|Mark Struberg||Oct 22, 2011 4:08 pm||.txt|
|Jakob Korherr||Oct 23, 2011 2:59 am|
|Mark Struberg||Oct 23, 2011 8:55 am|
|Jakob Korherr||Oct 23, 2011 9:14 am|
|Leonardo Uribe||Oct 23, 2011 11:27 am|
|Leonardo Uribe||Oct 23, 2011 12:08 pm|
|Mark Struberg||Oct 23, 2011 3:37 pm|
|Leonardo Uribe||Oct 23, 2011 7:35 pm|
|Mark Struberg||Oct 24, 2011 1:37 am|
|Leonardo Uribe||Oct 24, 2011 10:17 am|
|Mark Struberg||Oct 24, 2011 3:14 pm|
|Leonardo Uribe||Oct 24, 2011 3:29 pm|
|Leonardo Uribe||Oct 24, 2011 7:02 pm|
|Mark Struberg||Oct 24, 2011 11:38 pm|
|Mark Struberg||Oct 25, 2011 4:32 am|
|Leonardo Uribe||Oct 25, 2011 8:47 am|
|Mark Struberg||Oct 25, 2011 9:23 am|
|Subject:||Re: [DISCUSS] how to get rid of tons of duplicated code|
|From:||Leonardo Uribe (lu4...@gmail.com)|
|Date:||Oct 23, 2011 12:08:02 pm|
Ok, let's check the proposal
MS>> So the suggestion is: MS>> MS>> 1.) cleanup myfaces-shared. mf-shared has almost no checkstyle rules applied.
Yes, sounds good.
MS>> 2.) add unit tests for myfaces-shared. Currently there are not many...
Ok, sounds good too.
MS>> 3.) move the shared code parts back to myfaces-shared and add unit tests.
So, this means do one step back and move the code from myfaces-core "shared" to myfaces-shared project? This breaks effectively the changes done some months ago to make easier work with myfaces core itself.
In that time the conclusion was: "core has priority over anything else, so shared code must live in core, but myfaces-shared project should just copy the code from there and have its own lifecycle" (these are my own words as I understood).
So this point does not have practical sense, and go against everything discussed earlier.
MS>> 4.) import myfaces-shared via maven dependency and use <minimizeJar> and <relocations> to package the stuff
maven-shade-plugin is a good "tool" but doesn't fit well in this scenario. The reason is we need an alternate release lifecycle for the shared code between myfaces core and other projects.
Historically that was the very first intention behind myfaces-shared project. Any myfaces core release requires some additional steps to do (TCK), so that becomes a problem when you try to release other libraries that depends of shared. So, to fix that, "shared" was created, so the code can be released in a independent way, and prevent myfaces core becomes an obstacle to release any other project (tomahawk, portlet-bridge, ... ). So, to release tomahawk you release shared first and then tomahawk.
maven-shade-plugin requires a released artifact to do its job. So, use it impose that restriction. In "shared" case, preserve the original intention becomes "imperative", and that's the reason why a goal was created to copy the code from myfaces-core shared, so the release manager can run this goal, commit the changes and then run a release.
My proposal in this case is do the same we did for shared, but for "myfaces commons" case. Then we can use maven-shade-plugin in other projects, but not over shared, instead over a released version of myfaces-commons-utils. Keep tomahawk or portlet-bridge as is, using shared project, because by its nature, those projects require classes that are not meant to be used outside those cases.
Note do any hack in this part makes a little bit "obscure" how to make changes, because everything becomes "centralized", but makes easier maintain code. Additionally, we increase the risk of "side effects", because a change done in core could introduce a bug in other parts. So at the end this is a matter of how to keep our code "balanced", even if some times it becomes a decision about "choose the less inconvenient alternative".
2011/10/23 Leonardo Uribe <lu4...@gmail.com>:
2011/10/23 Jakob Korherr <jako...@gmail.com>:
+1 - that's exactly what I have been trying to accomplish some time ago (introducing common-shades ). Unfortunately, I was not successful back then.
It is clear we need to "split" myfaces-impl into multiple modules. There are some parts that are useful for other projects. The code you did on commons-shade was the attempt to solve the problem of the duplicate code used on myfaces-test.
