I understand that using Transformable data type is an easy way to store array or any custom object in Core Data. I would like to know when one should not use Transformable but should instead create another entity and use To-many relationship.
Say if it is an array of strings, is there a maximum number of elements or maximum length of the string that would cause significant performance problem?
I would like to know when one should not use Transformable but should instead create another entity and use To-many relationship.
You should only use transformable attributes when you absolutely have to. They are not a convenience or a short-cut but a resource intensive necessity in some circumstances.
You seldom use Core Data to store data structures like arrays or dictionaries because Core Data is primarily used for not for storage/persistence but rather for modeling/simulating. It's kind of useless for modeling data to turn an data structure into a big, logic-less blog of data.
Transformable attributes are usually used to store some class that itself actively manages the data it holds e.g. transforming a UIImage so that you can take an UIImage straight from the UI store and get it back all in one piece.
In answer to your main question:
I am more curious about how big the data can be until it affects performance and it's better to normalize them.
It depends mostly on the combination of size and complexity. Whenever you transform a bunch of existing objects into a data blob, you have to read the entire blog in by transforming. So if you store a 1mb array by transformation you get a 1mb array back in memory when run the reverse transform. Each transform, no matter how small, takes more processing time than accessing a normal attribute or even finding another managed object. Therefore, having a large number of small transforms that are often accessed will also cause quite a performance hit.
It is always better to decompose big lumps of data into entities, attributes and relationships. Doing so gives you all the flexibility and optimizations of Core Data for free. I find myself using Core Data in place of arrays and dictionaries because once you really wrap your head around Core Data, its just easier to use.
I would never use Core Data to store a transformed array of strings or the like. If the strings have no logic and there is just few dozen of them, you might as well write the array out to plist file. It will be quicker and easier than messing around with a transformable attribute.
It really comes down to how you want to use the data described by the transformable attribute.
Core Data will create a blob column in your SQLite database under the hood, so it will be nearly impossible to do any kind of real search or sort on the attribute.
Conversely describing the data you want to store with a logical entity (via a to-many relationship) will allow you to utilize search and sort functionality.
The entity has one other key advantage which is that it can be lazily loaded, with a transformable attribute your application will eat the cost of loading that data each time the object faults. If you're encoding a set of objects as a transformable attribute this could result in a significant performance issue as you ready the data off disk, even when you don't need it.
As far as size goes, you can keep as much data as you wish in the blob, it's unbounded. This again gets down to the performance impacts of such a design, where (especially on iOS devices) I/O is very expensive and memory is constrained your application may simply be unable to read the data out of the store and in to memory (people have tried sticking movies, in CoreData SQLite stores).
