Microsoft Answers “Top 10 Questions on Data”

I don't think that this message is so clear. L2S is a good LINQ query engine for SQL server with a leightweight ORM. If that's all you need, it's going to be easier to use than EF. That's its strength, so it wouldn't make any sense to cram more features into it.

In my opinion, the message on the data strategy is still the same it has been all along.

The server is great and will get better. .NET is great and will get better. The data access strategies are great, but Microsoft will continue to change its mind so often as to make them a huge moving target, thus even if some are the best thing since sliced bread, you cannot use them because you can't even have the slightest expectation of what Microsoft will shoot at tomorrow.

Take datasets for example. The idea behind them is absolutely great, and while many don't realize it (because no one uses them right =P), the are in many way revolutionary. They were a new way of thinking at the time though, so obviously there's a bunch of design mistakes in them. It could be fixed, but will it? Nope. New data access strategy from the ground up, with marginal effort on datasets (LINQ to Dataset). 10 years of development on Datasets could have kept them current, removed their shortcoming, etc. R.I.P datasets.

All these techs will go the same way. So while i love the Entity Framework (well, since 4.0 anyway), LINQ to SQL, ADO.NET Data services, and all of them...I can't build a lasting solution on them.

When there has been NUnit/xUnit, they invented stupid mstest; when there has been NAnt, they invented stupid msbuild; when there has been NHibernate, they invented EF.

to be fair, they're changing a bit. See: JQuery.

Entity Framework, IMO, is mostly a reaction to all the MS shops that will not use a data access solution if its not Microsoft branded. NHibernate doesn't help since the learning curve and iffy documentation really stings (we use NHibernate and I -love- it, but it has some serious issues). LLBLGEN doesn't have that issue, but it isn't free :)

NAnt doesn't have the following Ant does, so I can see where MSBuild comes from. But they're slowly pulling a "Dataset" stunt on that one: TFS Build 2010 is Windows Workflow Foundation based (it does use MSBuild, but its delegated to the back, like it is in Visual Studio, and a lot of people don't even realize its there...)

The Mono community is working on an API-compatible LINQ to SQL replacement: code.google.com/p/dblinq2007/

looks like it's not really production ready yet, but is still under active development. This may be a way forward for LINQ to SQL users if MSFT abandons their implementation.

There are a lot of developers who will howl in rage at this suggestion.

There was a famous petition from the Beta developers of Linq to EF, that basically said "use is sparingly and with caution". There were a bunch of issues, including that it's canonical / Least-Common-Denominator approach to the "intermediate" query language (suits everthing from SQL server, to XML and beyond) resulted in unnecessary bad performance. It's still online.

Some developers become advocates for L2EF others abandon it early and some have even converted to L2EF then gone back to L2S for performance reasons.

I don't have an up-to-date re-analysis of everything since .NET 4 was released but this isn't clear cut at all. L2S got updated for .NET 4 so reports of it's death are greatly exaggerated.

Then there's things like Plinqo. In that (and other products) third parties have taken L2S and added functionality that MS didn't. It might cost a bit more but there's an ardent band of smart people out there supporting L2S. Some corporations depend on it. Death is unlikely any time soon!

Just give it a try with something like LinqPad to give you a quick, thoroughly satisfying development experience. You'll see why Matt Warren's creation might just be here to stay. (And then Mono is seeking to emulate his long and sophisticated processing pipeline too. Taking this great implementation to a wider audience.)

(Of course if you want performance you need profiling and quite likely T-SQL / M / Whatever. You can integrate that into the LINQy things without throwing it all away. Thereby taking a possible performance dog and turning it into a performant solution, if that's what you need!)

Lastly the fact that L2S only does SQL Server, is not intrinsic in the design. It's a political / funding decision. (Matt has good posts about how you can plug your own data provider into this sort of thing.)

I was not saying L2S would die, quite to the contrary: It has found it's spot, but it's not likely to aggregate many new features as that would put it in competition with EF. Which would definately be a bad thing, because then nobody would dare to bet their money on either one, assuming that eventually one will win and one will die.

If you don't like EF, fine, neither do I. There are good alternatives out there, ranging from DbLinq to NHibernate in the free world, plus a whole range of commercial ones.

The main problem of L2S is its dependence on MS SQL Server. The way things are going today, I'm surprised it hasn't been packaged with SQL server. Not that it would make any difference at that point.

There's of course also Matt Warren's open-source-followup: IQToolkit...

iqtoolkit.codeplex.com/
blogs.msdn.com/b/mattwar/

MS have said that a reason why they provide alternatives to open source implementations is that many of their customers will not consider a technology at all unless Microsoft provides it. I work as a contract developer and I have certainly encountered this attitude at some client sites.