2012/1/20 Skogen, Espen <[email protected]>: > I disagree - I believe 1, 2, and 3 is handled adequately in the existing > module implementation. > Multi SDK versioning is also no problem as long as modules are compiled with > external SDK dependencies, and that the parent app is always of a later > release then the module.
Even if I used modules in a lot of applications, primarly to allow incremental loading of functions in large application, I think they are too limited. I don't like very much the current implementation because, AFAIK, a module has to be a visual element on the stage.. instead I'm using the term "module" here to include a set of compiled class definitions, visual or non-visual, that can be loaded and unloaded on demand, and the used to create object instances.. in short, any SWF. Let's call it "bundle", if you wish. Moreover, as you point out too, modules are not so easy to unload due to a number of quircks involving CSS style declarations, embedded fonts, registered remote aliases, etc. so point 3 is not also not adequately handled IMO. > Where this comes short however, is when module A has a dependency on library > c, and module B also has a dependency c - In the current scenario, if the > parent app do not have library c in its application domain, we will need to > merge the library in for both modules, leading to inefficient loading, and > potentially conflicts if module A and B needs to communicate and/or share > application domain. > This is where the OSGI registry becomes relevant. I'd like to consider "modules" and "libraries" as assets of the same type: just give me a clean, low level tool to load bytecode at runtime, and give me the chance to avoid locking that bytecode from being unloaded due to Flash Player or Flex internal behavior. But let me choose the best way to describe, handle and resolve the dependencies among all those modules (by hand, or using a framework). The OSGI approach could be one of the possible abstractions that solves this problem. > There's a number of an issue around unloading of modules - Mainly to do with > the Style chain, I believe, so it wouldn't hurt to look through the > StyleManager implementation to ensure that styles don't stop modules from > unloading. > Another bugbear of mine is the use of StyleManager2 in flex4. I'm not sure > what the best implementation would be, but the existence of StyleManager and > StyleManager2 is somewhat smelly in my mind. > > E > > > > Espen Skogen | Vice President | IB Tech Market | Investment Bank | J.P. > Morgan | 125 London Wall, EC2Y 5AJ, London, United Kingdom | T: > +442077420836 | [email protected] | jpmorgan.com > > > -----Original Message----- > From: Cosma Colanicchia [mailto:[email protected]] > Sent: 20 January 2012 13:12 > To: [email protected] > Subject: Re: OSGi bundle Plugin registry system > > I think we should first identify which low-level requirements that any > module system may want to address. > > I see the following ones, in ascending order of complexity: > > 1) bootstrap-time incremental loading of modules > 2) late and on-demand incremental loading of modules > 3) support for on-demand unloading of the modules > 4) support for sdk multi-versioning (modules compiled against > different SDK revisions) > > I think that the current RSL features only address requirement 1. > > AFAIK, many of the existing frameworks (e.g potomac) allow to go up to > requirement 2, and in this level we usually find the need for > additional high-level facilities such as registries, IoC and > dependency management features. However, my personal take is that > those facilities should not be embedded in the SDK but left open to > the application frameworks: I'd prefer to have only the low-level > handles in the SDK and in the Player to let these frameworks implement > their abstraction. > > Going to level 3 is currently not simple, partially because of > limitations at the Flash Player level (e.g. the remote aliases > registry managed at the SecurityDomain level, thus potentially related > to multiple ApplicationDomains with conflicting class definitions, or > the impossibility to remove entries of this registry once added). > > Level 4 is, on the other hand, pretty tied to the Flex SDK > implementation (it was the scope of the marshall plan). > > > I think that a discussion about the scope of any modules support > system could help us to better define the problem. > > > Cosma > > > 2012/1/19 Skogen, Espen <[email protected]>: >> In simple terms, the benefits would be this: >> >> Your main application loads a number of modules provided by a number of dev >> teams. >> Each module has a number of dependencies, and you do not know what those >> dependencies are at compile time of the parent app, so RSL is not an option. >> >> The registry allows loading of the dependencies of the modules, adds them to >> the application domain, and keeps a record of the dependency. When the >> second module depending on the same dependency gets loaded, there's no need >> to load the dependencies, because they already exist in the app domain. >> >> So in effect, it allows an on-demand rsl-like dependency loading behaviour >> which is supported on a module level instead of just at the application >> level. >> >> There are however issues around memory management and unloading which needs >> to be addressed, so although theoretically it sounds quite simple; there's a >> number of implementation issues that needs some thought. Haven't spent much >> time on this, but it would definitely be worth investigating. >> >> E:) >> >> >> Espen Skogen | Vice President | IB Tech Market | Investment Bank | J.P. >> Morgan | 125 London Wall, EC2Y 5AJ, London, United Kingdom | T: >> +442077420836 | [email protected] | jpmorgan.com >> >> >> -----Original Message----- >> From: Web DoubleFx [mailto:[email protected]] >> Sent: 18 January 2012 17:49 >> To: [email protected] >> Subject: RE: OSGi bundle Plugin registry system >> >> >>> From: [email protected] >>> This is the approach I've used in the past as well. Gotta love Parsley >> >> Sorry, for my english BTW, but can you explained what the gain with a >> registry ? >> >> If we have a bootstrap, we'll loose the possibility to optimize RSL loading >> (like in Gravity), and I wonder, what can bring us a registry ? >> Frédéric Thomas >> This email is confidential and subject to important disclaimers and >> conditions including on offers for the purchase or sale of >> securities, accuracy and completeness of information, viruses, >> confidentiality, legal privilege, and legal entity disclaimers, >> available at http://www.jpmorgan.com/pages/disclosures/email.
