[
https://issues.apache.org/jira/browse/ARROW-2051?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16932645#comment-16932645
]
Antoine Pitrou commented on ARROW-2051:
---------------------------------------
1) We don't have 128-bit numbers in Arrow. We do have fixed-size binary data.
2) Arrow now has extension types, so you could probably implement a UUID
extension type (though the Python API for extension types may still be in flux).
3) The answer to "why not..." questions generally is that it costs maintenance
time for us, so unless some contributor (such as you) wants to bear the
maintenance cost it probably won't happen if we don't find it useful enough.
> [Python] Support serializing UUID objects to tables
> ---------------------------------------------------
>
> Key: ARROW-2051
> URL: https://issues.apache.org/jira/browse/ARROW-2051
> Project: Apache Arrow
> Issue Type: Bug
> Components: Python
> Affects Versions: 0.8.0
> Reporter: Omer Katz
> Priority: Major
>
> UUID objects can be easily supported and can be represented as 128-bit
> integers or a stream of bytes.
> The fastest way I know to construct a UUID object is by using it's 128-bit
> (16 bytes) integer representation.
>
> {code:java}
> %timeit uuid.UUID(int=24197857161011715162171839636988778104)
> 611 ns ± 6.27 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
> %timeit uuid.UUID(bytes=b'\x124Vx\x124Vx\x124Vx\x124Vx')
> 1.17 µs ± 7.5 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
> %timeit uuid.UUID('12345678-1234-5678-1234-567812345678')
> 1.47 µs ± 6.08 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
> {code}
>
> Right now I have to do this manually which is pretty tedious.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
