Note I'm not really updating this anymore. Please see my new iteration of this Ensync.

This is a library for generating and executing SQL diff merge scripts. There are two use cases:

• Generate diff script by comparing an assembly or database with another database using DataModel.Compare. This functionality powers my ModelSync app, which is itself open source: ModelSync.WinForms.

• Create empty tables from C# model classes with DataModel.CreateTablesAsync.

Nuget package: AO.ModelSync.Library

DataModel.Compare

Use DataModel.Compare to generate a diff script from an assembly or database to another database. This example is adapted from this test:

using (var cn = GetConnection())
{
    var sourceModel = DataModel.FromAssembly(@"c:\users\adam\repos\whatever.dll");
    var destModel = await DataModel.FromSqlServerAsync(cn);
    var diff = DataModel.Compare(sourceModel, destModel);    
    string script = new SqlServerDialect().FormatScript(diff);
    Debug.Print(script);
}

Output might look like:

ALTER TABLE [child1] DROP CONSTRAINT [FK_child1_parentId]
GO
ALTER TABLE [child2] DROP CONSTRAINT [FK_child2_parentId]
GO
DROP TABLE [parent]

Source links: DataModel.FromAssembly, DataModel.FromSqlServerAsync

DataModel.CreateTablesAsync

Use DataModel.CreateTablesAsync to create empty tables as part of a component initialization. For example in my WorkTracker project, I create tables from a couple model classes Job and Error. This CreateTableAsync overload accepts a Func<IDbConnection> through which your database connection is opened.

await DataModel.CreateTablesAsync(new[]
{
    typeof(Job),
    typeof(Error)
}, GetConnection);

I have another example along these lines in my Dapper.CX Change Tracking stuff, which creates a couple tables from some model classes.

Background

This is a reboot of my SchemaSync project, which had run into issues I couldn't figure out.

The intent is the same: compare data models to generate a SQL diff merge script. The difference from other database diff apps out there is that this can treat .NET assemblies as data sources -- so that you can merge from C# model classes as well as to and from SQL Server databases. This enables you to use a single tool to migrate model changes end-to-end in your application.

The ultimate goal is to provide a GUI app for use as a Visual Studio "external tool" as an alternative to Entity Framework migrations. I never made peace with EF migrations, and have dabbled in alternative approaches for a long time. I believe it's a better dev experience to merge model classes on demand with a GUI tool rather than interrupting flow to write migrations. Besides being an interruption, migrations have their own administrative complications that, like I say, I never found acceptable.

That said, my approach to model sync works best with a shared development database rather than local database copies. This is because the diff scripting is not always perfect, so a little bit of manual intervention may be needed sometimes. By merging to a shared dev database, you minimize the "finesse" needed to accomplish some merges.

Things different this time around:

• No MySQL support. This will target SQL Server only for now. There are base abstract classes I can build upon later to target other platforms if I get around to it.
• No empty table rebuilds. A lot of the earlier issues I ran into before stemmed from dropping and re-creating empty tables in order to add columns. I did that because I wanted the table column order to match the order of properties in your source files as much as possible. But a lot complexity creeps in from this use case. So, this time around there won't be any table drops and rebuilds as part of the diff algorithm. Table rebuilds will be a standalone feature you can invoke periodically -- distinct from a diff merge -- when you want to align table column order with property order.
• No seed data support initially. Maybe later.
• A smaller dependency footprint. I had a lot of type load exceptions in my old GUI app due to dependencies getting out of date with my Postulate library. The app would have one version of Postulate, while projects might have a different version. I've refactored those dependencies into a new, smaller, more stable project called DbSchema.Attributes. I'm hoping this will clear up type load exceptions in the app as well as be more attractive to dependency-concscious devs. This is all to say that ModelSync is not pure POCO when it comes to inferring metadata like unique constraints and foreign keys. I'll have more to say about developing model classes for ModelSync compatibility.

The forthcoming GUI tool will be closed-source, but the library that powers it, this repo, will remain open source.