Daniel Simmons on ADO.NET Entity Framework April 2, 2007 Our Sponsors

Transcription

1 Carl Franklin and Richard Campbell interview experts to bring you insights into.net technology and the state of software development. More than just a dry interview show, we have fun! Original Music! Prizes! Check out what you've been missing! Text Transcript of Show # 226 (Transcription services provided by PWOP Productions) Daniel Simmons on ADO.NET Entity Framework Our Sponsors

2 Geoff Maciolek: The opinions and viewpoints expressed in.net Rocks! are not necessarily those of its sponsors or of Microsoft Corporation, its partners or employees..net Rocks! is a production of FranklinsNet which is solely responsible for its content. FranklinsNet - Training Developers to Work Smarter. (Music) Lawrence Ryan: Hey, Rock heads! Untape your gerbil from the ST s rocket and listen up, it s time for another stellar episode of.net Rocks!, the internet audio talk show for.net developers with Carl Franklin and Richard Campbell. This is Lawrence Ryan announcing show #226 with guest Daniel Simmons, recorded live Monday, April 2 nd, NET Rocks! is brought to you by FranklinsNet - Training Developers to Work Smarter, and now bringing the Just-In-Time Team System Class with Joel Semeniuk onsite for your development team online at Support is also provided by telerik - Combining the best in Windows Forms and ASP.NET controls with first-class customer service - online at and by CoDe Magazine - the leading independent magazine for.net developers, online at And now, the man who could have sworn the Magic Kingdom was a mushroom farm in Vancouver, Carl Franklin. Carl Franklin: Thank you very much and welcome to.net Rocks! This is the day after hump day, Thursday, and I'm here with Richard Campbell. Of course, we are not here, but we are here with you. Richard Campbell: We are somewhere. Carl Franklin: Yeah. Richard Campbell: I'm still shaking off DevConnections of course. Carl Franklin: DevConnections was a lot of fun; great interview with Dan Appleman. Richard Campbell: Ah, really enjoyed doing that -- and more to come too; that wasn t the only thing we were doing while we were there. Carl Franklin: So, if you are one of the people who know what Richard and Dan were talking about with all that chip stuff, send us an and try to outgeek Richard, because I don t know as if you could. That would be fun in there. Richard Campbell: We got down into the plumbing that day. Carl Franklin: It was fun. So, every once in a while we get an unusual request from one of our listeners, and this one takes the cake I'm afraid. This is from Chris Eyre and he says, Hi guys insert boilerplate praise and slash whatever. Richard Campbell: Very XML compliant there. Carl Franklin: I ve been a big fan of the show since around episode #70 although I eventually caught up. You even read out my Martian s flame. Anyway, I have an unusual request; I lost track of a developer friend of mine, Rupert Johnson. I used to work with him in Poole, England. I'm getting married next year and would like to invite him. It is likely that he listens to the show - or someone who knows him listens to the show. Please can you give a shout out for me? No. Richard Campbell: Yeah. Carl Franklin: Okay, so the answer is no. Next - Thanks again for putting out such a great show. Chris Eyre. And consider this the shout out. Rupert, if you are listening, just send us an - dotnetrocks@franklins.net and let us know where you are. We don t have to read your on the air but I damn sure would like to. Richard Campbell: Yup. There s is a mug in it for you man! Carl Franklin: Absolutely. So, good luck Chris. I hope he is listening. What you got Richard? Richard Campbell: I ve got another ORM . Can you believe it? Carl Franklin: No. Richard Campbell: It s true. Carl Franklin: Wow. Richard Campbell: Guys, friggin awesome show, I listen every week. This week s show in particular was really cool. I'm a huge fan of Rhino Mocks and the work that Ayende has done on this great framework. However, I'm not a huge fan of NHibernate for the very reasons that you guys brought up during the show. DBA has come back and yell at you for the way that most people use NHibernate and the queries it generates. We had actually used NHibernate on a production version of our app, but we ended up having to make workarounds so that we could interface with a DB in specific ways. Enter ibatis.net Carl Franklin: ibatis that s i-b-a-t-i-s. (00:04:54) Transcription by PWOP Productions, Page 2 of 17

3 Richard Campbell: That s right. And actually, I shrinksterized their site; and so, you can go take a look at shrinkster.com/nk0. And ibatis, like NHibernate, originally a Java version, now there s a.net version as well and even a Ruby version which is kind of cool. So, Ben goes on to say, ibatis is the best of both worlds; you have a nice ORM API with your objects to your DB, but you write the SQL that the ORM uses to interface with the DB. This allows developers to map Classes to Queries, as opposed to Tables. This means you can use stored procedures again. Carl Franklin: Right. Richard Campbell: For some reason NHibernate seems to be the only ORM that people are talking about and I feel there needs to be more discussion on the ibatis framework as it is just as clean as NHibernate without sacrificing the ability to customize at low levels. Thanks again for the awesome show; maybe one day I can be on it when my Open Source project NDemo at NDemo.NET takes off. Ben Monroe. Carl Franklin: Yeah, Scott Zischerk sent us a very similar which basically said the same thing. He says, It occurred to me a lot of ORMs and code generators tend to default to use inline SQL calls to access database objects, which means you must give the user or users select rights if not more destructive rights such as delete and update on the underlying tables. I ve been using stored procedures to access database objects so the user or users can only delete and update records in a way that would be safer. That being said, I would like it if you guys could do a show on the pros and cons of different ways to access database objects from.net. Richard Campbell: Do you see a theme here Carl? Carl Franklin: I certainly do; and I think it s going to culminate at the ORM Smackdown in DevTeach. Richard Campbell: DevTeach? Carl Franklin: DevTeach, yeah that conference up in -- where is it, up in your neck of the woods. Richard Campbell: Yeah, it might -- just like Seattle is in your neck of the woods right? Carl Franklin: It s your country not mine