3 unstable releases

new 0.2.0 Apr 25, 2024
0.1.3 Dec 23, 2023
0.1.2 Nov 20, 2023

#553 in Web programming

Download history 12/week @ 2024-02-19 31/week @ 2024-02-26 8/week @ 2024-03-11 9/week @ 2024-03-25 46/week @ 2024-04-01 143/week @ 2024-04-22

192 downloads per month

Apache-2.0

49KB
759 lines

dto_mapper

This is a library to create dynamic DTOs for rust based project. It has the same purpose with the known Java DTO Mapper Library called Model Java used mainly in Java SpringBoot applications.

DTO stands for Data Transfer Object. It is a Software Design pattern that involves mapping a parent object to a new one by reusing some of its properties into the newly created one. For example, if one application is manipulating sensitive data such as credit card information, passwords, or any other high sensitive information that shouldn't be sent(serialized) over the network or displayed to the user in any form. So, applications need a way to map this Entity to a new one by removing the properties or fields they wouldn't want to expose to a user in JSON or any other format.

This library makes it handy. It helps annotate a structure with special attributes to extract fields to compose new structure objects for re-use. This library is based upon the Syn Library and the power of macro derive attributes embedded in rust. The dtos are generated during compile time. There is no overhead during runtime for this as the final binary will contain the dto structure resulted after buil time. I would recommend using Visual Studio code with rust analyzer plugin installed for auto-completion and a nicer developer experience with preview of field names when hovering over a dto structure.

Summary

This library can be used by web applications which used to do data transformation between database and the controller that sends information to the user using DAO(Data Access Object) pattern. Let's consider a User Entity used by an application to store into and retrieve records for a user from a database. Let's describe our User entity like this:

    struct User{
        username: String,
        password: String,
        email: String,
        firstname: String,
        middle_name: Option<String>,
        lastname: String,
        age: u8,
    }

If we have to load a 'user' record from a database, we wouldn't want to send all this information back to a webpage for different scenario. We would want to remove the password information from the user if we need to reuse it to send it to a webpage. One will insist that we can use Json Serializer Library and annotate the password field in such a way we can ignore it. Well, let's push it a little bit further. If we have a Backend application that serves client for Authentication request,Profile Information request, or to request other information in form of other data mix that would want only the firstname, lastname and the age of that person. Json annotation wouldn't help. And we would have to create by hands each of those Objects and repeated the same information and implement proper methods to convert one to another. That sounds like too much work. This is where DTO Mapper Library comes handy.

Installation

dto_mapper library depends on unstringify and derive_builder which implements builder pattern for dto objects resulted from the dto mappers. By default, it generate builder for the dtos. You can use this instruction to install the latest version of dto_mapper library to your project

cargo add derive_builder
cargo add dto_mapper
cargo add unstringify

And import it to the rust source file you need to use it:

use dto_mapper::DtoMapper;

More details on how to use derive_builder crate here: https://crates.io/crates/derive_builder

Example

Let's say we want to create 3 special entities for our application

  • LoginDto that will contain only 2 fields from User such as username and password. we would like to rename username to login in LoginDto
  • ProfileDto that will contain all fields from User and will ignore only the password field.
  • PersonDto that will contain only 3 fields from User such as firstname, lastname, and email. But we would like to make the email field optional such that its final data type will be Option. That is if email is String from User, it will result in Option

It takes only those lines below to get this work done. And the conversion are being done automatically between one dto type to the original struct and vice versa.

  use dto_mapper::DtoMapper;

  /*** Use this declaration below in lib.rs if you're using a library crate , or in main.rs if you're using a binary crate.
   if your crate has lib.rs and main.rs. Use it instead inside your lib.rs.
  ***/
  #[macro_use]
  extern crate derive_builder;
  #[allow(unused)]
  use std::str::FromStr;
  #[allow(unused)]
  use unstringify::unstringify;

  fn concat_str(s1: &str, s2: &str) -> String {
      s1.to_owned() + " " + s2
  }

  #[derive(DtoMapper,Default,Clone)]
  #[mapper( dto="LoginDto"  , map=[ ("username:login",true) , ("password",true)] , derive=(Debug, Clone, PartialEq) )]
  #[mapper( dto="ProfileDto" , ignore=["password"]  , derive=(Debug, Clone, PartialEq) )]
  //no_builder=true will not create default builder for that dto
  #[mapper( dto="PersonDto" , no_builder=true,  map=[ ("firstname",true), ("lastname",true), ("email",false) ]  )] 
  #[mapper( dto="CustomDto" , no_builder=true , map=[ ("email",false) ] , derive=(Debug, Clone) ,
      new_fields=[( "name: String", "concat_str( self.firstname.as_str(), self.lastname.as_str() )" )]
  )]
  struct User{
      username: String,
      password: String,
      email: String,
      firstname: String,
      middle_name: Option<String>,
      lastname: String,
      age: u8,
  }

let login_dto : LoginDto = LoginDto::default();
let profile_dto: ProfileDto = ProfileDto::default();

