Quote these dates to prevent intermittent test failure. Suppose local time is 00:50 GMT+1. Without the quoting, the YAML parser would parse this as 00:50 UTC, into the local time of 01:50 GMT+1. Then, it would get written into the database in local time as 01:50. When it came back out the UTC date from the database and the UTC date of two weeks ago would be compared. The former would be 23:50, and the latter would be 00:50, so the two dates would differ, causing the assertion to fail. Quoting it prevents the YAML parser from getting involved.

This reverts commit 9d396ee8, reversing
changes made to fa2bfd83.

Reason: the change broke the build.

Add a proxy_association method to association proxies, which can be called by association extensions to access information about the association. This replaces proxy_owner etc with proxy_association.owner.

This reverts commit 04cc446d.

I reverted it because apparently we want to use: select([:a, :b])
instead of select(:a, :b), but there was no tests for that form.

Simplify the test by using id and name. `id` will be the only real sort criteria in any case as it's unique.

We were experiencing CI test failures, for example:

* 3-1-stable: http://travis-ci.org/#!/rails/rails/builds/79473/L407
* master: http://travis-ci.org/#!/rails/rails/builds/79507/L80

These failures only happened on 1.8.7-p352, and we were only able to
reproduce on the Travis CI VM worker. We even tried creating a new 32
bit Ubuntu VM and running the tests on that, and it all worked fine.

After some epic trial and error, we discovered that replacing the
following:

  fuu = Marshal.load(Marshal.dump(fuu))

with:

  marshalled = Marshal.dump(fuu)
  fuu = Marshal.load(marshalled)

seemed to prevent the failure.

We have NO IDEA why this is. If anyone has some great insight to
contribute then that is welcome. Otherwise, hopefully this will just
help us get the CI green again.

Many thanks to @joshk for help with sorting this out.

We have to use Observer#update rather than Observer#send since the enabled state is checked in #update before forwarding the method call on.

Make it the responsibility of the connection to hold onto an ARel visitor for generating SQL. This improves the code architecture generally, and solves some problems with marshalling. Adapter authors please take note: you now need to define an Adapter.visitor_for method, but it degrades gracefully with a deprecation warning for now.

On Oracle disable_referential_integrity before execution of block will disable foreign key constraints and after block will enable them but when constraints are enabled then they are validated. Therefore created record with invalid foreign key should be deleted before enabling foreign key constraints.

This reverts commit c5448721.

Quote these dates to prevent intermittent test failure. Suppose local time is 00:50 GMT+1. Without the quoting, the YAML parser would parse this as 00:50 UTC, into the local time of 01:50 GMT+1. Then, it would get written into the database in local time as 01:50. When it came back out the UTC date from the database and the UTC date of two weeks ago would be compared. The former would be 23:50, and the latter would be 00:50, so the two dates would differ, causing the assertion to fail. Quoting it prevents the YAML parser from getting involved.

This reverts commit 9d396ee8, reversing
changes made to fa2bfd83.

Reason: the change broke the build.

Add a proxy_association method to association proxies, which can be called by association extensions to access information about the association. This replaces proxy_owner etc with proxy_association.owner.

This reverts commit 04cc446d.

I reverted it because apparently we want to use: select([:a, :b])
instead of select(:a, :b), but there was no tests for that form.