doc: book section on lending row iterators

Author: molpopgen

book/src/table_collection_row_access.md

@@ -1,6 +1,43 @@
 ## Accessing table rows
 
-We may also access entire table rows by a row id:
+The rows of a table contain two types of data:
+
+* Numerical data consisting of a single value.
+* Ragged data. Examples include metadata for all tables,
+  ancestral states for sites, derived states for mutations,
+  parent and location information for individuals, etc..
+
+`tskit` provides two ways to access row data.
+The first is by a "view", which contains non-owning references
+to the ragged column data.
+The second is by row objects containing *copies* of the ragged column data.
+
+The former will be more efficient when the ragged columns are populated.
+The latter will be more convenient to work with because the API is a standard
+rust iterator.
+
+### Row views
+
+To iterate over all views we use *lending* iterators:
+
+```rust, noplaygound, ignore
+{{#include ../../tests/book_table_collection.rs:get_edge_table_rows_by_lending_iterator}}
+```
+
+#### Lending iterators
+
+The lending iterators are implemented using the [`streaming_iterator`](https://docs.rs/streaming-iterator/latest/streaming_iterator/) crate.
+(The community now prefers the term "lending" over "streaming" for this concept.)
+The `tskit` prelude includes the trait declarations that allow the code shown above to compile.
+
+rust 1.65.0 stabilized Generic Associated Types, or GATs.
+GATs allows lending iterators to be implemented directly without the workarounds used in the `streaming_iterator` crate.
+We have decided not to implement our own lending iterator using GATs.
+Rather, we will see what the community settles on and will decide in the future whether or not to adopt it.
+
+### Row objects
+
+We may access entire table rows by a row id:
 
 ```rust, noplaygound, ignore
 {{#include ../../tests/book_table_collection.rs:get_edge_table_row_by_id}}

tests/book_table_collection.rs

@@ -69,6 +69,7 @@ fn add_node_handle_error() {
 
 #[test]
 fn get_data_from_edge_table() {
+    use tskit::prelude::*;
     use rand::distributions::Distribution;
     let sequence_length = tskit::Position::from(100.0);
     let mut rng = rand::thread_rng();
@@ -119,6 +120,23 @@ fn get_data_from_edge_table() {
         panic!("that should have worked...");
     }
     // ANCHOR_END: get_edge_table_row_by_id
+    //
+    // ANCHOR: get_edge_table_rows_by_lending_iterator
+    let mut edge_table_lending_iter = tables.edges().lending_iter();
+    while let Some(row_view) = edge_table_lending_iter.next() { 
+        // there is only one row!
+        assert_eq!(row_view.id, 0);
+        assert_eq!(row_view.left, left);
+        assert_eq!(row_view.right, right);
+        assert_eq!(row_view.parent, parent);
+        assert_eq!(row_view.child, child);
+        assert!(row_view.metadata.is_none()); // no metadata in our table
+    }
+    // ANCHOR_END: get_edge_table_rows_by_lending_iterator
+
+    assert!(tables
+        .check_integrity(tskit::TableIntegrityCheckFlags::default())
+        .is_ok());
 
     // ANCHOR: get_edge_table_rows_by_iterator
     for row in tables.edges_iter() {