Parquet code generation for Rust

This project provides tools for generating Rust code to work with Parquet files using the Rust implementation of Arrow. It includes both a code generation crate (parquetry-gen) and a small runtime library required by the generated code (parquetry).

Please note that this software is not "open source", but the source is available for use and modification by individuals, non-profit organizations, and worker-owned businesses (see the license section below for details).

Example

Given a schema like this:

``` message user { required int64 id (integer(64, false)); required int64 ts (timestamp(millis, true)); optional int32 status;

optional group user_info {
    required byte_array screen_name (string);

    optional group user_name_info {
        required byte_array name (string);

        optional group user_profile_info {
            required int64 created_at (timestamp(millis, true));
            required byte_array location (string);
            required byte_array description (string);
            optional byte_array url (string);

            required int32 followers_count;
            required int32 friends_count;
            required int32 favourites_count;
            required int32 statuses_count;

            optional group withheld_in_countries (list) {
                repeated group list {
                    required byte_array element (string);
                }
            }
        }
    }
}

} ```

The code generator will produce the following Rust structs:

```rust

[derive(Clone, Debug, Eq, PartialEq)]

pub struct User { pub id: u64, pub ts: chrono::DateTime, pub status: Option, pub user_info: Option, }

[derive(Clone, Debug, Eq, PartialEq)]

pub struct UserInfo { pub screenname: String, pub username_info: Option, }

[derive(Clone, Debug, Eq, PartialEq)]

pub struct UserNameInfo { pub name: String, pub userprofileinfo: Option, }

[derive(Clone, Debug, Eq, PartialEq)]

pub struct UserProfileInfo { pub createdat: chrono::DateTime, pub location: String, pub description: String, pub url: Option, pub followerscount: i32, pub friendscount: i32, pub favouritescount: i32, pub statusescount: i32, pub withheldin_countries: Option>, } ```

It will also generate an instance of the parquetry::Schema trait for User with the code for reading and writing values to Parquet files.

Usage

The example directory provides a fairly minimal example, and the generated code is checked in there. In most cases a build.rs like thie following should be all you need:

```rust use std::{fs::File, io::Write};

fn main() -> Result<(), parquetrygen::error::Error> { for schema in parquetrygen::ParsedFileSchema::opendir( "src/schemas/", Default::default(), Some(".parquet.txt"), )? { println!("cargo:rerun-if-changed={}", schema.absolutepath_str()?); let mut output = File::create(format!("src/{}.rs", schema.name))?; write!(output, "{}", schema.code()?)?; }

Ok(())

} ```

By default the generated code is formatted with prettyplease and is annotated to indicate that it should not be formatted by Rustfmt, but if you'd prefer to use Rustfmt yourself, you can set format to false in the configuration.

Status and scope

These tools support schemas with most physical and logical types, and with arbitrary nestings of lists, optional fields, and structures.

Missing features that I might add at some point:

• 8- and 16-bit logical integer types (trivial, I just haven't needed them)
• DATE, TIME, INTERVAL, and UUID (same as previous)
• DECIMAL (at least for the INT32 and INT64 representations)
• ENUM (not really useful in this context since the schema doesn't enumerate the variants?)
• Maps (a little more work but probably worth having)

Features that will probably never be supported:

• The INT96 physical type (which has been deprecated)
• Timestamps with local semantics (isAdjustedToUTC = false)
• Legacy list shapes
• Legacy map shapes
• A way to avoid the chrono dependency or support for other time libraries

This project differs from parquet_derive in a few ways:

• Both generate reading and writing code, but this project generates Rust structs from the schema, as opposed to the reverse.
• This project does not use parquet::record::RecordWriter (it just didn't seem all that useful and I wanted more flexibility).
• This project supports nested structures.

In general the two projects have different use cases, and if you just want to store some Rust values in Parquet, I'd recommend choosing parquet_derive.

License

This software is published under the Anti-Capitalist Software License (v. 1.4).