorialized in the Parquet file, without the
fromConvertedType flag of course, but correctly and unambiguously.
It's only when constructing the logical type from a Parquet file that has only
converted types where fromConvertedType would be true. Then passing the
Parquet node thru to an Arrow field we could create a timezone-naive timestamp
field (restoring the behavior requested in this issue). After that we have a
well-formed Arrow field, so we can goto the first paragraph of this comment.
[The reason for declining to write a TimestampLogicalType that came from a
converted type is so that a non-Arrow parquet user who does a read followed by
write doesn't change the semantics of the TIMESTAMP converted type by coercing
it to a TimestampType with UTCAdjusted := true (as the parquet.thrift spec
currently demands). For that user, whatever TIMESTAMP meant before, it still
means and it's still the controlling annotation for the column.]
Hope that helps. (It helped me, so thanks!)
Unless I hear a hard "No" on this, I'm going to start hacking on this first
option (mostly to gain confidence that it'll work), but also to see how
non-intrusive a change I can make it.
> [Python][C++] Parquet backwards compat for timestamps without timezone broken
> -----------------------------------------------------------------------------
>
> Key: ARROW-5889
> URL: https://issues.apache.org/jira/browse/ARROW-5889
> Project: Apache Arrow
> Issue Type: Bug
> Components: C++, Python
> Affects Versions: 0.14.0
> Reporter: Florian Jetter
> Priority: Minor
> Labels: parquet
> Fix For: 0.14.1
>
> Attachments: 0.12.1.parquet, 0.13.0.parquet
>
>
> When reading a parquet file which has timestamp fields they are read as a
> timestamp with timezone UTC if the parquet file was written by pyarrow 0.13.0
> and/or 0.12.1.
> Expected behavior would be that they are loaded as timestamps without any
> timezone information.
> The attached files contain one row for all basic types and a few nested
> types, the timestamp fields are called datetime64 and datetime64_tz
> see also
> [https://github.com/JDASoftwareGroup/kartothek/tree/master/reference-data/arrow-compat]
> [https://github.com/JDASoftwareGroup/kartothek/blob/c47e52116e2dc726a74d7d6b97922a0252722ed0/tests/serialization/test_arrow_compat.py#L31]
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