pal. Richard Campbell: You are talking about DevTeach at Montreal, Quebec, May 14 th to 18 th. You can visit there at devteach.com and Carl and I will be there. It sure sounds like we are going to end up with an ORM panel at some point -- or something. Carl Franklin: Like I said last Tuesday, if we can t get it together like on stage, we are going to go out to a bar and turn on the microphones but we will have that conversation. Richard Campbell: While we are calling out conferences, let s talk about Mix 07. Carl Franklin: Yeah. Richard Campbell: April 30 th to May 2 nd at the Venetian in Las Vegas. Carl Franklin: That s going to be a huge conference and we also know as we have said before that Microsoft is going to be announcing something very interesting that nobody knows. Richard Campbell: Nobody knows; not even us. Carl Franklin: We don t have any idea. Richard Campbell: Our friend Scott Guthrie is going to be there keynoting again. Carl Franklin: Good show if you are going to go. What else we got? TechEd coming up in Orlando in June... Richard Campbell: That s June 4 th to 8 th. Carl Franklin: That s coming up. Richard Campbell: You better hurry; they sell out early every year. Carl Franklin: Of course we d like to mention the New York tour that we mentioned on Tuesday s show. Greg Brill at Infusion, New York City, is offering a free apartment for a year in Manhattan to work for him plus a New York City salary - shrinkster.com/kh6. Also, there s a gig in Washington D.C. for ASP.NET gurus; a great opportunity there as well. You can read about it at shrinkster.com/mmj. Alright Richard lets introduce our guest today. Daniel Simmons is a Dev Lead working on ADO.NET at Microsoft. He has been at Microsoft for about 10 years working on a variety of products including Outlook, Excel, LiveMeeting, and some challenging but less fortunate projects like NetDocs and WinFS. Before coming to Microsoft he spent several years as the founder of the second-ever ISP in Idaho as well as consulting with a number of small to medium-sized organizations on custom development and IT projects. Welcome to the show Dan. Transcription by PWOP Productions, Page 3 of 17

4 Daniel Simmons: Thanks a lot; it s great to be here. Carl Franklin: Great to have you on. Wow, that s quite a list of products that you ve worked on. Daniel Simmons: Yeah, I never set out to be the guy who moved around Microsoft and did a million different things, but somehow that happened and it s been an exciting experience. Carl Franklin: Your focus is on data; has it always been so? Daniel Simmons: Well, it has been for a long time; even though I worked on a bunch of the Office applications and things like that, most of what I have always done has been the data side of those applications. And so, one of the things that was really cool when I came to the SQL Server and Data Programmability organization was to say, let s bring the experience of looking at applications and how do we approach writing apps that use data to these guys who are spending all of their time down in building the plumbing only. Carl Franklin: And that culminated in ADO.NET. Is that what you are trying to get at here? (00:09:48) Daniel Simmons: Yeah absolutely. So, I originally joined to work in the WinFS team on the API side of that, and then we went through this process of realizing that a lot of the innovation that was going into WinFS applied across all different kinds of backend database storage, and how could we sort of merge that into the general effort and down into ADO.NET. Carl Franklin: Awesome. Well, you did a great job; let me just say, I love ADO.NET. Daniel Simmons: Very cool, I appreciate that. I certainly can t claim responsibility for all of that, but it has been great to join this organization with a lot of great people who have done a lot of great things, and then ride on that and see what new things we can build. Carl Franklin: Do you have any -- I know Richard is chomping at the bit here to ask you a couple of questions, but I got to ask. Are there any keywords or features or anything that you could take credit for, or at least partial credit for - things you would recognize? Daniel Simmons: When you talk about the new stuff coming, the ADO.NET Entity Framework, that s the place where I have been really, in the center of the object services layer of that, essentially from the beginning. Prior to joining the ADO.NET team, all of my work pretty much was in non-managed native coding, so less of it -- the.net keywords there. Carl Franklin: Okay. All right, so you weren t working on ADO.NET 1.0; you re actually working on 3.0 or 4.0. Daniel Simmons: That s right. Carl Franklin: What version are we up to now? Daniel Simmons: Well, I guess it s actually going to be 3.5. Carl Franklin: Okay. Daniel Simmons: So, when they shipped Orcas, the point is that.net framework will be sort of all 3.5, and that s in the next ADO.NET. Richard Campbell: Now, I got to tell the story. So, I first heard from Dan when he fired us an when we were talking about the ORM Smackdown... Carl Franklin: Ah, right. Richard Campbell: Right? Carl Franklin: Yes. Richard Campbell: And ae has talked about the ADO.NET Entity Framework, and I thought that was very -- I mean I ve heard about the Entity Framework for a while now, and so I thought, well, we re going to the MVP Summit, so we ll have a chance to meet him. So, I just said, I ll be there, I ll make sure to see that talk -- not that he was giving it, but he d be there. So I go -- and I am on the ASP.NET side -- and the MVP Summit tends to put you into your team. The big thing that I think the MVP Summit gives more than anything else for MVPs is direct access to your team; the guys who are doing the work, you can get to talk to them and see what they re working on. It s very exciting stuff. So, I crossed over actually to change buildings over to building 42, found this session, this ADO.NET Entity Framework introduction session, sat in the back, and watched. And of course, MVPs -- for the most part, you really know each other, and there are the usual suspects. Carl Franklin: Right, okay. Richard Campbell: So, in a little cluster in the center on the right, I called them the NHibernate Mafia; and it was James Kovacs, Scott Bellware, Jeffrey Palermo, Jean-Paul Boodhoo, all in a little cluster, and they are just hammering these guys. Transcription by PWOP Productions, Page 4 of 17

