ARROW-9304: [C++] Add "AppendEmpty" builder APIs for use inside StructBuilder::AppendNull

[tianchen92]

No description provided.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-9304

[emkornfield]

@pitrou does this seem ok now?

[pitrou]

As for naming, according to various dictionaries, "empty" as a noun (and "empties") is rather informal.
It seems AppendEmptyValue and AppendEmptyValues may be more appropriate (and more explicit), but I'm not a native English speaker.

[tianchen92]

As for naming, according to various dictionaries, "empty" as a noun (and "empties") is rather informal.
It seems AppendEmptyValue and AppendEmptyValues may be more appropriate (and more explicit), but I'm not a native English speaker.

Fine, fixed.

[tianchen92]

Hi, @emkornfield , please also take a look at this patch when you have time.
This patch add AppendEmptyValue/AppendEmptyValues used in StructBuilder#AppendNull/AppendNulls, which would set child null bitmap slot to true rather than false( http://arrow.apache.org/docs/format/Columnar.html#struct-layout).
And now this test ArrowReaderWriter#SingleColumnNullableStruct failed, could you please help fix this since you're export in parquet :) Thanks very much! error message as below:

Failed
Unequal at absolute position 2
@@ -1, +1 @@
-{Backward: 0}
+{Backward: 10}
Expected:
-- is_valid:
[
false,
true
]
-- child 0 type: int64
[
0,
10
]
Actual:
-- is_valid:
[
false,
true
]
-- child 0 type: int64
[
null,
0
]
[ FAILED ] ArrowReadWrite.SingleColumnNullableStruct (7 ms)

[pitrou]

Thanks for doing this. It looks a lot better now! Here are some more comments.

[pitrou]

@tianchen92 The Parquet failure is probably a bug in the nested Parquet reader. I've pushed changes to your branch, be careful to pick them up.

[emkornfield]

@tianchen92 sorry I missed the tag. Is it possible the failure could be related to https://issues.apache.org/jira/browse/ARROW-9603?filter=-2

[tianchen92]

@pitrou Thanks for your patch and careful review, I updated PR according to the comments above.
@emkornfield doesn't matter :), pitrou helped disabled the failed test and now seems unblocked.

[nealrichardson]

@tianchen92 do you want to pick this up again? the bug that @emkornfield mentioned has been fixed.

[tianchen92]

@emkornfield @nealrichardson @pitrou Sorry for the delay due to National Day Holiday, I reverted parquet related changes and rebased this PR. Seems tests all passed (one unrelated failing). Please help take a look again, thanks!

[emkornfield]

@pitrou any objections to merging now (I would expect this might conflict with Decimal256 PR).

[pitrou]

Working on this.

[pitrou]

Added some fixes and tests. +1.