Code-First or Database-First, how to choose?

Question

Let us suppose we are going to start new project - application that contains some business logic, user interface on ASP.NET, WPF or both of them. We'd like to use ORM or DAL code generator and implement our business logic in .NET classes. There are several fundamental ways how we can express our ideas of business domain:

• Implement business classes on .NET and let ORM generate appropriate database schema
• Create database schema manually and generate .NET classes by code generator
• Use some kind of visual designer, that can generate business classes and database structure or script

What do you prefer to write: "Create Table Persons ( ... )" or "public class Person { ... }"?
What are Pros and Cons of those ways?
Maybe there are some special situations where one way is better than another?
How to choose optimal way in a particular project?

I am quite familiar with "Code-First" (or "Model-First") way, but it seems most of ORMs are designed as code generators or mappers, that suppose that I will manually implement both database structure and business classes.

Answers based on expirience and examples of ORM's are especially welcome.

Edit: Note, the question is not "What should I do first when starting new project?", but "What should be manually declared / automatically generated, domain classes or database structure?"

Answer 1

I think the appropriate approach to system analysis and design is to start by modeling your objects and the relations between them first. If you're creating a library system you should think of the phrases Book, Author, Publisher, ISBN as objects not as database tables or attributes. I believe this is the way it should be. That been said, let's admit that code generators save way a lot of time, and those require a relational database in order to generate the model and map it to the DB objects. I think this is the major reason why developers tend to start by the D.B. What could prove my point more is that code generators developers is trying hard to reverse the currently implemented operation (i.e. You provide a business model-objects and classes- and the generator creates the DB with the appropriate schema for this).

Edit:
Here's an example of domain-first generators (ADO.NET Entity Framework itself) Model First :

Visual Studio 2010 has to ability to generate a DDL and create a database to store the entity data model. The developer has complete control over the entire process being able to customize the DDL, or to select the database he desires, or fine tune the mapping process.

Answer 2

Analysis of requirements first, and then some documentation of those needs and an overview of the data aspects of this?

Then you know what data you'll be capturing and how it relates to other data, and can design a database schema or data structure to match it (as logical objects/tables of related content, not "tab1_data", "tab2_data" matching the data capture process which could change, but you know that!). You could even design a .xsd first and generate code and a database schema from that. It's all fun and games these days, depending on your skillset.

As the database schema in my mind stores the data, and that is the really important thing for a business to have, I would design that first - multiple tools may access it in time, maybe the original system would be replaced down the line (e.g., migration to newer tools/languages/interfaces). If you know nothing about database theory, then maybe that's not your best option but I would still get any generated schema verified by someone else.

Answer 3

Why not interface-first?

Too many apps start with a program-first mentality. That's a bad idea. Programming is the heaviest component of building an app, meaning it's the most expensive and hardest to change. Instead, start by designing first.

Design is relatively light. A paper sketch is cheap and easy to change. html designs are still relatively simple to modify (or throw out). That's not true of programming. Designing first keeps you flexible. Programming first fences you in and sets you up for additional costs.

This is from Chapter 9 of Getting Real by 37signals.

Answer 4

Start by not directly thinking about either, rather "model" (preferably on paper) what parts your application will have from the users point of view.

If you have a clear mental picture of that model, You can divide the parts up in common and specific little elements which you can translate to both object definitions and database tables.

I find this method to cut database normalization time and effort significantly.

Answer 5

In my opinion there is no correct answer to this. I guess it mostly boils down to your own personal or your teams preferences. All mentioned approaches (database first, code first, interface first) have their own advantages and disadvantages.

I'd probably sit down with pen and paper and sketch up the general structure and the main functions of the application before i do anything specific, be it code or database tables. A simple drawing of the basic user interface also helps a lot.

Answer 6

“Show me your flowcharts and conceal your tables, and I shall continue to be mystified. Show me your tables, and I won’t usually need your flowcharts; they’ll be obvious.”

— Fred Brooks in “The Mythical Man-Month”

Answer 7

In most cases it won't matter much. It is more up to personal preference and skill than anything else. Most apps are not going to suffer much either way, use whatever your team is comfortable with. Where the choice really matters it should be obvious which approach to go for.

That said, my personal opinion is that "database first" is generally the safer choice. If the data is in any way important, especially if it is important outside the scope of your particular app, you want to have full control over how it is stored.

"Code first" (implied: leaving the database in the hands of some automatic tool) is in my mind really a shortcut, one you should use when (and only when) you know for sure you can get away with it.

Answer 8

To answer your edited question (manual db/auto classes or manual classes/db), I'd choose "neither". Autogenerated code of both kinds are to be avoided for a number of reasons, first of all YAGNI. You end up with code you never wrote but are nonetheless responsible for, code you'll never use, and (in my experience) code you'll end up spending more time refactoring than if you'd designed and written it yourself in the first place. And they both keep your focus far away from the most important location - the User.