Treat undef like blessed objects for the purpose of looking for special handling

[jonathancast]

Suppose foo is an inflated timestamp column and $row->foo is set to a true DateTime (representing a date, say, 2016-06-23T16:42:34). In the present code, these two code snippets:

$row->set_inflated_column(foo => undef);
say 'The value is: '.$row->get_column('foo');
say 'The value is: '.$row->foo;

$row->set_inflated_columns({ foo => undef, });
say 'The value is: '.$row->get_column('foo');
say 'The value is: '.$row->foo;

would do different things. The first prints

The value is: 
The value is: 

the second prints

The value is: 
The value is: 2016-06-23T16:42:34

Because update uses set_inflated_columns, the same behavior is visible there.

This change makes set_inflated_columns call set_inflated_column when the column is inflated and new value is undef (which set_inflated_column handles correctly), which causes the two functions to behave identically in this case, which seems like the correct behavior.

[ribasushi]

Hi, and thank you for the report/patch!

A similar problem was reported couple weeks earlier, I haven't gotten around to looking at the other report yet: I am hoping to be able to do so in 2~3 days from now.

Your fix seems incorrect: it kinda sorta feels like you should be doing the cache invalidation at the right spot, instead of giving undef inflated status (which is likely to mess with other logic N steps removed). I can not put my finger on it yet, but then I haven't dug too deep.

Unrelated and regardless of the above: DBIC is at a stage of its life where single line fixes like the proposed can no longer be applied without an accompanying fails before / doesn't fail after test. While I am busy elsewhere, could you please put an addition to a file of your choice in t/icdt/* that will exercise the changeset in question?

Thank you!

[ribasushi]

@jonathancast yup, this is roughly what I was looking for, thanks! (there are a few snags wrt extra dependencies, but I will fix these myself, no need to worry about them).

I will get back to you on the fix really soon.

[jonathancast]

Added test that fails in the old version and passes with the changes