So, my first patch was crap. It was actually failing because of a number of typos. However, I have a second patch that both reveals the problem and now fixes it, I think. I'm dropping it here to get feedback on how it looks, mostly because the new second-run.t test I've added to TestApp::DatabaseBackedModels is kind of kludgey and wondered if anybody had a better way.
The new second-run.t test checks to see if a second run of Jifty after Jifty
has been bootstrapped will load the models correctly. It does not in the
current repository since collections aren't necessarily required before they
are used. To do this, I fork inside the test after including Jifty::Test to
bootstrap the database and run a second-run.pl child process with
essentially the same tests as bootstrap.t. In order to make the second run
work without bootstrapping, I've also built a custom subclass of Jifty::Test
called TestApp::DatabaseBackedModels::Test that overrides setup() to run
without the database setup and _testfile_to_dbname() so that the test will
use the same database as the parent process already bootstrapped.
I could probably eliminate the need for TestApp::DatabaseBackedModels::Test
with an environment flag similar to the JIFTY_FAST_TEST variable.
Cheers,
Andrew
On Nov 19, 2007 11:41 PM, Andrew Sterling Hanenkamp <[EMAIL PROTECTED]>
wrote:
> Jesse,
>
> I'm working on a problem I've discovered in the V-M branch. I've attached
> a patch that contains a new bootstrap.t test and modified Bootstrap.pm for
> TestApp::DatabaseBackedModels to reveal the basic issue. But here's the
> jist. The patch basically modifies the existing boostrapper to add an
> additional "Warehouse" model similar to these declared models:
>
> package Warehouse;
> schema {
> column location => type is 'text';
> column widgets => references WidgetCollection by 'warehouse';
> };
>
> package Widget;
> schema {
> column name => type is 'text';
> column inventory => type is 'int';
> column warehouse => references Warehouse;
> };
>
> This code when used works fine until you try to call the "widgets" method
> of a "Warehouse" object. When you do that you get:
>
> WARN - DBD::SQLite::db prepare failed: no such column: widgets(1) at
> dbdimp.c line 271 at /Library/Perl/5.8.6/Jifty/DBI/Handle.pm line 474.
> WARN - TestApp::DatabaseBackedModels::Handle=HASH(0x1f69ea8) couldn't
> prepare the query 'SELECT widgets FROM warehouses WHERE id = ?'no such
> column: widgets(1) at dbdimp.c line 271
>
> As you can see the anonymous subroutine used to define "widgets" is
> incorrectly using a call to __value() to fetch the value rather than using
> _collection_value() as it should. I've not pinned this all the way down yet,
> but I'm speculating that the issue is that the "widgets" column is being
> defined prior to "WidgetCollection" becoming available inside
> Jifty::ClassLoader::require_classes_from_database().
>
> What I don't know is how to fix it, if this is indeed the case. I'm
> currently working my way through Jifty::DBI::Schema to understand how the
> _init_column() method and related code works to see if I can bring the same
> or similar logic into the class loader, but I'd be grateful for some advice.
>
>
> Cheers,
> Andrew
>
databasebackedmodels-load-order-problem-fix.patch
Description: Binary data
_______________________________________________ jifty-devel mailing list [email protected] http://lists.jifty.org/cgi-bin/mailman/listinfo/jifty-devel
