Le 5 juin 2014 19:44, "Denis Gervalle" <[email protected]> a écrit :
On Thu, Jun 5, 2014 at 3:51 PM, Jeremie BOUSQUET <
wrote:
Hi,
2014-06-05 15:10 GMT+02:00 Denis Gervalle <[email protected]>:
On Thu, Jun 5, 2014 at 2:27 PM, Thomas Mortagne < [email protected]
wrote:
On Thu, Jun 5, 2014 at 2:14 PM, Denis Gervalle <[email protected]> wrote:
What happen if you also use dependency A not just because of B ?
You put a dependency on A.
But you may not see that so easily when you change a few line in an existing module. Nothing will complains until you remove your deps to B.
How do you determine "because of B" ?
By thinking.
Ok, I rephrase my question. Could you define what you consider a usage of dependency A because of B and the opposite ?
If I understood, you depend on B, and by transitivity, you also depend on A (indirectly). Then, from your module you use some classes of A (brought to you indirectly, so "magically", it compiles). So you never explicitly declare the dep on A from your module. If one day you remove dependency on B for any reason, you will also loose A (and also you can say it's an explicit dependency that is not "explicit", though it's a matter of taste or point of view to consider transitive deps not to be explicit ...).
The opposite would be to declare explicit dep on A from your module, while you don't need it at all - only B needs it.
You get it fully. The question is, could we define in a few word when the transitivity could be used, because the usage of A is just "because of" B ? Else, we should add a deps on A as soon as we import any class from A IMO.
And what to you think of xwiki-commons-test-component that is a
deps
of
xwiki-platform-core ?
It's wrong IMO. Any forced dependency is wrong IMO.
Should we remove it ?
Yes we should remove it.
Why do we get it ? Its removal could become an nightmare... but if we agree on that, we should remove it ASAP.
What about deps for logging ?
Depends how you use it, the logger used with @Inject is an official feature of our component framework so xwiki-commons-component-api should be enough.
And could we add xwiki-commons-stability (probably provided scope)
to a
high level pom to avoid adding/removing it all the time ? (or forget it, since it come with xwiki-commons-component-api currently) ?
It's far from being used everywhere and there is no rule forcing to use it, you set @Unstable when an API is unstable, it's not forbidden to not go through @Unstable. Plus you are supposed to remove that annotation after some time.
Ok, so not deps at any scope in any high level poms. This seems opposite to what Sergiu proposed, but it would be nice to agree on a rule.
To sum up, currently I am not sure the exception rule "because of" is clear enough to not create confusion. I also agree with Sergiu that we should list all (no warning of used deps not declared in dependency:analysis), this make the rule clear at least. I am not against factoring common infrastructure in a single place, but Thomas seems to be clearly -1.
I'm not sure it relates to what you describe above. Factoring deps in a common infrastructure is more a matter of using dependencyManagement at correct level (top-most usually), but the rule to declare only deps that are really needed per module is a very widely used best practice.
From a contributor (not committer) point of view, it's sometimes annoying to have to update your builds because of this kind of changes. Obviously it would be difficult to consider those dependency graphs as APIs :) , but at the same time when they evolve too frequently, it can be considered painful to follow by devs in general ... Some are not impacting, but for example regarding @Unstable, if it's not brought anymore by xwiki-commons-component-api, it will break my build when I upgrade xwiki versions.
This is bad, since I do not see any reason for xwiki-commons-component-api to bring xwiki-commons-stability to you.
With Maven, it's not really a "bad" practice, to consider that some poms are used mainly to bring to you a set of logically related dependencies
by
transitivity (there's even the "import" scope, though it's not the best sample of good practice I agree). Question is more (case by case), is it really "bad" to bring @Unstable by default to everyone that develops an xwiki component, even if he will never use it / don't use it anymore ?
I would be positive, especially because it is not a runtime deps, and we can simply use the "provided" scope. Thomas do not seems to agree.