5 I mean they are deeply -- I ve teased these guys because they are all friends of the show without a doubt -- and Jeffrey Palermo s party, by the way, at the MVP Summit - off the hook - it was crazy. I didn t even make it there, and I kept hearing about it. Carl Franklin: It s too bad I missed that. Richard Campbell: Yeah, it was a heck of a party. So, these guys -- I mean they buy ORM in a very deep way; and they are all NHibernate believers -- they are in there. and so, they are looking at this product in its very early stages, which I get -- they are just getting started on this, it s just coming along now, been thought through a few times, and they are comparing it to NHibernate and just boom, boom, boom. And Dan, I d never met him before; he was sitting in the left, right at the front, and I figured out because there was couple of his guys in the back that it was him. And when the session broke, he made the mistake of popping up a couple of times and answering questions. So now they knew who he was; he had the target pinned on his head. When the session broke, it took him an hour to get out of the room. The boys were just on him; and even when he got out of the room, they were still on him. So, I think we exchanged like four sentences, and I said, Look, I really got to get you on this show. Can I book you right away? And he said, Absolutely. He says, I am going to leave you alone, because I ve been in your hot seat where you are surrounded by guys who are just on you. You go to work, this is the gig. I am going to go; I am going to go get something to eat. Daniel Simmons: It was a cool discussion; it was so good in fact that a group of us said, Hey, you guys, clearly we need to talk some more. So the next day we scheduled a lunch after one of the other sessions, and just all sat around in a room in front of a whiteboard with a big group, and just went at it. Carl Franklin: That is so cool. Daniel Simmons: I think one of the things that was very apparent to us -- several of the folks from the ADO.NET team talking to these folks was that, there is this impression that this group of people have that we just completely don t get their style of working. We don t get what they do with NHibernate; we re trying to build this other product. (00:15:06) Carl Franklin: That s an impression, as you say. Daniel Simmons: Yeah. So, the more we talk about it, we work through that. At one point I stopped Jeffrey Palermo when I sat down and talked for maybe an hour straight, and I said, Okay, here, let s stop for a second and let me give you a vision for a few releases out, this is what it could be like. And I just started describing this and he is like, Yeah, perfect, you ve got it, okay. Now let me describe to you why that s not coming this release, and I am sorry, yes, you will stick with NHibernate a little longer. Carl Franklin: I seem to remember talking to -- I think it was Jimmy Nelson about the entities product, and how he blogged about it, and after he blogged about it he says, he has no idea what happened, but all of a sudden all traces of it were gone from Microsoft s website. Like, the product just disappeared. And so, he sort of kids everybody, and people kid him that he killed the product. Daniel Simmons: So, it was funny to me. I have been listening to the show for a while, and I remember listening to this exact session and laughing on my way to work that, yeah, nobody on the product team was even aware of his blog entry, when that thing went out and came back. He was like, oh well, yeah, that s interesting; maybe I should go read that blog entry there. It s a typical stuff that happens with these -- when you re trying to forge a lot of new ground, and there are a lot of different teams inside Microsoft and a lot of stuff going on, and somebody says, Hey, I want to share some stuff about this. So, something goes up, and then, oh, wait somebody, no, wait, wait -- you know, got to get the coordinated message. Richard Campbell: Yeah. It is a complicated machine, the inside of Microsoft. Carl Franklin: So, you are talking about the Entity Framework; let s define that before we go any further. Daniel Simmons: Sure. So, if we take a look at the Entity Framework, it s kind of a broad topic. So, I ll give a little bit of vision, and then we can sort of dive in and talk about what actually appears in Orcas and in future releases. So, the idea of the Entity Framework is to create a new data model that s a higher abstraction than what you get from the Relational data model. That s sort of closer to the way that we actually think about right applications, right business logic, but that has the same kind of theoretical underpinnings that -- we ll give it sort of long term legs that you get out of the Relational model. So, the basis of this is the Entity-Relationships kind of model that was defined by this guy, Dr. Peter Chen a long time ago, 1976 it was when he Transcription by PWOP Productions, Page 5 of 17

6 published the first paper on it. But usually the way people use the Entity Relationships in practice is, they draw pictures on a whiteboard or maybe in an electronic document that tell how they re going to reason about things, but then they translate that into some completely different form when it comes time to actually build the Relational database and build your application logic. So, the idea of the Entity Framework is that we make a real runtime representation of the structure -- the model of your data and the constraints on that, that s in terms that are much closer to the way your application runs. Richard Campbell: More like the way people think about it. And I keep seeing the whiteboard drawings we do where we just draw sales - and you know, that s a raft of tables. Daniel Simmons: Right. Carl Franklin: So, it would be fair to say that in general this falls into the ORM category. Daniel Simmons: It does, but there is a pretty key distinction; because one of the things that we re trying to do is capture just the structure and the relationships and constraints on the data as a separate thing before you get to the point of actually building your Objects that have Behaviors, and that have business logic on them. The reason why we do that is that we have this realization in a company, not only we have this ORM problem that s been going for some time where people want to write business logic, and they need to do it in terms of their Objects, but you ve also had this same kind of problem showing up in a variety of other things. Like, if you go look at Replication and you say, I am going to Replication from SQL Server to a client machine because I want to have an offline operation. So I am going to use Merged Replication