Read-only fields.

rubyonrails-talk
1

I have a field in my table that I do not want ActiveRecord to update. I would like to retrieve that field, and have it displayed.

However, when I perform update, and ActiveRecord generates the SQL, I don't want it to include the field in it's update. I haven't figured out a way to do this.

Any ideas?

2

If the attribute is only updated under the hood, with triggers, Model.update_all, etc. then

   attr_readonly :foo

is what you need, but according to

   http://dev.rubyonrails.org/ticket/6896

is scheduled for 1.2.4 (I thought it was accepted for 1.2).

-- fxn

3

Dean Holdren wrote:

I have a field in my table that I do not want ActiveRecord to update. I would like to retrieve that field, and have it displayed.

However, when I perform update, and ActiveRecord generates the SQL, I don't want it to include the field in it's update. I haven't figured out a way to do this.

Why not write unit tests that check the value never changes, then write it liberally? Who cares if the magnetism actually hits that database file storage block, if the users never see the difference?

DBs were not made for the C++ 'const' concept...

4

Problem are race conditions:

   1. user = User.find(id)

   2. another process udates his record

   3. user.save

Now the field is overwritten with the previous value, so the record is broken. Even if you perform an extra read from the database in a before_save to get the value at that moment there's risk for a race condition.

In a muti-process environment there are race conditions if you don't lock resources, but if the data is updated always by actions the heuristic "last update wins" is normally good enough. When there are updates that only come from the database that doesn't apply.

-- fxn

5

Xavier: Most of those are solved with a lock_version field and optimistic locking. However, there are other conditions where a bulk update might not be ideal.

6

There are situations where you have an attribute that is _always_ maintained outside AR. There, optimistic locking is not what you need. To program that way you need to catch an exception, reload that special value, preserve the rest, and loop until eventually some save succeeds. That's cumbersome.

What you really need is a transparent solution that provides a regular model.save that ignores the field altogether. That's why there's a patch to add attr_readonly in the cue.

-- fxn

7

Yup, I agree 100%. That’s just one of the many “other conditions” I hinted at. I misinterpred your example (another process) to mean the same program, not an outside entity.