HasColumnName works incorrectly with inheritance

[impworks]

Using HasColumnName with class inheritance works incorrectly: instead of configuring the value for current type, each config overwrites the column name for the entire inheritance tree.

Consider the following model classes:

public class Account
{
    public int Id {get; set;}
    public string Email {get; set;}
}

public class UserAccount: Account
{
    public string Group {get; set;}
}

public class AdminAccount: Account
{
}

And the following configuration:

public class AccountConfig : IEntityTypeConfiguration<Account>
{
    public void Configure(EntityTypeBuilder<Account> builder)
    {
        builder.Property(x => x.Id).ValueGeneratedOnAdd();
        builder.Property(x => x.Email).IsRequired();
        builder.ToTable("Account");
    }
}

public class UserAccountConfig : IEntityTypeConfiguration<UserAccount>
{
    public void Configure(EntityTypeBuilder<UserAccount> builder)
    {
        builder.Property(x => x.Id).HasColumnName("AccountId");
        builder.ToTable("UserAccount");
    }
}

public class AdminAccountConfig : IEntityTypeConfiguration<AdminAccount>
{
    public void Configure(EntityTypeBuilder<AdminAccount> builder)
    {
        builder.Property(x => x.Id).HasColumnName("AccountId");
        builder.ToTable("AdminAccount");
    }
}

Querying code is simple:

var account = await db.Accounts.FirstOrDefaultAsync(x => x.Email == "foo@example.com");

So this is the generated query:

exec sp_executesql N'SELECT [t].[AccountId], [t].[Email], [t].[Discriminator]
FROM (
    SELECT TOP(1) [a].[AccountId], [a].[Email], CASE
        WHEN [a0].[AccountId] IS NOT NULL THEN N''AdminAccount''
        WHEN [u].[AccountId] IS NOT NULL THEN N''UserAccount''
    END AS [Discriminator]
    FROM [dbo].[Account] AS [a]
    LEFT JOIN [dbo].[AdminAccount] AS [a0] ON [a].[AccountId] = [a0].[AccountId]
    LEFT JOIN [dbo].[UserAccount] AS [u] ON [a].[AccountId] = [u].[AccountId]
    WHERE [a].[Email] = @__email_0
) AS [t]
ORDER BY [t].[AccountId]',N'@__email_0 nvarchar(100)',@__email_0=N'foo@example.com'

The LEFT JOIN clauses are incorrect: the correct field name in Account table is [a].[Id], not [a].[AccountId].

While trying to investigate, I tried to use incorrect, but unique column names (UserAccount.AccountId1 and AdminAccount.AccountId2). This is the request I got:

exec sp_executesql N'SELECT [t].[AccountId2], [t].[Email], [t].[Discriminator]
FROM (
    SELECT TOP(1) [a].[AccountId2], [a].[Email], CASE
        WHEN [a0].[AccountId2] IS NOT NULL THEN N''AdminAccount''
        WHEN [u].[AccountId2] IS NOT NULL THEN N''UserAccount''
    END AS [Discriminator]
    FROM [dbo].[Account] AS [a]
    LEFT JOIN [dbo].[AdminAccount] AS [a0] ON [a].[AccountId2] = [a0].[AccountId2]
    LEFT JOIN [dbo].[UserAccount] AS [u] ON [a].[AccountId2] = [u].[AccountId2]
    WHERE [a].[Email] = @__email_0
) AS [t]
ORDER BY [t].[AccountId2]',N'@__email_0 nvarchar(100)',@__email_0=N'foo@example.com'

Notice how every single reference to the key property is now referred to as AccountId2, even when it should be Id in the base table and AccountId1 in UserAccount.

EF Core version: 5.0.0-rc.2.20475.6
Database provider: Microsoft.EntityFrameworkCore.SqlServer 5.0.0-rc.2.20475.6
Target framework: .NET Core 3.1
Operating system: Windows 10
IDE: Visual Studio 2019 16.4.3

[ajcvickers]

Duplicate of #19811

[impworks]

@ajcvickers I came across this bug in actual production code when porting it from EF6 to EF Core. Is it possible to at least display a comprehensible configuration error until the feature is properly implemented?

[AndriySvyryd]

@impworks Due to the way the API is architected we can't detect on which entity type you started the configuration by the time HasColumnName is called. And changing this would be a major breaking change as it would involve changing the return type of Property