Merge branch 'alamb/parquet_decoder' of github.com:alamb/arrow-rs into alamb/parquet_decoder

Author: alamb

parquet/src/arrow/in_memory_row_group.rs

@@ -256,10 +256,14 @@ impl RowGroups for InMemoryRowGroup<'_> {
     }
 }
 
-/// An in-memory column chunk
+/// An in-memory column chunk.
+/// This allows us to hold either dense column chunks or sparse column chunks and easily
+/// access them by offset.
 #[derive(Clone, Debug)]
 pub(crate) enum ColumnChunkData {
-    /// Column chunk data representing only a subset of data pages
+    /// Column chunk data representing only a subset of data pages.
+    /// For example if a row selection (possibly caused by a filter in a query) causes us to read only
+    /// a subset of the rows in the column.
     Sparse {
         /// Length of the full column chunk
         length: usize,

parquet/src/arrow/push_decoder/reader_builder/filter.rs

@@ -47,7 +47,9 @@ pub(super) struct FilterInfo {
 /// but it owns the ProjectionMask and RowGroupCache
 #[derive(Debug)]
 pub(super) struct CacheInfo {
-    /// The columns to cache in the predicate cache
+    /// The columns to cache in the predicate cache.
+    /// Normally these are the columns that filters may look at such that
+    /// if we have a filter like `(a + 10 > 5) AND (a + b = 0)` we cache `a` to avoid re-reading it between evaluating `a + 10 > 5` and `a + b = 0`.
     cache_projection: ProjectionMask,
     row_group_cache: Arc<Mutex<RowGroupCache>>,
 }
@@ -69,9 +71,9 @@ impl CacheInfo {
 }
 
 pub(super) enum AdvanceResult {
-    /// advanced to the next predicate
+    /// Advanced to the next predicate
     Continue(FilterInfo),
-    /// no more predicates returns the row filter and cache info
+    /// No more predicates returns the row filter and cache info
     Done(RowFilter, CacheInfo),
 }
 

parquet/src/arrow/push_decoder/reader_builder/mod.rs

@@ -211,7 +211,7 @@ impl RowGroupReaderBuilder {
     /// call to [`Self::take_state`].
     ///
     /// Any code that calls this method must ensure that the state is put back
-    /// before returning, otherwise the reader error next time it is called
+    /// before returning, otherwise the reader will error next time it is called
     fn take_state(&mut self) -> Result<RowGroupDecoderState, ParquetError> {
         self.state.take().ok_or_else(|| {
             ParquetError::General(String::from(