Let's consider we have a User struct value and we would like to convert it back to a dto object:

        let user = User{
            username : "dessalines".to_string(),
            email: "dessalines@mail.ht".to_string(),
            password: "XXXXXXXXXXXXX".into(),
            firstname: "Dessalines".to_string(),
            lastname: "Jean jacques".to_string(),
            age: 50,
            ..User::default()
        };

        //clone user as into moves user after into() operation in order to reuse user in subsequent calls
        let lg_dto_user : LoginDto = user.clone().into();
        let pf_dto_user: ProfileDto = user.clone().into();

        println!("User to LoginDto = {:?}",lg_dto_user);
        println!("User to ProfileDto = {:?}",pf_dto_user);

Let's consider now we have a PersonDto and we'd like to convert it back partially to a User object knowing that PersonDto misses some fields:

        let person = PersonDto{
            firstname: "Dessalines".to_string(),
            lastname: "Jean Jacques".to_string(),
            email: Some("dessalines@mail.ht".to_string()),
        };

        let user_from_person: User = person.into();

Let's consider building LoginDto with the builder pattern object generated by dto_mapper:

        let mut login_dto_builder = LoginDtoBuilder::default();
        let login_dto = login_dto_builder
            .login("capois-lamort".into())
            .password("hello123".into())
            .build()
            .expect("Failed to build login dto");
        println!("LoginDto built with a builder: {:?}", login_dto);

Here is how vscode prints the code generated by DTO mapper for the LoginDto and ProfileDto

pub struct LoginDto {
    pub login: String,
    pub password: String,
} // size = 48 (0x30), align = 0x8

pub struct ProfileDto {
    pub username: String,
    pub email: String,
    pub firstname: String,
    pub middle_name: Option<String>,
    pub lastname: String,
    pub age: u8,
} // size = 128 (0x80), align = 0x8

pub struct PersonDto {
    pub email: Option<String>,
    pub firstname: String,
    pub lastname: String,
} // size = 72 (0x48), align = 0x8

Let's consider building CustomDto by adding a new field called name which will be initialized with concatenation of firstname and lastname fields from User struct with the help of dto_mapper:

        let user = User {
            firstname: "Dessalines".into(),
            lastname: "Jean Jacques".into(),
            ..User::default()
        };

        println!("{:?}", user);
        let custom_dto: CustomDto = user.into();
        println!("{:?}", custom_dto);

Here is how vscode prints the code generated by DTO mapper for CustomDTO and the Into Trait implemented by User for it

pub struct CustomDto {
    pub email: Option<String>,
    pub name: String,
}

impl Into<CustomDto> for User {
    fn into(self) -> CustomDto {
        CustomDto {
            email: Some(self.email),
            name: concat_str(self.firstname.as_str(), self.lastname.as_str()),
        }
    }
}

You can install 'expand' binary crate and use cargo expand command in order to print out the DTO structures generated as shown above:

cargo install expand
cargo expand

Description of macro derive and attributes for Dto Mapper

First, DTO Mapper library requires that the source struct implements Default traits as the library relies on it to implement Into Traits for conversion between DTO and Source struct. If not it will result in error in your IDE. It is a must to to derive or implement Default for your source struct.

#[derive(DtoMapper,Default,Clone)]
struct SourceStruct{ }
  • #[mapper()] attributes

    mapper attributes can be repeated for as many dtos needed to be created. Each mapper represents a concrent dto struct.
    • Required fields will result in build errors if not present.
      • dto : name for the dto that will result into a struct with the same name. Example : dto="MyDto" will result into a struct named MyDto. dto names must be unique. Otherwise, it will result into build errors.

      • map : an array of field names from the original struct to include or map to the new dto as fields. map=[("fieldname:new_fieldname", required_flag )]. fieldname:new_fieldname will rename the source field to the new one. It is not mandatory to rename. you can have map=[("fieldname",true)] required_flag can be true or false. if required_flag is false it will make the field an Option type in the dto.

        if required_flag is set to true, the destination dto field will be exactly of the same type with the source one in the struct.

    • Optional fields
      • ignore : an array of fieldnames not to include in the destination dtos. ignore=["field1", "field1"] if ignore is present , then map field becomes optional. Except if needed rename destination fields for the dto
      • derive : a list of of macro to derive from. derive=(Debug,Clone)
      • no_builder: a boolean flag to turn on or off builders for the dto. Default value is false. If the Dto name is "MyDto" , the builder will create a struct named "MyDtoBuilder" that can be used to build "MyDto" struct.
      • new_fields : an array of declaration of new field names to include to the resulted dto structure. new_fields=[("fieldname:type"), ("initialize_expression") )]. fieldname:type will create a new field with the fieldname specified and the type. It is not mandatory to rename. you can have map=[("fieldname",true)] initialize_expression is used an initialize value or expression to use when converting the original structure to the dto. For instance new_fields=[("name:String"), ("concat_str(self.firstname,self.lastname)") )] will create a new field in the dto called name which will be initialized with the concatenation of the original struct firstname and lastname fields. See the example above. I would strongly suggest to use function as initialize_expression for more complex scenarios in case parsing is failing when writing complex inline expression directly. This will reduce code complexity!!

Dependencies

~0.9–1.4MB
~30K SLoC