Re: isolate the pipeline component from rest of cocoon
Vadim Gritsenko skrev: Daniel Fagerstrom wrote: If we want to factor out the pipeline functionality to an own block that doesn't depend on the rest of Cocoon it will certainly be a lot of work. But I think it would be worthwhile anyway as it would increase usability and make the architecture easier to follow and maintain. I'm not that sure that pipelines should be extracted first. One would think that it is a sitemap engine that is logically sitting on top of pipelines. +-+ | Env Impl + Sitemap Servlet | +-+ | Sitemap Engine | +-++ | Environment API + Pipelines | Sitemap Components | +-++ | Container + Core Components | +--+ Agree about your view, not certain that there is much use in splitting up the sitemap stuff in sitemap engine and env impl + sitemap servlet though. We need names for the blocks, what about cocoon-sitemap, cocoon-pipeline, cocoon-core (or cocoon-container) and cocoon-sitemap-components? We should IMO strive towards a layered architecture where we have a sitemap (treeprocessor) block that depends on a pipeline block. Where the pipeline block contains the pipeline, pipeline components and the environment as well as numerous classes supporting them. Sounds good. Only point is that we should start with outermost layer first - extracting CocoonServlet first, following by sitemap engine, and only then pipeline block. Yes. Although I think we could split out the two upper layer in your picture at once and split them in a servlet and engine part later if needed. Next we need to plan the work so that it doesn't disturb the rest of the work to much, and so that it can be performed incrementally. I think that the Pipeline part is the largest part of Cocoon core so it would probably be simplest to move away the sitemap, container and sitemap component part from the core and keep the sitemap part. The sitemap components should be the simplest part to move out. Something that can be complicated and tedious to keep track on during the work is all the dependencies in the poms that depend on cocoon-core. One way would be to start with moving all code within cocoon-core to e.g. cocoon-pipeline and let cocoon-core depend on cocoon-pipeline, cocoon-sitemap, cocoon-sitemap, cocoon-container and cocoon-sitemap-components. Then we can narrow the dependencies when we have finished the work. If some part of the work need larger or more complicated refactorings, we probably need to do such work in a branch. At last we of course need to be a number of people that are prepared to work on it. I volunteer to do a part of the job. Taking a look at the pipeline code there are some dependencies on that would need to be removed. The ProcessingPipeline interface depends on the class o.a.c.sitemap.SitemapErrorHandler that in turn depend on treeprocessor stuff. It would be better to create an interface e.g. PipelineErrorHandler that the ProcessingPipeline depends on. +1 There is also a Processor.InternalPipeline descriptor that is used for the error handling within the pipeline implementations. It would IMO be better to move this internal class from the Processor interface to the components.pipeline package. That's new for 2.2, I don't see it in 2.1. No objections for the move here also. Both these changes affects interfaces, but I would be surprised if anyone have implemented Processor or ProcessingPipeline outside our code base, so it shouldn't matter that much. Since InternalPipelineDescription is 2.2 only, it was not released yet. SitemapErrorHandler can implement PipelineErrorHandler without affecting anybody. The AbstractProcessingPipeline gets its SourceResolver through EnvironmentHelper.getCurrentProcessor().getSourceResolver(), we need to use dependency injection for it instead to make the implementation easier to reuse outside the treeprocessor. // TODO (CZ) Get the processor set via IoC this.processor = EnvironmentHelper.getCurrentProcessor(); AbstractProcessingPipeline does not even use processor, only resolver. (Did not check derived classes though). OK, then we can start with this part of the work right away. /Daniel
Re: isolate the pipeline component from rest of cocoon
