# AR: save! doesn't redo object save after child save fails **URL:** \ **Category:** Rails\ **Created:** [August 30, 2006, 11:41pm UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021 "2006-08-30T23:41:43Z")\ **Posts on this page:** 8\ **Page:** 1
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 30, 2006, 11:41pm UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/1 "2006-08-30T23:41:43Z")
I have two AR descendant classes, A and B, where A has\_many B’s. I have an action where I create a new (unsaved) A. It then renders to another page that in turn posts to an action that creates several B’s that are then associated to A (but not saved yet since A has not been saved). If I call A.save! and there is a problem saving one of the B’s, I redisplay my form and prompt the user to fix the data. Looking at the log file, I can see the apparently successful SQL call to save the row for A as well as the SQL error for B. (Note that throughout the process of fixing the B objects, the A object is not changed.) Once the data is fixed, the call to A.save! happens again, and there is no INSERT statement for table A. The save for B fails with a foreign key constraint failure (no A row to point to). There are no A rows in the database so this makes sense. Calling A.new\_record? before A.save! shows that the A object thinks it is not saved. However, no INSERT statement for A ever occurs after that first, successful save. I’m guessing that when the first save! call occurs, that the A object thinks it saved to the database even though it didn’t (because of the child row failures). Then, later, there is no attempt to re-save the A object and the B objects fail to save because there is no A object available. What’s confusing is that A returns “true” to new\_record? and then no INSERT occurs. I think I can get around this problem by doing A.save! before I add any B’s to it. But this seems like a bug. Anybody have any insight into the intricacies of save! with a has\_many relationship? Thanks, Wes ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 30, 2006, 11:51pm UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/2 "2006-08-30T23:51:47Z")
I ran a test and here’s what I saw. - 1st call to save!: The INSERT for table A succeeded because the INSERT for table B tries to use foreign key id 191 as the foreign key to A. The INSERT for table B fails. - 2nd call to save!: No INSERT statement is executed for A. The INSERT for table B tries to use foreign key id 191 as the foreign key to A. The INSERT for table B fails because there is no such A row with id 191. Not only is the INSERT not being retried for A, but the child rows are remembering the original A id and failing for that reason. WG ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 12:22am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/3 "2006-08-31T00:22:24Z")
On 8/30/06, Wes G. [removed\_email\_address@domain.invalid](mailto:removed_email_address@domain.invalid) wrote: > Not only is the INSERT not being retried for A, but the child rows are > remembering the original A id and failing for that reason. The database rollback on a failed save doesn’t rollback your object also. As far as it knows, it was inserted fine… jeremy ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 12:24am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/4 "2006-08-31T00:24:15Z")
So then is the correct thing to do to always start from scratch with a fresh object to save? Wes ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 12:58am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/5 "2006-08-31T00:58:31Z")
On 8/30/06, Wes G. [removed\_email\_address@domain.invalid](mailto:removed_email_address@domain.invalid) wrote: > So then is the correct thing to do to always start from scratch with a > fresh object to save? Yup - much easier to deal with. Another option is to check .valid? on parent and children before saving. jeremy ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 2:00am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/6 "2006-08-31T02:00:38Z")
On 8/30/06, Wes G. [removed\_email\_address@domain.invalid](mailto:removed_email_address@domain.invalid) wrote: > 1. But I believe the real problem is that I’m trying to build up my > object in two different actions that post from 2 different pages. So > I’m going to create a separate model object (not to be saved) to act as > a value object to help me build my real object in the 2nd step. > > Because I’m trying to build my object across multiple submits, I’m > saving it to the flash and trying to keep up with its state as errors > occur in various submits. And clearly, it isn’t working. An easy approach is to use a candidate record that you save as you go. Then you can use separate validation rules for the candidate and completed models - pretty convenient - and easily ‘complete’ the candidate when the wizard’s done. Quickly: Would Rails benefit from the use of form model objects - > objects whose purpose is to keep track of the state of a form? (I think > the answer is yes). Yes, but I think they would be mis- and over-used. Consider how far Rails has come without them. Should we always go down the path of one form ↔ one model object no > matter what (you may have many intermediate steps that you need to do in > many forms in order to build your object - how is that best > accomplished)? I like the candidate record approach. jeremy ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 1:29am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/7 "2006-08-31T01:29:13Z")
Jeremy, Thanks for the responses. There are a couple of issues here. 1. The database error is happening because the length of one of the columns is too long. Perhaps some validates\_length\_of is appropriate here. 2. But I believe the real problem is that I’m trying to build up my object in two different actions that post from 2 different pages. So I’m going to create a separate model object (not to be saved) to act as a value object to help me build my real object in the 2nd step. Because I’m trying to build my object across multiple submits, I’m saving it to the flash and trying to keep up with its state as errors occur in various submits. And clearly, it isn’t working. * * * This all has me thinking about the RESTful thing and I will post about that at a higher level soon. Quickly: Would Rails benefit from the use of form model objects - objects whose purpose is to keep track of the state of a form? (I think the answer is yes). Should we always go down the path of one form \<-\> one model object no matter what (you may have many intermediate steps that you need to do in many forms in order to build your object - how is that best accomplished)? Thanks again, Wes ---
**Author:** ![weyus](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/w/7ab992/32.png) [@weyus](https://www.ruby-forum.com/u/weyus)\ **Post date:** [August 31, 2006, 2:06am UTC](https://www.ruby-forum.com/t/ar-save-doesnt-redo-object-save-after-child-save-fails/67021/8 "2006-08-31T02:06:08Z")
Candidate record is a nice way to say it. I can see your point about the mis and over-use of form models - on the other hand, aren’t we just misusing regular models and/or views and/or controller methods when we use regular models to hold non-trivial form state? In my particular case, I have an upload file step and then a data transformation step, and it is totally valid to deal with two objects when that much work is going on. Wes