[
https://issues.apache.org/jira/browse/ARROW-8749?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17103475#comment-17103475
]
Wes McKinney commented on ARROW-8749:
-------------------------------------
Definitely not good. This may be an artifact of the fact that the way that we
attribute dictionary id's to fields is not very robust. Is this a regression (I
marked for 0.17.1 in case it is)?
> [C++] IpcFormatWriter writes dictionary batches with wrong ID
> -------------------------------------------------------------
>
> Key: ARROW-8749
> URL: https://issues.apache.org/jira/browse/ARROW-8749
> Project: Apache Arrow
> Issue Type: Bug
> Components: C++
> Affects Versions: 0.17.0
> Reporter: David Li
> Priority: Major
> Fix For: 1.0.0, 0.17.1
>
>
> IpcFormatWriter assigns dictionary IDs once when it writes the schema
> message. Then, when it writes dictionary batches, it assigns dictionary IDs
> again because it re-collects dictionaries from the given batch. So for
> example, if you have 5 dictionaries, the first dictionary will end up with ID
> 0 but be written with ID 5.
> For example, this will fail with "'_error_or_value11.status()' failed with
> Key error: No record of dictionary type with id 9"
> {code:cpp}
> TEST_F(TestMetadata, DoPutDictionaries) {
> ASSERT_OK_AND_ASSIGN(auto sink, arrow::io::BufferOutputStream::Create());
> std::shared_ptr<Schema> schema = ExampleDictSchema();
> BatchVector expected_batches;
> ASSERT_OK(ExampleDictBatches(&expected_batches));
> ASSERT_OK_AND_ASSIGN(auto writer, arrow::ipc::NewStreamWriter(sink.get(),
> schema));
> for (auto& batch : expected_batches) {
> ASSERT_OK(writer->WriteRecordBatch(*batch));
> }
> ASSERT_OK_AND_ASSIGN(auto buf, sink->Finish());
> arrow::io::BufferReader source(buf);
> ASSERT_OK_AND_ASSIGN(auto reader,
> arrow::ipc::RecordBatchStreamReader::Open(&source));
> AssertSchemaEqual(schema, reader->schema());
> for (auto& batch : expected_batches) {
> ASSERT_OK_AND_ASSIGN(auto actual, reader->Next());
> AssertBatchesEqual(*actual, *batch);
> }
> }{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
