Aaaaah, big missunderstanding here! I don't have a web project, I have a WPF rich client. I'm using the Session-per-screen concept (with my own UnitOfWork implementation which handles the session), so:
1. Screen is opened. 2. New NHibernate session is opened (FlushMode = Never). 3. Data is loaded to domain objects. 4. Data is passed to ViewModel objects. 5. User modifies ViewModel objects (through WPF data binding). 6. ViewModel objects check if the data is valid and if yes, modify the domain objects. 7. Screen wants to know if it should enable the save button, which results in a session.IsDirty() call (after some layers of abstraction). 8. User clicks save, which results in a session.Flush() call. 9. Session is closed. On 29 Okt., 13:07, John Davidson <jwdavid...@gmail.com> wrote: > I did not say that IsDirty should not be used for Domain objects. I did say > it should not be used for ViewModel objects. > > The proper sequence of events should demonstrate why IsDirty checking should > not be attempted with ViewModel objects. > > 1. Start a Web Page request > 2. Open a new NHibernate session > 3. Get Domain objects from DB (using NHibernate) > 4. Create ViewModel objects from Domain data (if NHibernate then hydrate via > projections) > 5. End NHibernate Session > 6. Display Web page > 7. Users interacts with Web Page and Starts New Web Request > 8. Open a new NHibernate Session (this session knows nothing about prior > session) > 9. Get Domain objects from DB (using NHibernate) > 10. Move ViewModel data to Domain objects (using business layer) > 11. Persist Domain model changes to DB (using NHibernate) > 12. Update ViewModel with business information > 13. End NHibernate Session > 14. Display Web Page to User. > > The ViewModel objects are not part of what NHibernate is managing, so it > makes no sense to try to have ISDirty checks from NHibernate on them. > IsDirty checks can be performed on the Domain Model objects and then handled > in the the business layer. > > If you are trying to have the ViewModel objects managed by NHibernate are > you trying to keep a Session open between Web Requests? > > John Davidson > > On Fri, Oct 29, 2010 at 1:29 AM, cremor <cre...@gmx.net> wrote: > > Neither is my View Model identical to my Domain Model, nor is the > > Domain Model simple. > > > So, if I understand your advices correctly, you are basically saying > > that IsDirty shouldn't be used to check if there are any changes in > > the domain entities, rather I should implement my own change tracking? > > Then I'll have to ask, what is the purpose of IsDirty? > > > On 29 Okt., 00:17, John Davidson <jwdavid...@gmail.com> wrote: > > > NHibernate is meant to manage the Object-Relational interface between > > your > > > Domain Model and your persistence layer (RDBMS). > > > > It is not used to manage the ViewModel interface with the Domain Model. > > Any > > > functionality that seems to support this interface is incidental and as > > you > > > are finding will not work the way you are hoping. There are samples where > > > the ViewModel and the Domain Model are identical, and I wish these > > samples > > > had never been created, because they only work correctly because of the > > > simplicity of the Domain Model, and everything falls apart when a more > > > complex Domain Model and ViewModel are attempted. > > > > I would seriously recommend you rethink your architecture. NHibernate is > > not > > > a panacea for poor design. > > > > John Davidson > > > > On Thu, Oct 28, 2010 at 11:43 AM, cremor <cre...@gmx.net> wrote: > > > > I am using ViewModel objects for the binding to my screens. But my > > > > ViewModel objects must write the data to the domain (NHibernate) > > > > objects, how would I use the business logic otherwise? > > > > Also, if I only write the data to the domain objects on save, I'd have > > > > to implement change tracking myself. NHibernate already implements > > > > change tracking, so why not use it? > > > > > On 28 Okt., 17:28, John Davidson <jwdavid...@gmail.com> wrote: > > > > > This is the reason for using a DTO rather than the NHibernate object. > > The > > > > > DTO data is only moved to the NHibernate object after the user has > > > > pressed > > > > > save. Screen objects should not be NHibernate objects. > > > > > > John Davidson > > > > > > On Thu, Oct 28, 2010 at 9:31 AM, cremor <cre...@gmx.net> wrote: > > > > > > Just for curiosity I switched my ID generator from Sequence to > > > > > > SeqHiLo. Calling session.IsDirty() now doesn't execute a select > > > > > > statement for the ID anymore (which is expected with SeqHiLo), but > > it > > > > > > still queues the insert statement. > > > > > > Then I even implemented my own IIdentifierGenerator which just > > returns > > > > > > an increasing number in Generate(), and IsDirty() STILL queues the > > > > > > wrong insert statement! > > > > > > > What's the reason for this? > > > > > > > I highly doubt that this behaviour is as designed. Because, in > > > > > > addition to the problem with not null or check constraints and the > > > > > > inconsistent behaviour I explained in my first post, this even > > results > > > > > > in the following behaviour, that is clearly a bug in NHibernate in > > my > > > > > > opinion (please tell me if I'm wrong): > > > > > > > Use-case: > > > > > > 1. Persistent Parent is loaded. > > > > > > 2. A new Child is added to the collection. > > > > > > 3. session.IsDirty() is called (UI checks, if it should enable the > > > > > > save button). > > > > > > 4. NHibernate queues an insert statement for the Child using its > > > > > > current (default) values. > > > > > > 5. User deletes the just created Child again (it is removed from > > the > > > > > > collection in the Parent). > > > > > > 6. User saves the changes, session.Flush() is called. > > > > > > > Now NHibernate does the following: > > > > > > 1. The queued insert statement for the new Child is executed (let's > > > > > > ignore the not null constraint for now). > > > > > > 2. Parent version column is updated. > > > > > > 3. The child (which shouldn't be in the DB) is commited. The next > > load > > > > > > of the Parent will return the Child which shouldn't be there. > > > > > > > I'd really appreciate any advice on this topic - or a NHibernate > > > > > > bugfix ;-) > > > > > > > On 7 Okt., 15:49, John Davidson <jwdavid...