RE: [jira] Commented: (MNG-415) allow exclusion of certain depend encies from inclusion in an archive

[Maczka Michal <mi...@imtf.ch>]

Hey Brett!


>my comments inline. To me, you seem to be viewing all of dependency 
>management through the eyes of a single use case (being war 
>exclusions), which is where we are disagreeing.
>
>Maczka Michal wrote:
>
>  
>
>>And it indeed requires some amount of extra work in a very limited set 
>>of cases ( "triples what needs to be specified " is not really 
>>accurate - see below).
>>
>>    
>>
>To allow for multiple values:
><scope>foo</scope>
>becomes
><scopes>
>  <scope>foo</scope>
></scopes>
>if we are to retain the conventions used so far.
>
>  
>

This is just one of many possibilities.

Some of others are:

a) use predefined tags (do you want to have more scopes then "compile",
"runtime" and "test"?) :

<dependency>
    <compile/>
    <test/>
</dependency>

b) use "flat" structure
<dependency>
    <scope>test</scope>
    <scope>compile</scope>
</dependency>

c) change the rules which exists and add one more predefined scope

one more scope maybe introduced (e.g. named "global")

and your rules:

compile --> runtime   (--> means "imply")
runtime --> test

can be replaced with:

global --> compile
global --> runtime
runtime --> test
compile --> test


And have a default scope set to "global"

I am sure that there is bunch of much better options then those which I
found quite quickly.

But the question is: do you want to still consider changes in that part of
m2 or for you this part is fixed?
At the moment the number of POMs which are using scope tag is very low. 
So this will be no almost breaking change.
So if you are saying: scopes in m2 are not going to change and this is not
something which we are ever going to consider - I think that this discussion
is useless and I am not going to waste my precious time anymore.

I also believe that the best option here will be this one, which will be the
most intuitive for users. So they can guess what are the rules without even
looking at the documentation. So my proposition c) and what exists now as
for me seem to be an inferior solutions as users have to learn some
unintuitive and very arbitrary rules. But to know that for sure what's easy
and intuitive you must actually ask your users what is simple for them. As
something which is simple for you or for me does not have to be simple for
avg. user.


You wrote:

"Hugo - if you make "runtime" imply "compile" you can never have a runtime
only dependency, which is a valid and common use case (i.e. anything where
you only call into it via reflection - like a JCE provider, etc). "

I can replay to that

if you make "compile" imply "runtime" you can never have a compile time only
dependency, which is a valid and common use case. 

I don't think that any of those two usecase is more important and that's why
it would be nice to support both.



[...]

>Or for simplicity, let's make servletapi a default exclude for WAR and 
>not configure anything. It's not worth changing the defiition of the 
>dependencies for a single JAR.
>
>  
>
you may have classes from servletapi incorporated into some other jars (e.g.
j2ee). And you still need to exclude some other jars.

>>Sure. But war is an assembling plugin and if you have project 
>>packaging set to "war" you are actually not likely going to use the 
>>same pom for creating other assemblies and you definitly know 
>>something about environment in which this artifact will be used 
>>(servlet container). It is just a question if standard attributes of 
>>dependencies are sufficient or not and if they are not sufficient how it
can be done in the other way.
>>Anyway scope tag (scope=test) is already used for filtering out 
>>artifacts
>>    
>>
>>from assemblies so I don't see what's actually the problem with that.
>  
>
>> 
>>
>>    
>>
>I'm probably missing something, but I can't see how what you are saying 
>here is related to my point at all. I'm saying you can't change the 
>definition of scope for the entire system to fit the war use case.
>
>  
>

>>>>From a pure point of view a compile time dependency is always needed 
>>>at
>>runtime (unless the code is never hit, or the server provides the jar, 
>>all of which is environmental, not by >definition)
>>
>>Not true. For example servletapi is a jar agaist which you want to 
>>compile but most servlet containers have classes from that jar bundled 
>>in their own jars which have completely different names.
>> 
>>
>>    
>>
>Read what I wrote again. The servletapi, needed at compile time, is 
>also needed at runtime. You've just said so yourself. Just because the 
>container happens to provide it, doesn't change anything.
>  
>
One more time: no! The container doesn't have to provide _the_ servletapi
jar which was used at compile time.
It just needs to provide the compatible substitute. It maybe be a different
jar (e.g. something like geronimo-spec) or servletapi with higher version.  