Vadim Gritsenko skrev: Renaud Bruyeron wrote: Therefore I would really like to be able to isolate the pipeline from the rest of cocoon, and be able to configure and instantiate it alone. I would like to do things like this: Generator producer = MyFactory.newGenerate(...); XMLConsumer consumer = MyFactory.newConsumer(...); pipeline.generate(producer, consumer); It is possible to use pipelines directly, it would look something like: ...setup container... ProcessingPipeline pipeline = new [Non]CachingProcessingPipeline(); pipeline.enableLogging(...); while (...) { pipeline.setup(new Parameters()); // Pipeline will get generator from container pipeline.setGenerator(file, ...); pipeline.addTransformer(xslt, ...); pipeline.setSerializer(html, ...); Environment env = new BackgroundEnvironment(...); pipeline.process(env); pipeline.recycle(); } I am willing to dig around in SVN, however it would help a lot with a few pointers. I could not find anything beyond 2.1.x on the wiki/official site. Cocoon 2.2 (spring based) is in trunk, Cocoon 2.1 (avalon based) is in branch. Do you have any pointers to things I should look at in SVN, or even if what I am asking for is remotely possible? It might be more complex than what you want but it's possible. If we want to factor out the pipeline functionality to an own block that doesn't depend on the rest of Cocoon it will certainly be a lot of work. But I think it would be worthwhile anyway as it would increase usability and make the architecture easier to follow and maintain. We should IMO strive towards a layered architecture where we have a sitemap (treeprocessor) block that depends on a pipeline block. Where the pipeline block contains the pipeline, pipeline components and the environment as well as numerous classes supporting them. Taking a look at the pipeline code there are some dependencies on that would need to be removed. The ProcessingPipeline interface depends on the class o.a.c.sitemap.SitemapErrorHandler that in turn depend on treeprocessor stuff. It would be better to create an interface e.g. PipelineErrorHandler that the ProcessingPipeline depends on. There is also a Processor.InternalPipeline descriptor that is used for the error handling within the pipeline implementations. It would IMO be better to move this internal class from the Processor interface to the components.pipeline package. Both these changes affects interfaces, but I would be surprised if anyone have implemented Processor or ProcessingPipeline outside our code base, so it shouldn't matter that much. The AbstractProcessingPipeline gets its SourceResolver through EnvironmentHelper.getCurrentProcessor().getSourceResolver(), we need to use dependency injection for it instead to make the implementation easier to reuse outside the treeprocessor. /Daniel
Re: isolate the pipeline component from rest of cocoon
On 22.10.2006 11:36, Daniel Fagerstrom wrote: Both these changes affects interfaces, but I would be surprised if anyone have implemented Processor or ProcessingPipeline outside our code base, so it shouldn't matter that much. Even if ... when to do it if not now? Jörg
Re: isolate the pipeline component from rest of cocoon
Renaud Bruyeron wrote: Vadim Gritsenko wrote: Renaud Bruyeron wrote: It is possible to use pipelines directly, it would look something like: ...setup container... Is that the avalon container, or is it a spring container? Is there an example of this somewhere? Avalon in 2.1, spring in 2.2. Avalon is setup somewhere in CocoonServlet/Cocoon objects; in 2.2, spring container is setup as usual + cocoon specific spring namespace handlers in spring config. ProcessingPipeline pipeline = new [Non]CachingProcessingPipeline(); pipeline.enableLogging(...); while (...) { pipeline.setup(new Parameters()); // Pipeline will get generator from container pipeline.setGenerator(file, ...); pipeline.addTransformer(xslt, ...); pipeline.setSerializer(html, ...); Environment env = new BackgroundEnvironment(...); pipeline.process(env); pipeline.recycle(); } This looks simple enough. The recycle() method is not part of the ProcessingPipeline interface, but all its implementations are Recyclable (I guess this is avalon speak for to be used in a pool, and therefore I can assume that the implementations are not threadsafe, right?). Yes, pipeline interface is inherently not thread-safe. Am I correct in assuming that cocoon pools instances of the pipelines (one per thread) when in a webapp context ? One pipeline object per ... ahem ... pipeline being assembled. Usually this means one pipeline per Request (meaning, internal requests get own pipeline object), which can total to more than one per thread. Vadim
isolate the pipeline component from rest of cocoon
