Updating only changed columns in ActiveRecord

Does anyone know of a way to get ActiveRecord to update only the
columns which have changed? For example, consider a table foos with
columns x,y,z. If I do

    f = Foo.find(123)
    f.update_attributes(:x => 99)

then I want AR to generate

    UPDATE foos SET x=99 WHERE id=123

rather than what it currently does:

    UPDATE foos SET x=99, y=2, z=3 WHERE id=123

(assuming y=2 and z=3 were the values read in by Foo.find)

Why do I want this? Well, I have tables which are updated
asynchronously due to external events, as well as by the user looking
at the record in Rails. Both the frontend and the event-handling
backend are using ActiveRecord. There are some fields which are
updated by the operator, and others which are updated by the backend.
But using ActiveRecord, there's a race condition between reading a row
and writing it back out again, which means that changes made by the
other party may be lost.

I searched for a plugin at
but couldn't find anything relevant.

So, the options I can think of are:

(1) Bypass ActiveRecord's 'save' and 'update' methods entirely

    f = Foo.find(123)
    ... compare f with params, determine that only 'x' has changed
    Foo.update_all(['x=?',params[:foo][:x]], ['id=?',123])

But this has to be done consistently everywhere. I guess I can write
my own version of update_attributes which does this.

(2) Read each record twice:

    f1 = Foo.find(123)
    ... compare f1 with params, determine that only 'x' has changed
    f2 = Foo.find(123, :select => 'x')

This seems like unnecessary overhead.

I had a quick look through the code path for 'update_attributes' but
it seems to be intimately bound to 'save', and 'save' seems to have no
concept of which attributes have changed.

Thanks for any suggestions.