>Since we seem to be stuck on servletapi, let's go to a practical 
>example. You run some tests using a mock library that calls your code, 
>which utilises the servletapi. No container. What scope does the 
>servletapi have in your proposal? You'll probably answer "compile, test"
>which in this example is valid. What if these tests were not junit 
>tests, but some other set using the "runtime" classpath? So you add 
>runtime to servletapi and now it is automatically included in the war - 
>exactly what you didn't want.
>  
>
Can you make this example bit more clear? (my comments maybe not accurate as
I am not sure if I understood you)

I thought that we were looking for simple way of addressing common use cases
(80/20 principle), and for finding sensible defaults for them, weren't we?
This example seems to be artificial and such tests can be always moved to
another module (with m2 this will be very easy).
With maven2 it is also every easy to move entire java code and tests for it
to another module and in war project work only with jars.

>We can't mix concepts. scope != bundling, so we can't try and abuse it 
>to make it the same. It will give you the right default 99% of the 
>time, the other 1% you need to configure it, under the war plugin in 
>terms of bundling.
>
I agree that there might be some limited cases when scope has little to do
with bundling. But there is a lot of cases when those two concepts are
practically overlapping. And those exceptional cases when scope tag is not
sufficient can be threaten in exceptional way.


>  
>
>>I think this case qualifies for such change :)
>> 
>>
>>    
>>
>That's much easier for you to say since you aren't doing the work :)
>
>You can't argue the point in terms of one use case. You need to address 
>how all the counter examples raised would work under your scheme too.
>  
>
Before proposing that solution of course I had already checked some other
uses cases.
And I am pretty confident that it will work well for the most common cases
(wars,
ejbs). I am also using 2-3 other containers and component technologies and
for them this approach seems to be working fine.  
I am not claiming that this is going to work in every case.  What I am
saying is: this might be helpful 
in the case of some plugins and I feel that in the case of plugins when it
won't be helpful it won't be harmful. 
After all it just gives you no more or less then a possibility for more
precisely describing your dependencies. 
I don't see a single issue with that. If you cannot use the knowledge
regarding scopes it does not mean that this
info must be always omitted.
But if you can use it sometime  - why not to do this?
If you agree that this is  something which at least can be investigated
further and still be considered for m2 I hope we that my opinion can be
confirmed (or rejected!) in the practice.


>Then, if there are others who agree that is better, we can look into it.
>But I think the confusion here is just how it is being communicated, 
>not that the concept is broken.
>  
>
What was surprising to me is that you have disqualified this proposition in
2 minutes without even reading with attention what I have written in my jira
comment and judged it as something "too complicated".


best regards

Michal





---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org

Re: [jira] Commented: (MNG-415) allow exclusion of certain depend encies from inclusion in an archive

[Brett Porter <br...@apache.org>]

Maczka Michal wrote:

>What was surprising to me is that you have disqualified this proposition in
>2 minutes without even reading with attention what I have written in my jira
>comment and judged it as something "too complicated".
>  
>
I started to reply to your invidual points until I got to this. I've 
given your responses careful consideration and detailed replies, and 
spent reasonable time on it. Sorry you feel this way.

>So if you are saying: scopes in m2 are not going to change and this is not
>something which we are ever going to consider - I think that this discussion
>is useless and I am not going to waste my precious time anymore.
>  
>
This is not only my call to make, so I will bring this up again in the 
design discussions next week, and see if there is some consensus among 
people working in here. What you do with your time is up to you.

The problems you have highlighted centre around believing there is a 
need for a compile time dependency that is not runtime. This stems from 
believing runtime equates to bundling. Runtime is not only bundling - it 
is also used for classpath construction, such as for the execution of 
javadoc.

An alternative has been proposed which we will discuss alongside the 
other design items: scope = container.  This takes the bundling into 
consideration, separate from runtime as it should be. A container scoped 
dependency would be available at compile, run and test time, but would 
not be bundled. This was a quick idea from Jason, but its the best 
alternative I've heard so far.

- Brett



---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@maven.apache.org
For additional commands, e-mail: dev-help@maven.apache.org