Simplified code to infer arrow from parquet

guide/src/io/parquet_read.md

@@ -44,3 +44,51 @@ by delegating all CPU-intensive tasks to separate threads.
 This can of course be reversed; in configurations where IO is bounded (e.g. when a
 network is involved), we can use multiple producers of pages, potentially divided
 in file readers, and a single consumer that performs all CPU-intensive work.
+
+## Apache Arrow <-> Apache Parquet
+
+Arrow and Parquet are two different formats that declare different physical and logical types.
+When reading Parquet, we must _infer_ to which types we are reading the data to.
+This inference is based on Parquet's physical, logical and converted types.
+
+When a logical type is defined, we use it as follows:
+
+| `Parquet`         | `Parquet logical` | `DataType`    |
+| ----------------- | ----------------- | ------------- |
+| Int32             | Int8              | Int8          |
+| Int32             | Int16             | Int16         |
+| Int32             | Int32             | Int32         |
+| Int32             | UInt8             | UInt8         |
+| Int32             | UInt16            | UInt16        |
+| Int32             | UInt32            | UInt32        |
+| Int32             | Decimal           | Decimal       |
+| Int32             | Date              | Date32        |
+| Int32             | Time(ms)          | Time32(ms)    |
+| Int64             | Int64             | Int64         |
+| Int64             | UInt64            | UInt64        |
+| Int64             | Time(us)          | Time64(us)    |
+| Int64             | Time(ns)          | Time64(ns)    |
+| Int64             | Timestamp(\_)     | Timestamp(\_) |
+| Int64             | Decimal           | Decimal       |
+| ByteArray         | Utf8              | Utf8          |
+| ByteArray         | JSON              | Binary        |
+| ByteArray         | BSON              | Binary        |
+| ByteArray         | ENUM              | Binary        |
+| ByteArray         | Decimal           | Decimal       |
+| FixedLenByteArray | Decimal           | Decimal       |
+
+When a a logical type is not defined but a converted type is defined, we use
+the equivalent convertion as above, mutatis mutandis.
+
+When neither is defined, we fall back to the physical representation:
+
+| `Parquet`         | `DataType`      |
+| ----------------- | --------------- |
+| Boolean           | Boolean         |
+| Int32             | Int32           |
+| Int64             | Int64           |
+| Int96             | Timestamp(ns)   |
+| Float             | Float32         |
+| Double            | Float64         |
+| ByteArray         | Binary          |
+| FixedLenByteArray | FixedSizeBinary |

src/io/parquet/read/file.rs

@@ -10,7 +10,7 @@ use crate::{
     error::{ArrowError, Result},
 };
 
-use super::{get_schema, read_metadata, FileMetaData, RowGroupDeserializer, RowGroupMetaData};
+use super::{infer_schema, read_metadata, FileMetaData, RowGroupDeserializer, RowGroupMetaData};
 
 type GroupFilter = Arc<dyn Fn(usize, &RowGroupMetaData) -> bool>;
 
@@ -47,7 +47,7 @@ impl<R: Read + Seek> FileReader<R> {
     ) -> Result<Self> {
         let metadata = read_metadata(&mut reader)?;
 
-        let schema = get_schema(&metadata)?;
+        let schema = infer_schema(&metadata)?;
 
         let schema_metadata = schema.metadata;
         let fields: Vec<Field> = if let Some(projection) = &projection {

src/io/parquet/read/mod.rs

@@ -54,7 +54,7 @@ mod utils;
 pub use file::{FileReader, RowGroupReader};
 pub use row_group::*;
 pub(crate) use schema::is_type_nullable;
-pub use schema::{get_schema, FileMetaData};
+pub use schema::{infer_schema, FileMetaData};
 
 use self::nested_utils::{InitNested, NestedArrayIter, NestedState};
 use deserialize::page_iter_to_arrays;