tree e418d2ee3534e8eba9c69fb51715c80868853106
parent 4b09ac703bc75fee72f94bed8ecfe571096b04c1
author Toby Shaw <labuser@LABWKS10> 1505978907 +0200
committer Uwe L. Korn <uwe@apache.org> 1505978907 +0200

PARQUET-1037: allow arbitrary size row-groups

The main change made is that you don't have to specify the size of a row-group upfront when writing it. This is signalled through a "row_count_determined" flag which is threaded through the relevant classes.

The AppendRowGroup(int64_t num_rows) method should have identical behaviour as before.
The AppendRowGroup() method should fix its row group after completing a single column, and enforce all
future columns match that. Empty row groups are still not allowed.

(You'll definitely want to squash these commits. Apologies for the poor git usage.)

Author: Toby Shaw <labuser@LABWKS10>
Author: Wes McKinney <wes.mckinney@twosigma.com>

Closes #378 from TobyShaw/master and squashes the following commits:

5ba1b75 [Wes McKinney] * Remove number of expected rows from ColumnWriter in a backwards compatible   way, refactoring * Fix issue where column length was being checked twice after failed write
14bc5e8 [Toby Shaw] Allow arbitrary size row groups