Now the objective is find a way about how to reuse code in myfaces core between multiple projects effectively.
However, there is a slight problem with moving all this stuff into MyFaces shared, which I want to point out: code size. If we really put all the code that is shared across any MyFaces subproject into shared, it will get fat and ugly (even more than it is right now). In addition, if we continue including the whole shared project into MyFaces core, MyFaces core impl will get bigger and bigger.
Yes, the problem basically is MyFaces shared does not have any order or any notion of API. There are code that is used only in tomahawk but not intended to use in any other place. There are some useful utitlities but sometimes without documentation, and there are some other code that is just obsolete. It it clear a cleanup of that location is necessary, but note priorities comes first, so this task has been delayed in order to deal with other important stuff. Now it is a good time to fix this.
Thus I'd like to suggest something similar which I wanted to accomplish with common-shades: Introduce a new shared module, which consists of many submodules that each handle a specific functionality instead of being one fat module. With this approach each MyFaces subproject would be able to pick out only the stuff it really needs. Furthermore we would see more easily which code in shared is not used anymore (I guess at the moment there is a lot of it), just by checking which modules are still used in our poms.
That is the big question, how to split myfaces-impl and shared. Precisely the intention of myfaces-commons-utils projects was take the stuff that is useful from shared and build an usable API for developers outside MyFaces.
For example, MyFaces HTML5 subproject was a good experiment to see which code is useful and should be added in a API. Some weeks ago I checked and removed all duplicate code to use myfaces-commons-utils. So the 1.0.2 release contains those classes taken from shared.
2011/10/23 Mark Struberg <stru...@yahoo.de>:
Hi! While working on the mafyces-commons cleanup I figured that we have tons of duplicated code spread over MyFaces.
As an example I like to mention myfaces-commons-resourcehandler. There are 43 classes in total, and 35 of them are just 1:1 copied from other projects to provide resource management, zip, etc. For me this is an absolute no-go. Those classes have neither tests nor any documentation where they got forked from. Nor will any bug which gets fixed in another module make it's way over to all the other projects containing that very forked code. That's just ... unbelievable unmaintainable.
There are 2 different ways to solve this (depending on the problem):
A.) drop the functionality and provide a generalized solution. The GZIP of myfaces-commons-resourcehandleris an obvious example: We now copy this code over the 4th time or even more. Instead of doing this, we should rather do it in the classic unix fashion: do one thing, but do it well. Which means I'd rather see all the GZIP stuff factored out into an own mf-commons module as a Servlet Filter. This can then get applied to what ever other mechanism you like. This could also (commonly) cover cases like detecting http UserAgents which are not able to handle zipped resources, etc. That way we could provide this logic ONCE and have complete freedom over the configuration.
B.) code reusable components once and use them in other projects (ev via shading it in). ClassLoaderResourceLoader would be a perfect candidate! I grepped through only the few pits which I have checked out locally and found this class 7 SEVEN times! I just can't believe that we can't move this stuff to a shared modul...
Same for FacesServletMapping. 6 times copied around, WebConfigProviderFactory 5 times, ... There are whole packages with 10++ classes which got copied 1:1!
I really could cry seeing this :(
What can we do to solve this?
Theoretically myfaces-shared should contain this stuff. This is exactly what it is for! Historically there have been some hand forged tweeks and ugly hacks, but nowadays we have the maven-shade-plugin to make our live easier.
So the suggestion is:
1.) cleanup myfaces-shared. mf-shared has almost no checkstyle rules applied. 2.) add unit tests for myfaces-shared. Currently there are not many... 3.) move the shared code parts back to myfaces-shared and add unit tests. 4.) import myfaces-shared via maven dependency and use <minimizeJar> and <relocations> to package the stuff
[+1] fine go ahead (ideally: yes, what parts can I help with?)  dont care [-1] wont work because ...
I've attached a file which contains all Classes which name exists multiple times in MyFaces. The number is the cound how often they exist in MyFaces. I excluded current20. Please note that classes with the same name do not necessarily have the same content - but quite a lot actually do have! (scroll to the bottom of the file ...)
-- Jakob Korherr