that s not replicating everything in the table, and suddenly I want to pick pieces of several different tables to make up one logical entity and make sure that that thing replicates all at once or not at all. Richard Campbell: I remember this. Dave Campbell brought this up at the beginning of the session, exactly that scenario that Merge Replication just sort of opened this can of worms up, because suddenly you wanted to keep this data in sync between these two machines; you didn t necessarily need it all, and you didn t want partials. It had to be -- the reality of Replication is, you re always further away than you wanted to be; and the connection fails and you wanted to be able to know where to rollback to. So, you ve got to be able to define a transaction around that Replication. Carl Franklin: Yeah, that s a big challenge. (00:19:55) Daniel Simmons: Well, it s a challenge that the Replication team has been taking on for a while and they had been building their own way of defining the pieces that go together to make up a logical unit for Replication. And then we said, Okay, that sounds kind of like what we are doing for Objects, and then we look a little farther around the company and we come to people working on Reporting. And they say, well I have the exact same problem; I want to pick these pieces from a table and I want to secure them with a security token so that I can filter on them and only show certain ones that then reports to certain people, and I want to manipulate them and have an authoring language that operates at this higher level. So, they defined their own thing that talks about how the pieces go together. And then you go talk to the WCF guys and they are defining the pieces that go together, that show up in an XML document or some other form when you re Remoting. And suddenly you find that this same problem is being redone in separate silos all around. We d like to find one representation that you can say, Okay, I want to build the model of the Logical Entities that work in my application and the way they are related to each other, and then I want to reuse that information in all of these different places. Carl Franklin: Who said it? Scott Hanselman has a famous quote that s like, Every problem can be solved by adding one more layer of indirection. Daniel Simmons: Absolutely. Carl Franklin: I don t know who said it. Daniel Simmons: Hopefully we find a way to do that, so that that one more layer not only added value to the programmer time, but it didn t hurt the performance. That s the big challenge. Richard Campbell: That s always the battle with more indirection is, it s more difficult to understand, more challenging to code, and everything runs slower. Daniel Simmons: So, one of the ways that people do this today is they say, well we ll build it all in Object Relational Mapping, and then we ll take these sets of objects, and we ll use them as the basis for writing all of our other things. Then they discover well, Gee, the performance didn t work really well from our report. So, then I ll go write -- optimize that by hand, and throw out all of my Mapping, and write my own queries. Transcription by PWOP Productions, Page 6 of 17

7 Carl Franklin: Just got to get one of those 90 processor core chips from Intel or whatever. Daniel Simmons: And deploy it to every one of your customers. Carl Franklin: Absolutely. Richard Campbell: And several terabytes of RAM; you ll be fine. Carl Franklin: No problem, problem solved. Next. Daniel Simmons: So, in the interim Carl Franklin: That s why I get the big bucks man. Daniel Simmons: So, this is the strategy that is sort of, I think one of the unique contributions of the Entity Framework is to sort of separate this idea of the model and the structure from actually going to the Objects. So, we talk about having a concrete model, which is the way your database is actually structured, and then having a conceptual model, which is still -- you access it just like you access other data in ADO.NET. So you have a connection and you have a command, and you get back a reader, and it s all highly optimized so that as you ask for each column and each row, it only pulls the data over the network as it needs to, and all of these things. We do the same thing at a next layer up that does mapping and reshaping, so that you get data that is in terms of your conceptual model, but it s still super efficient, and it works and acts and smells just like ADO.NET. Carl Franklin: Awesome. Daniel Simmons: Then we add another layer on top of that if you want, for some part of your application where you can actually build your objects and write business logic that way. So, there are sort of these two steps. One of the things that we encounter with all of these folks who are coming from NHibernate, because they are focused so much on the Objects, they re a little bit frustrated that what we have added at that layer doesn t yet have all the features they want. Carl Franklin: I see. Daniel Simmons: And one of our messages is to say, yes, what we really want is to have you understand this foundational piece where we get this mapping piece and it can be reused in all these different ways, and it meets your needs and all of the rest of our customers. Then over time, we can improve on just the object piece -- and by the way, that also gets easier, because it doesn t have the mapping in the middle of it. Carl Franklin: Right, and by the time that s done, it s going to have more meat underneath. So it ll actually be a lot more powerful and cover a lot more ground. Where does it fit in terms of the user interface? Daniel Simmons: So, interestingly enough, one of the things that happens today I think is that people end up writing applications where they put explicitly extra layers, because they don t have a good way to do Object Relational Mapping, and merge it with their Domain Objects. So, they ll have these DTO Objects that are just carrying around the data, and then they ll wrap every one of them in a business object, and then they ll put a UI at the next layer up on top of that. And one of the things that s interesting that we can do here is, we can say, the mapping gets it into the right shape, and then you don t have to have two layers of objects; the Domain Objects come directly out of that, and then you can build your UI on that, which maybe work very naturally on top of that. We can have Data Binding services where if your Entity is an order and you want to have a grade of orders to data bind directly to those objects, or it might be that