Table mapping for UniqueColumn uses unstable GetHashCode() method

[Scooletz]

The snippet below presents the way how Table class generates UniqueColumn name. This value is used in several code paths, like CreateForeignKey and/or generating an unique column name itself.

nhibernate-core/src/NHibernate/Mapping/Table.cs

Lines 859 to 872 in c4fb3dc

	public string UniqueColumnString(IEnumerable iterator, string referencedEntityName)
	{
	// NH Different implementation (NH-1399)
	int result = 37;
	if (referencedEntityName != null)
	{
	result ^= referencedEntityName.GetHashCode();
	}
	foreach (object o in iterator)
	{
	result ^= o.GetHashCode();
	}
	return (name.GetHashCode().ToString("X") + result.GetHashCode().ToString("X"));

string.GetHashCode is not a stable hashing function, according to the documentation

The hash code itself is not guaranteed to be stable. Hash codes for identical strings can differ across versions of the .NET Framework and across platforms (such as 32-bit and 64-bit) for a single version of the .NET Framework. In some cases, they can even differ by application domain. This implies two subsequent runs of the same program may return different hash codes.

Creating the script then, can have a different outcome when run in two different environments, like .NET Core 2.1 and net461 or even across 32/64 bits.

[fredericDelaporte]

Anyway using a hashcode for attempting getting something unique is a bug on its own, since hashcodes are never guaranteed unique. Another naming algorithm has to be put in place.

[ramonsmits]

Alternatively use something like https://github.com/LogosBible/Logos.Utility/blob/master/src/Logos.Utility/GuidUtility.cs to generate a cross platform stable uuid

[maca88]

Hibernate fixed this problem by postponing the generation until the reference table is known, after that they generate the name by concatenating the table and the columns names and hashing the result with the MD5 algorithm in order to avoid too long names.

[fredericDelaporte]

We should likely do the postponing. But then, the hashing part can be troublesome, since collisions may occur, even if it is quite unlikely. Moreover it guarantees a name only shorter than thirty characters, while we have dialects having MaxAliasLength as small as eighteen.

[fredericDelaporte]

Worse for the .Net Core case, as written in #1791, the generated names change at each run under .Net Core. So .Net Core is in the following case:

In some cases, they can even differ by application domain. This implies two subsequent runs of the same program may return different hash codes.

The following code yields different strings at each run under netcoreapp2.0:

[Test]
public void UniqueName()
{
	var tbl = new Table { Name = "name" };
	Console.Write(tbl.UniqueColumnString(new[] { "col1", "col2" }, "ent1"));
}