[
https://issues.apache.org/jira/browse/ARROW-8065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17056322#comment-17056322
]
Ben Kietzman commented on ARROW-8065:
-------------------------------------
If a {{Fragment}}'s schema is pulled from its dataset then we'll have to
document carefully that it *doesn't* correspond to the backing file's physical
schema (accessed through {{FileFormat::Inspect}} impls)
> [C++][Dataset] Untangle Dataset, Fragment and ScanOptions
> ---------------------------------------------------------
>
> Key: ARROW-8065
> URL: https://issues.apache.org/jira/browse/ARROW-8065
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++ - Dataset
> Reporter: Francois Saint-Jacques
> Priority: Major
>
> Currently: a fragment is a product of a scan; it is a lazy collection of scan
> tasks corresponding to a data source which is logically singular (like a
> single file, a single row group, ...). It would be more useful if instead a
> fragment were the direct object of a scan; one scans a fragment (or a
> collection of fragments):
> # Remove {{ScanOptions}} from Fragment's properties and move it into
> {{Fragment::Scan}} parameters.
> # Remove {{ScanOptions}} from {{Dataset::GetFragments}}. We can provide an
> overload to support predicate pushdown in FileSystemDataset and UnionDataset
> {{Dataset::GetFragments(std::shared_ptr<Expression> predicate)}}.
> # {{Fragment::schema}} property should be set at construction time, usually
> extracted from the fragment's Dataset.
> This will lessen the cognitive dissonance between fragments and files since
> fragments will no longer include references to scan properties.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
