X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=TODO;h=9295085e89d94c79a54132992102397068a68865;hb=ff0e97358b6a21b59d3826b6a5fe9d0a9486e004;hp=e96c161b52abe7670fecd866b3079b8c3587f723;hpb=79a3f1420ecfa48ebd86f01f1f43bc9cfc13c978;p=catagits%2FCatalyst-Runtime.git diff --git a/TODO b/TODO index e96c161..9295085 100644 --- a/TODO +++ b/TODO @@ -1,13 +1,166 @@ -Compatibility: - - Deal with component downwards compatibility in 5.71 (make 5.71 call - mooses construction logic so that attributes get initialized etc). - - Provide an extension so that components can have a Moose::Object - constructor and an @ISA which makes c3 happy on both 5.7X and 5.8X. +# Known Bugs: -Testing: + - Bug ->go or ->visit causes actions which have Args or CaptureArgs called + twice when called via ->go or ->visit. - - Run more smokes + Test app: http://github.com/bobtfish/catalyst-app-bug-go_chain/tree/master -Profiling: +# Compatibility warnings to add: + + - $self->config should warn as config should only ever be called as a + class method (TESTS). + +# Proposed functionality / feature additions: + +## Log setup needs to be less lame + +So Catalyst::Plugin::Log::* can die +in a fire. Having $c->log_class would be a good start. kane volunteered +to do some of this. + +Simple example: Catalyst::Plugin::Log::Colorful should just be a +subclass of Catalyst::Log, no ::Plugin:: needed. + +See also: Catalyst::Plugin::Log::Dispatch and +http://github.com/willert/catalyst-plugin-log4perl-simple/tree + +# REFACTORING + +## The horrible hack for plugin setup - replacing it: + + * Have a look at the Devel::REPL BEFORE_PLUGIN stuff + I wonder if what we need is that combined with plugins-as-roles + +## App / ctx split: + + NOTE - these are notes that t0m thought up after doing back compat for + catalyst_component_class, may be inaccurate, wrong or missing things + bug mst (at least) to correct before trying more than the first 2 + steps. Please knock yourself out on the first two however :) + + - Eliminate actions in MyApp from the main test suite + - Uncomment warning in C::C::register_action_methods, add tests it works + by mocking out the logging.. + - Remove MyApp @ISA controller (ask metaclass if it has attributes, and if + so you need back compat :/) + - Make Catalyst::Context, move the per request stuff in there, handles from + main app class to delegate + - Make an instance of the app class which is a global variable + - Make new instance of the context class, not the app class per-request + - Remove the components as class data, move to instance data on the app + class (you probably have to do this for _all_ the class data, good luck!) + - Make it possible for users to spin up different instances of the app class + (with different config etc each) + - Profit! (Things like changing the complete app config per vhost, i.e. + writing a config loader / app class role which dispatches per vhost to + differently configured apps is piss easy) + +## GSOC + +### Next large steps, planned: + +For all components that have been discovered, in whatever way, we create a service: + - that's a catalyst component service + - which is basically just a constructor injection + - except the constructor name is COMPONENT + - and we're "implicitly" passing along some constructor args + - Lifecycle => Singleton + + - We make a 'components' sub container in the main container. + - This gets the ConstructorInjection COMPONENT services, as model_Foo. + - Lifecycle of these services is Singleton + - This simplifies the code for MyApp->components, as it need only list one sub-container + + - We create a second service (depending on the first one) for ACCEPT_CONTEXT + - This has a custom service which calls ACCEPT_CONTEXT when the instance is fetched + - Not Singleton lifecycle + + Note - ACCEPT_CONTEXT can be called on app values - if you have a Model::Foo, with an ACCEPT_CONTEXT + and you call MyApp->model('Foo') then ACCEPT_CONTEXT gets invoked with a $c of 'MyApp' (this is not\ + the normal case, but we need to preserve for compat) + +### Next steps - less planned: + + - Creating service()-like sugar for component + + - Test cases for extending the container in an application. + - Using the sugar added in the previous item + - Test when Model::Foo depends_on Model::Bar + + a) configure additional services in that container + - super simple container $default_container => as { more services }; + class MyApp::Container extends Catalyst::Container { + use Bread::Board; # Or our own sugar? + method BUILD { container $self => as { + service model => ...; # some constructor injection to MyApp::Model or something + container Model => as { + component Foo => (dependencies => ['/model']); # As per default! + component Bar => (dependencies => ['/model/Foo']); # Magic! + }; + # Note - implementation of BB may need to be changed to support making sure existing + # services actually get overridden. not sure how the default container behaves when doing that + # above code would build the constructor injection as it currently does, + # defaulting to the class name in the right namespace as declared by the surrounding container + # as well as adding using the catalyst-specific service class + } } + }; + + let's consider the usage patterns we actually want to enable by doing the whole B::B thing + what happens if i make the "per-app" service for a component life only for the duration of the request? + or be instanciated every time i look up the component? + (or scoping it per session, or having a block injection, or something) + + say you override the app service to be per-request + now the wrapper for the per-request variant doesn't make sense anymore. does it? + because you're only overriding one half of what has been generated automatically + + ah, so you have basically ended up with getting a request scoped thing to be used to construct a request scoped thing, which is pointless? Would/could you not just override the + service which is actually getting looked up instead, and make it not depend on the auto-generated per-app scope service, which will then just never be built? + + yes, you could. but then you'd have to be aware of the distinction + which is what i hoped to be a barely visible backcompat thing + but which i'm afraid it won't be if we go for two actual separate services + + what stops the sugar we give from not just making you specify the lifecycle, and giving you the obvious name / wiring? + i.e. everything looks like 'Foo', so you don't have to know COMPONENT/Foo exists + + my hopes of not needing any sugar at all, i guess + + only a couple of new lifecycles to be registered with the container - ?? + + +### To polish off / t0m review + + - + $class->container->get_sub_container('model')->make_single_default; + + $class->container->get_sub_container('view')->make_single_default; + + get_components_names_types + + locate_components + + +# FIXME - t0m, how do you feel about this name? + +# also, do you think I should draw it here, or just return the data structure? + +sub get_components_names_types { + + + MyApp->config->{ 'Plugin::ConfigLoader' }->{ substitutions } = { + + +# FIXME - just till I understand how it's supposed to be done + +# Made this so that COMPONENT is executed once, + +# and ACCEPT_CONTEXT every call. + +has instance => ( + + is => 'rw', + + # This is ok?? + +my $applevel_config = TestAppContainer->container->resolve(service => 'config')->{applevel_config}; + +__PACKAGE__->config(applevel_config => 'foo'); + + + accept_context_args - where does this come from? + +### Known issues + + - Broken $instance->expand_modules() in setup_components and figure + out later how to bring it back + + - expand_component_module - - vs 5.70 and optimisation as needed on perl 5.8 (5.10 is already faster!).