you want to do a higher order transformation on top of that, because your UI needs to look at it in yet another form. (00:25:06) Carl Franklin: In terms of the developer UI, does it live in Visual Studio? Is there any kind of -- does it hook into the designer Daniel Simmons: Yeah. So, today what we re doing is, in Orcas what we re going to ship is definitely some integration with Visual Studio, but probably won t have all of the tools that we want long term. So, you can just go into Visual Studio with like -- if you get the March CTP of Orcas, and you can say, I want to add a new item to my project, and you say, I want an ADO.NET Entity item, and you get this little wizard, and you can pick either Go to a database, and you give it a way to connect to a database and you pick the tables, and it will sort of auto-generate a starting point model for you. Or you can say, take me to an empty conceptual schema definition file, and you can write your conceptual model, sort of, from scratch, and then do a separate exercise to go and map that model down to the database. The current plan is that what actually ships in the box in Orcas is pretty much just that experience, but that we re going to ship sort of, out of band on the Web, more visual tools around visually designing the schemas, and maybe even getting to a point of visually specifying the mapping. Transcription by PWOP Productions, Page 7 of 17

8 Carl Franklin: Okay. Daniel Simmons: Today you have to do that in a declarative format, and long term, I think we believe that we re going to need a whole variety of these; you re going to need the visual one, you re going to need one where you actually author the XML files that we have today; we are also going to want to have the ability to say, Hey, just go to my classes and put a few attributes on them, kind of, like the experience you can get in some of the other products. In any of those ways that you specify it, it still turns out into the same runtime representation that we have of the Metadata system that s part of the Entity Framework that knows about all of these mappings. Carl Franklin: Now, since it s based on or related to ADO.NET, I assume that it can use whatever data provider that you want to use. Richard Campbell: That s exactly what I was thinking is like -- so we are not bound, this doesn t all have to be from SQL Server; this is anything we can cram into a dataset. Daniel Simmons: Yeah. So, this is a critical point, and unfortunately it turns out, we do need a change to the providers, but the change that s needed is relatively small, and we ve been working with the providers. We had multiple provider fests already, where people came in, and we said, all right we want you to work on making changes to your provider in the course of this week, and by the end of the week they had queries up and running. Carl Franklin: Provider Fest - I got to going to look that up in Wikipedia. That s a good one, I love it. Daniel Simmons: So, the basic technical point that we discovered is, this problem of writing queries that run against multiple backend providers and do it in a predictable fashion is a very difficult one. And ADO.NET took a different stance on this from previous data access technologies in Microsoft in that we said, You know what, we are not in that business; we just hook you up and then you give us a string and we just send it down to the provider, and we don t know what that string means and we ll get back to you. But suddenly we have this problem that we have to understand what that string means, and we have to reshape it and do all these mapping kinds of exercises. So, we defined this thing that we call a Canonical Query Tree; it s just a tree format of your query, that s in terms of Relational algebra and in terms of the schema of the provider. The one change we needed, the provider needs to take that tree and translate that into their actual SQL. But because the tree is so SQL-like, that translation process is really quite straightforward. Carl Franklin: The other question that comes to mind is, how does this work with LINQ or DLINQ? Daniel Simmons: Absolutely, it s a good question. First we ll talk about how LINQ works over the Entity Framework, and then we can talk about how that compares to what s going on in DLINQ or LINQ to SQL. So, because we have this transformation to the conceptual model, and then we have the ability to have objects that sit on top of that, it s very natural then to take these enumerations of objects. They are my Logical Entities that I can query from the database and think about wanting to do LINQ queries on top of them so you get strong typing and those kinds of things. But the big challenge of course is that you want those LINQ queries to be translated into backend queries rather than get back all my data and then filter it in memory. So, that s a key concept in LINQ of being able to define an interface that s IQueryable, and get a handle on that specification rather than the compiler actually compiling it. And then we do a runtime translation of that query specified into the same query format, the same query trees, and then it runs through all of the rest of our stack the same as if you had specified your query any other way. (00:29:51) In particular, one of the big innovations in the Entity Framework is the ability to say, not only do you have LINQ as an option, but you also, for those cases where you really have to have a dynamically generated query, we have a rich query language that we call Entity SQL -- it s very close to SQL, but it has extensions, so it understands the things in the Entity Data Model like Inheritance and Navigating Relationships and things like that. But it is a full representation query language. So it s got -- you can compose your queries, you can do joins, you can do very, very rich query capabilities - and then those get translated. So, you can do it sort of either way; you can either buy into LINQ, which for some solutions makes very great sense and for some solutions not as much sense, or you can go textual based query option. Richard Campbell: So, this is again a classic feature of the strength of the.net framework that if you need to get under the hood, you can. It s not going to be simple; you got to work hard to do it, but it s worth it, because it means there are really no roadblocks. I think the big thing that people fear in the whole ORM concept is that you re going to abstract me away from my data, Transcription by PWOP Productions, Page 8 of 17

