We need to do something to ensure view that are generated on one platform won't be considered 'out-of-date' on a platform that gives a different reflection order for properties (and therefore a different column order in SSDL).
We originally did alphabetical sorting of columns in Code First which avoided this issue... but we got a **LOT** of feedback that column order should match the order that properties are defined in the class. Reflection order doesn't guarantee this but we aren't seeing any complaints about ordering with the current implementation.
We've come up with two ways to fix this:
* Just use alphabetical sorting in SSDL but add an attribute with desired ordering in the database we create so that columns in the database are generated in reflection/configured order.
* Improve the view hashing algorithm to ignore column ordering (this is part of some larger view hashing improvements we are tracking - http://entityframework.codeplex.com/workitem/606)
Comments: We tracked it down to properties in a second partial class that are not processed equally on all our machines. It seems as if sometimes the properties in the second partial class are discovered before and sometimes after the properties in the "main" partial class. HTH
We originally did alphabetical sorting of columns in Code First which avoided this issue... but we got a **LOT** of feedback that column order should match the order that properties are defined in the class. Reflection order doesn't guarantee this but we aren't seeing any complaints about ordering with the current implementation.
We've come up with two ways to fix this:
* Just use alphabetical sorting in SSDL but add an attribute with desired ordering in the database we create so that columns in the database are generated in reflection/configured order.
* Improve the view hashing algorithm to ignore column ordering (this is part of some larger view hashing improvements we are tracking - http://entityframework.codeplex.com/workitem/606)
Comments: We tracked it down to properties in a second partial class that are not processed equally on all our machines. It seems as if sometimes the properties in the second partial class are discovered before and sometimes after the properties in the "main" partial class. HTH