@gmail.com> wrote: > > > > > > > You probably need to do an id check yourself first, before the > > > > IsDirty. > > > > > > > > John Davidson > > > > > > > > On Thu, Oct 7, 2010 at 7:32 AM, cremor <cre...@gmx.net> wrote: > > > > > > > > Sadly, a different ID generator is no option for me. My > > application > > > > is > > > > > > > > not the only one accessing the DB and the other systems rely on > > > > these > > > > > > > > sequences. > > > > > > > > > What is the reason for getting an ID and flushing the Child > > when > > > > > > > > checking IsDirty()? Shouldn't it be enough for NHibernate that > > the > > > > ID > > > > > > > > of the Child has an unsaved-value? > > > > > > > > > On 7 Okt., 12:56, John Davidson <jwdavid...@gmail.com> wrote: > > > > > > > > > The use of sequence is similar to identity in that the db > > > > generates > > > > > > the > > > > > > > > id > > > > > > > > > stored rather than the other id generators which create the > > id > > > > within > > > > > > > > > NHibernate. For your problem the issue identified in jira > > NH-2136 > > > > is > > > > > > the > > > > > > > > > same. > > > > > > > > > > The solution is to use an NHibernate created id from hi-lo or > > > > guid > > > > > > based > > > > > > > > > generators. > > > > > > > > > > John Davidson > > > > > > > > > > On Thu, Oct 7, 2010 at 4:09 AM, cremor <cre...@gmx.net> > > wrote: > > > > > > > > > > I currently have the problem that NHibernate creates an > > insert > > > > > > > > > > statement for new items in a collection of a persistent > > entity > > > > when > > > > > > > > > > calling session.IsDirty(). > > > > > > > > > > > My configuration: > > > > > > > > > > * NHibernate 3.0.0.Alpha3 > > > > > > > > > > * An entity "Parent" with a collection of "Child" entities. > > > > One-to- > > > > > > > > > > many side is mapped as set, inverse, lazy and cascade > > > > all-delete- > > > > > > > > > > orphan. Many-to-one side is mapped as lazy. > > > > > > > > > > * All IDs are using the sequence generator. > > > > > > > > > > * Optimistic lock is by version column. > > > > > > > > > > * Session.FlushMode is Never. > > > > > > > > > > > My use-case: > > > > > > > > > > 1. Persistent Parent (already containing some Child > > entities) > > > > is > > > > > > > > > > loaded. > > > > > > > > > > 2. A new Child is added to the collection. > > > > > > > > > > 3. session.IsDirty() is called (UI checks, if it should > > enable > > > > the > > > > > > > > > > save button). > > > > > > > > > > 4. NHibernate executes a select statement for the ID of the > > new > > > > > > Child > > > > > > > > > > and queues an insert statement for the Child using its > > current > > > > > > > > > > (default) values. (Problem starts here: Why is it needed to > > get > > > > the > > > > > > ID > > > > > > > > > > of the new Child and queue an insert statement?) > > > > > > > > > > 4a. (Optional step) Some more new Childs are added. No more > > > > select > > > > > > > > > > statements are executed or insert statements are queued > > when > > > > > > calling > > > > > > > > > > session.IsDirty() because IsDirty() returns early because > > it > > > > knows > > > > > > it > > > > > > > > > > already has queued statements (very inconsistent behaviour > > in > > > > my > > > > > > > > > > opinion!) > > > > > > > > > > 5. User has to change a property of the Child to get rid of > > UI > > > > > > > > > > validation errors (a property is not allowed to be empty, > > but > > > > is by > > > > > > > > > > default to force the user to input something meaningful). > > > > > > > > > > 6. User saves the changes, session.Flush() is called. > > > > > > > > > > > Now NHibernate does the following: > > > > > > > > > > 1. Select statements for the new IDs of additional Childs > > from > > > > step > > > > > > 4a > > > > > > > > > > are executed. > > > > > > > > > > 2. The queued insert statement for the first new Child is > > > > executed. > > > > > > > > > > This statement uses the default values of the class and > > fails > > > > with > > > > > > an > > > > > > > > > > not null constraint error. > > > > > > > > > > 3. (If I disable the not null check in the DB) Parent > > version > > > > > > column > > > > > > > > > > is updated. > > > > > > > > > > 4. Additional childs from step 4a are inserted (with the > > > > already > > > > > > > > > > changed properties, so these statements are ok). > > > > > > > > > > 5. The first new Child is updated with the changed > > property. > > > > > > > > > > > I found the following Jira issue which already describes > > the > > > > > > problem, > > > > > > > > > > but it was rejected because it's "Expected behavior using > > > > > > identity": > > > > > > > > > >http://216.121.112.228/browse/NH-2136 > > > > > > > > > > But I'm using sequence, > > ... > > Erfahren Sie mehr ยป -- You received this message because you are subscribed to the Google Groups "nhusers" group. To post to this group, send email to nhus...@googlegroups.com. To unsubscribe from this group, send email to nhusers+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/nhusers?hl=en.