9 and sooner or later you re going to burn me for that, there is going to be no way. Carl Franklin: You are going to black box me. Daniel Simmons: Absolutely. Being the ADO.NET team, we have lots and lots of feedback from the customers who are using Relational Data Access today, and one of the things that we tried really hard in doing is -- I think is another way that the Entity Framework is very different from some other ORM efforts is that we re saying, it s very much in the spirit of ADO.NET. So it s a series of components for the public and there s a architecture, that you can look at all the different pieces and you can buy into as much or as little as you want. And it does everything according to ADO.NET principles; so, for example, we never make a query unless you know when it s going to happen; they re very explicit; we never talk to the database unless it s explicit. If you want to build something that does it automatically and then puts it on top, we provide the hooks, so that you can do that, but you have that kind of control to be able to optimize your application and go on... Richard Campbell: You don t want to break that cardinal rule of Don t do stuff I didn t expect - and tell me when you are going to do something. Daniel Simmons: Exactly. And this is a problem that most ORMs struggle with, because they re trying so hard to give this transparent experience to the business logic developers that there s no database here, and I just -- when I go to my customer, I ve got a collection of orders, and I just go enumerate it, and magically their orders are all here. But at the same time, then that introduces all of these unexpected behaviors into your application, because it s very difficult to get the performance and all of the characteristics correct. Richard Campbell: You mentioned like the business logic to your -- I can see that this is not a replacement for CSLA.NET, Rocky Lhotka s Framework -- that it s really something that would work underneath that framework, as business objects are a different creature from Entities, and these are only one small part of the equation. Daniel Simmons: Absolutely. I think for some class of applications, you might say that I am going to make my Entities and my business objects be the same. And I think for the first time, we begin to have enough abstraction from the database and enough control that you could actually entertain that for some real projects, but clearly, there are projects that go beyond that, and you re going to say, no, absolutely, I want to have a separate layer. Carl Franklin: Have you ever felt envy for the new slick interfaces introduced in Windows Vista? I am sure you want to have something similar in your apps, but unfortunately, that s quite hard to achieve with Windows Forms. There is WPF of course, but that requires you to adopt a whole new programming model. Wouldn t it be nice if you could have scaling, rotations, animations, alpha blending, complex gradients, and all that in classic Windows Forms? How cool would your application be then? Well, it s going to stand out - and it s definitely going to look nice. Stop envying and start delivering great experiences today. telerik r.a.d. control suite for WinForms offers the first Vista-style controls for Windows Forms. Pick a Vista piece of UI and try to implement it with the telerik controls; chances are that you can do it. Join the telerik WinForms challenge today to explore the controls in a fun and engaging way. The challenge is a mini-quiz that shows off the unique features of the Controls. In just minutes you can see how you can make your desktop apps much more appealing, and you can win a product license by simply answering five questions correctly. And everybody who completes the challenge is automatically entered into the drawing for the grand prize - get this, a 50 Plasma TV. Check out telerik r.a.d. controls for WinForms and join the WinForms challenge today at So, what kinds of projects would this be applicable to, to have your business logic and your Entities one and the same? Like, where would that be a good idea? (00:34:55) Daniel Simmons: Well, I think it has to do with how complicated your business logic is. I think we re striving for the world where you can have the flexibility that you need with as fewer layers as possible. So, once we have the ability to transform from my underlying Relational model into my Logical Conceptual model, and even have the ability to after the fact, tune my database and change my mapping without having to change my conceptual model. Suddenly I get the opportunity then to say, well, if I put my business logic here, I don t have to worry about having to refactor all of that business logic, because I have to change my database to add another Denormalization to get the perf right or something. So, I would suggest that for a great many applications you might find that hey, I can actually write the business logic that knows to work in my domain, directly in my Entity Objects, and I ve gotten a big win in reducing complexity. That said, there are also times where you say, Hey, my application has -- even though I have Transcription by PWOP Productions, Page 9 of 17

10 Conceptual Entities, I have two ways of looking at the world; I ve got the way, when I am writing the Rich Client or the Mid-Tier, and then I ve got the way when I remote it because I need to deal with bigger packets at once. Carl Franklin: Right. Logic is going to change. Richard Campbell: One element that business objects definitely lend themselves to that Entities also do is this idea of encapsulating the larger picture of a block of data that s related. But you can also see where business objects go in different directions; and the first thing that hit me was stuff like Workflow that if you re using this set of Entities then you must involve one of X other Entities based on the data you re working with -- and that just gets out of the scope of Entities pretty quickly. Carl Franklin: Well, that s something that we haven't really defined. What is the scope of an Entity as opposed to a Business Object? Daniel Simmons: Well, that s kind of interesting. I think this example that Richard was giving is a pretty good way to think about this. The Entity becomes a logical unit that is basically persistent. So, it s a customer, it s an order, it s a Carl Franklin: Some data class. Daniel Simmons: some data class, but it can have, because it s an object now, and because it merges all of the pieces that go together to make up that Entity for real, you can actually write business logic in there, and you can