[
https://issues.apache.org/jira/browse/PARQUET-712?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15475169#comment-15475169
]
Deepak Majeti commented on PARQUET-712:
---------------------------------------
As long as the current parquet-cpp reader / writer API remains unchanged, this
should be okay.
I understand that parquet-cpp will benefit from Arrow's richer I/O layers, file
system layers, memory allocators, utility classes (encoding / compression).
Reversing the dependency makes sense for this benefit.
It is not clear to me why generating Arrow records in this project will benefit
in any way.
It could make things complicated since parquet-cpp uses {{exceptions}} and
Arrow uses {{Status}}.
> C++: Read into Arrow memory
> ---------------------------
>
> Key: PARQUET-712
> URL: https://issues.apache.org/jira/browse/PARQUET-712
> Project: Parquet
> Issue Type: New Feature
> Components: parquet-cpp
> Reporter: Uwe L. Korn
>
> We want to reverse the dependency between arrow-cpp and parquet-cpp. Thus
> parquet-cpp should depend on arrow-cpp and read/write to/from arrow memory.
> There are two options the implementation could go ahead:
> * Optionally depend on arrow-cpp and have some methods that read to arrow
> memory
> * Strictly depend on arrow-cpp: This would avoid some code duplication
> between the two projects, i.e. we could use Arrow's allocator, I/O layer,
> utility classes instead of having our own ones in parquet-cpp which do
> exactly the same, just live in another namespace.
> (I'll create some more JIRAs so we can proceed a bit more step-by-step but
> this one should be the one for coordination and discussions)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