Hi there, I have a bit of cocoon knowledge from the 2.0.x days, and I remember that the whole pipelining system was deeply integrated inside cocoon. I have a need for an XML pipeline system and cocoon's SAX pipeline technology is an obvious candidate. However I am not so interested in the rest, i.e. the blocks, the servlet integration, etc. Therefore I would really like to be able to isolate the pipeline from the rest of cocoon, and be able to configure and instantiate it alone. I would like to do things like this: Generator producer = MyFactory.newGenerate(...); XMLConsumer consumer = MyFactory.newConsumer(...); pipeline.generate(producer, consumer); I am willing to dig around in SVN, however it would help a lot with a few pointers. I could not find anything beyond 2.1.x on the wiki/official site. Do you have any pointers to things I should look at in SVN, or even if what I am asking for is remotely possible? best regards, - Renaud
Re: isolate the pipeline component from rest of cocoon
I am willing to dig around in SVN, however it would help a lot with a few pointers. I could not find anything beyond 2.1.x on the wiki/official site. Do you have any pointers to things I should look at in SVN, or even if what I am asking for is remotely possible? There have been discussions around this very topic now and then.But there is not much to point to. You know were the classes are ..just something that needs to be done. MO this would be more than useful!. cheers - Torsten
Re: isolate the pipeline component from rest of cocoon
On 10/20/06, Torsten Curdt [EMAIL PROTECTED] wrote: ...MO this would be more than useful!... Sure - anyone who makes this happen is going to get a Hero Plate I guess. -Bertrand
Re: isolate the pipeline component from rest of cocoon
Renaud Bruyeron wrote: Therefore I would really like to be able to isolate the pipeline from the rest of cocoon, and be able to configure and instantiate it alone. I would like to do things like this: Generator producer = MyFactory.newGenerate(...); XMLConsumer consumer = MyFactory.newConsumer(...); pipeline.generate(producer, consumer); It is possible to use pipelines directly, it would look something like: ...setup container... ProcessingPipeline pipeline = new [Non]CachingProcessingPipeline(); pipeline.enableLogging(...); while (...) { pipeline.setup(new Parameters()); // Pipeline will get generator from container pipeline.setGenerator(file, ...); pipeline.addTransformer(xslt, ...); pipeline.setSerializer(html, ...); Environment env = new BackgroundEnvironment(...); pipeline.process(env); pipeline.recycle(); } I am willing to dig around in SVN, however it would help a lot with a few pointers. I could not find anything beyond 2.1.x on the wiki/official site. Cocoon 2.2 (spring based) is in trunk, Cocoon 2.1 (avalon based) is in branch. Do you have any pointers to things I should look at in SVN, or even if what I am asking for is remotely possible? It might be more complex than what you want but it's possible. Vadim
Re: isolate the pipeline component from rest of cocoon
It is possible to use pipelines directly, it would look something like: ...setup container... Sorry, but please without that ;-) cheers -- Torsten
Re: isolate the pipeline component from rest of cocoon
Torsten Curdt wrote: It is possible to use pipelines directly, it would look something like: ...setup container... Sorry, but please without that ;-) Where would you get dependencies from? SourceResolver, Cache, Store, what else? And, AFAIU, setting up Spring is way easier than Excalibur. Vadim
Re: isolate the pipeline component from rest of cocoon
Vadim Gritsenko wrote: Renaud Bruyeron wrote: It is possible to use pipelines directly, it would look something like: ...setup container... Is that the avalon container, or is it a spring container? Is there an example of this somewhere? ProcessingPipeline pipeline = new [Non]CachingProcessingPipeline(); pipeline.enableLogging(...); while (...) { pipeline.setup(new Parameters()); // Pipeline will get generator from container pipeline.setGenerator(file, ...); pipeline.addTransformer(xslt, ...); pipeline.setSerializer(html, ...); Environment env = new BackgroundEnvironment(...); pipeline.process(env); pipeline.recycle(); } This looks simple enough. The recycle() method is not part of the ProcessingPipeline interface, but all its implementations are Recyclable (I guess this is avalon speak for to be used in a pool, and therefore I can assume that the implementations are not threadsafe, right?). Am I correct in assuming that cocoon pools instances of the pipelines (one per thread) when in a webapp context ? - Renaud