add operations and behaviors to them. I can have on my order object that comes straight from the database, the ability to submit it to some next step, but then I might have a Richard Campbell: Or do the tax calculation. Carl Franklin: So, really, when we say Entity, we just mean a persistable object, whether that object is a business object or not. Is that a good way to describe it? Daniel Simmons: Yeah, I think it is a good way to describe it. And in particular, in the case of the Entity Framework in Orcas, we re talking about ones that you re persisting to the database. In future releases, it might be -- that definition may even more generally be true, because you might say, hey, we have other ways to take Entity Data Model and persist them. Maybe it s that I transform it to XML, and I send it to merge to something else. Carl Franklin: Yeah, I was just going to say, you transform them to XML, and then you can persist them on the disk. Now, you re sort of getting into like serialization stuff, right? I mean, you re sort of getting into that murky area. Daniel Simmons: Well, absolutely. One of the interesting things about serialization is that you could say, well I might build all of my objects, and then serialize them; and for some set of applications that makes sense, because I need to run business logic before I serialize to decide if I am going to do it or something like that, but I might have some other scenarios where I say, you know what, all I want to do is get these logical things out of the database, and get them on the wire as fast as possible, and then something like the Entity Provider going direct ADO.NET, and not ever building objects could be dramatically more preferment. Richard Campbell: I could also see it is a great stubbing solution. Why wait for the database to finish being built, when I could define these Entities and what they look like ahead of time, and work against those. Carl Franklin: Sort of like Mock Objects? Daniel Simmons: Well, in a sense you could say that. Carl Franklin: They re mocking the database, not the logic though, that s the difference. Daniel Simmons: Well, yeah. People do this today when they just go build their domain objects, but one of the interesting things that happens when you introduce a conceptual model is that you get extra constrains that parts of the framework can enforce for you. So, for example, you can define a relationship between two of your Entities, and you can specify that the multiplicity of one end of the relationship can be many, but the other one must be one, and then the framework can provide Classes for you for the Collections and References that automatically enforce and do fix up so that when I add a new Order to my Customer s Orders Collection, that Order automatically gets a reference back to the customer, and you don t have to write any code to do that. So, you can get way down the road of getting your business logic and enforcing the same constraints that your database is going to enforce in a completely declarative fashion. (00:40:08) Richard Campbell: Yeah, more and more declarative structures. Just Rocky -- I know you re listening out there -- I haven't forgotten about you. I think there s still a distinction here in CSLA.NET that I think what Entities is bringing to the table now is that those very simple apps, the Transcription by PWOP Productions, Page 10 of 17

11 simpler class of apps that were built in CSLA.NET, because that was all we had, may well be switching to the Entity Framework. But there s a level of sophistication here where CSLA.NET is going to live happily; it is another layer of abstraction that, if both exist, if Entity objects and CSLA both exist in the same project, that CSLA would be the larger level of abstraction over that. Daniel Simmons: Absolutely, and you can certainly imagine -- and one of the things I did about this design point was, bind into as much as makes sense. You might say, you know what, I don t even want the object services layer from the Entity framework because I already have CSLA.NET, but I do want this transformation, this ability to do mapping to a conceptual model underneath. Richard Campbell: I know you re not here for this version, so I am going to commit you to the feature, but the prospect that I d be able to do those Declarations in the Entity Framework, and then hand it to Replication and say, This is a unit of work; make sure that gets replicated, that s a lot of code I didn t have to write. Carl Franklin: Sure. Daniel Simmons: Absolutely, and that s really where the vision is, and that s why in this release we re saying, hey, let s get the mapping and the definition of this Entity Data Model really nailed down. So, we have this foundation, so that in future releases we can start delivering just a whole realm of services on these that you just get automatically. You build your objects, and you do a little bit of information to let us know what the model was, maybe from just the definitions of your objects, or maybe because you opted them separately, and then you get to leverage that from a whole bunch of layers. Carl Franklin: Is there any way that you can hook into or maybe you already have or are planning to hook into the White Horse Class Designer? I called it White Horse because that was the code name - but the Class Designer in Visual Studio. Daniel Simmons: So, right now we re not so much hooking into the Class Designer because we re thinking more in terms of, there s at least some set of customers that wants to go design their models first, and not necessarily even thinking about Classes, sort of into a separate exercise of moving to Classes. So, I think the first goal for us is to say, Let s have a first level Entity Data Model designer, but over time there s lot of opportunities for tools, and going into the Class Designer and having a way to just sort of decorate and say, yeah, this hooks up to these Entities... Carl Franklin: Or have it reverse-engineer the way it does already with Classes; to have it represent those things as little things dangling off the Classes that will bring up the Entity Designer. Daniel Simmons: Absolutely. This is a very interesting point about generation; one of the bits of feedback we get from folks that s right on the money is, Hey the kind of automatic generation we have now, goes from the database up to the model, but there s this other kind that a lot of people want absolutely, that is, going from the Objects down, and wouldn t it be nice if you could, from your object, say, Generate me a database or Hey, I added this Property to my Object, and would you please not just generate the Database but generate the modification script that would add the column to the Database, and then if I decided to roll it back, generate a second script that will pull it out. Carl Franklin: Let me give you my thoughts on that, because I have talked to quite a few people about this; there are all sorts of tools that have done this throughout time, code generation tools that are sort of modeled down. What most people do, especially the ones using UML, because UML is now just like in the realm of whiteboards; you design something, maybe you print it out to look at it, and then you go write your code. There isn t any way to directly -- rational for Java, okay, but for.net there s no Events or anything like that. So, they have the Class Designer. But I don t see a lot of people using it, just because after a while, something will happen and it will sort of just blow up. It just seems like it s very hard for these designers to just go all the way to allowing me to just work with them throughout the whole lifecycle of the application development. Daniel Simmons: I agree. I think one of the interesting stories is the one that says, I am not trying to generate my code from my model, but what I am trying to do is, I am writing my code, and then maybe putting a little bit of hints in it, so that we can generate the model out of the code. (00:44:56) Carl Franklin: Yeah, that s a very useful thing to do; let me take a bird s eye view of this code and see what it looks like. Daniel Simmons: Absolutely. Richard Campbell: And I do think the key to all of this or sort of the requirement, if anybody is going to go down this path is this nature of bidirectionality that if I tweak the model -- you can Transcription by PWOP Productions, Page 11 of 17

12 tweak my code but don t mangle it, and if I modify my code it needs to be reflected in the model. Daniel Simmons: Right. Carl Franklin: Absolutely. Daniel Simmons: So, I would say, in Orcas we re getting sort of some of this direction; we re getting the story that says like many of the Visual Studio designers, hey, there s this way you can describe the model, and then we ll do Codegen of partial Classes, and if you put your data in a separate side of that partial class, everything works. But you also have the option if you don t want to do that of writing your classes from scratch, implementing a few interfaces that allow us to be aware of what s going on, and then writing the model by hand. In a later release, we want to be able to say, okay let s go to the next step, which is to say, you just write your classes - - I mean you tell us to build the model for you either one time or in a make or meddle fashion, because that s -- the truth is in the Classes. We clearly have some set of customers who say the truth is in the model, because I am a big corporation and we have a backend IT group, and they have written a model, and then they have different departments that write apps against that model. So, that s one thing we want to support, but we also want to support the group of folks that says, hey, the Objects are the truth, and just always generate the model from there, and then we can work down. Richard Campbell: The code is always the truth guys; it is the manifestation of it one way or the other; you may be able to cast the truth down to us from on high, and let us know if we match, but what we ultimately wrote is the truth even if you didn t like it. Daniel Simmons: Of course. I am surprised to hear you saying that Richard being the database guy, because there s another camp who would say, no, no, no, no! Richard Campbell: No. The data is the truth. Daniel Simmons: The truth is what s on disk. Richard Campbell: Yes. What I stored is what matters. I am an IT as a service kind of guy. In the end, I don t create the data, I don t really let you know what you ve got, and that means that the guys who create the data are really the guys who matter in the end. It s important that we let them know we captured what was important, but you have to deal with the realities as well. Carl Franklin: Also, instead of having the database, you have a family and children, and other intelligent Entities to look after unlike some DBAs that all their life is the database - Hey, get out of there; put that down; stop slapping your brother. Richard Campbell: Well, at least we wash our hair. Carl Franklin: True. Now we re going to get hate mail from DBAs all over the world. Richard Campbell: I don t wash my hair. Daniel Simmons: One of the scary possibilities that we re entertaining and/or maybe exciting is -- depending on how you want to think about it, is that if you take this Entity Framework vision farther and farther forward, you eventually say, well, why did I need to specify the Relational model when I wrote my database at all. Why didn t I hand my Entity Data Model directly to SQL and say, Store this; and do it fast. Richard Campbell: That s a very developer thing to say. Daniel Simmons: Interestingly enough, from a SQL Server implementer standpoint, there s good and bad. Absolutely, there s a good reason why people go and tune their Relational models and things like that, but there are certain kinds of tuning we could do in the database if we didn t have to honor the Relational model sometimes. Richard Campbell: Yeah, and the trick is of course popping that up and saying, I could make this faster if you are willing to relax X rule. Daniel Simmons: Exactly. Richard Campbell: And believe me, I really dig the idea -- just putting on my DBA hat for a moment - if you handed me a SQL Script that was generated from ADO.NET Entity Framework or more importantly the client tool that sort of wraps over the top and said, hey, we had to add this new Property, and the framework thinks we should stick the column here, here is the script to run it. You ve now narrowed down my effort to add that column a lot. You said, this is where we think it ought to go, and put it in a coherent form. I may not be totally happy with it, but you ve also cut a big chunk of work out of my effort to implement what you need. Daniel Simmons: Sure. Well, and part of the point is, imagine a world even farther in the future where I said, hey, every application I ever do with this data, always goes through this conceptual model, because my reporting and replication, and my business logic and everything uses it. So, I am going to tell you DBA, this is what I want, and Transcription by PWOP Productions, Page 12 of 17