I agree with Thomas, forced dependencies in your build are bad. The only forced dep I ever used in my root poms is junit, but here the objective was really to force it ;) In this case above, it would not be a pom really part of your build, but a pom to be used as a dependency by devs users of xwiki apis. That's why I talked about "import" scope, it's the typical use-case for it. But I agree that in that case it's more maintenance work, additional poms and deps graphs, so maybe not worth it at all. I'm not sure I'd do it if I were you, even if I 'd find it nice as contributor ;)
All I really want at the end, is that we have clear rule. Unless we can define what "because of" is, the only rule I see valid currently, would be to explicitly depends on any module we do an import from.
It would be nice to have more feedback from other committers ! This is
not
a minor aspect of our best practice IMO.
WDYT ?
On Thu, Jun 5, 2014 at 11:42 AM, Thomas Mortagne <
wrote:
For me the rule to apply is simple: when you require dependency A because of another dependency B (B expose A in it's API, you
implement
an interface of A to be found by B, etc.) you should not
explicitly
depend on A.
On Thu, Jun 5, 2014 at 11:32 AM, Denis Gervalle <[email protected]> wrote: > Hi devs, > > I am reviving this proposal since we never came to a conclusion, and I have > the feeling that our deps become more and more convoluted. > > To resume what I get from past history with my own vision: > > 1) Since our modules are getting really modular, it should never silently > depends on transitive dependency of another module (with IMO some > exception, see 3). So any undeclared deps found by dependency:analyse > should be explicitly declare in the effective pom (Vincent POV as well) > 2) Apply maven principle, we should reuse and apply > convention-over-configuration > over configuration, so common dependency like slf4j, xwiki-commons-stability > ? ... should be in a high level parent pom > 3) We may rely on some very tight transitive dependency, for exemple, it > would be really curious that xwiki-commons-component-api stop providing > javax.inject, or that xwki-commons-test-components stop providing junit, > mockito, and al. > > It would be nice to add those rules in our best practice and to always > check our pom upon finishing changes in a module. The best would be to > enforce those rules, but this is not easy since static analysis is limited > and could create false positive. > > WDYT ? > > > On Mon, Aug 15, 2011 at 10:31 PM, Thomas Mortagne < [email protected] >> wrote: > >> On Mon, Aug 15, 2011 at 9:25 PM, Vincent Massol < [email protected]> >> wrote: >> > >> > On Aug 12, 2011, at 4:45 PM, Sergiu Dumitriu wrote: >> > >> >> On 08/12/2011 07:50 AM, Vincent Massol wrote: >> >>> Hi devs, >> >>> >> >>> Running mvn dependency:dependency-analyze produces interesting results. >> >>> >> >>> For example: >> >>> >> >>> [INFO] >>
>> >>> [INFO] Building XWiki Commons - Properties 3.2-SNAPSHOT >> >>> [INFO] >>
>> >>> … >> >>> [INFO] --- maven-dependency-plugin:2.3:analyze (default-cli) @ >> xwiki-commons-properties --- >> >>> [WARNING] Used undeclared dependencies found: >> >>> [WARNING] org.slf4j:slf4j-api:jar:1.6.1:compile >> >>> [WARNING] javax.inject:javax.inject:jar:1:compile >> >>> [WARNING] Unused declared dependencies found: >> >>> [WARNING] >>
org.xwiki.commons:xwiki-commons-component-api:jar:3.2-SNAPSHOT:compile
>> >>> [WARNING] org.xwiki.commons:xwiki-commons-test:jar:3.2-SNAPSHOT:test >> >>> [WARNING] org.hibernate:hibernate-validator:jar:4.2.0.Final:test >> >>> [WARNING] org.hamcrest:hamcrest-core:jar:1.1:test >> >>> [WARNING] org.jmock:jmock:jar:2.5.1:test >> >>> >> >>> The question is (for this module but more generally for all others): >> >>> * Should we add slf4j and javax.inject reps in the pom.xml for this >> module? (for ex today slf4j and javax.inject are found in the component-api >> dep) >> >>> >> >>> I think we should, wdyt? >> >> >> >> +1 as well. >> > >> > hmm actually we need to decide about the following: >> > >> > * In order to simplify writing pom.xml for modules having components >> (i.e. depending on xwiki-commons-component-api) I had added the following >> to xwiki-commons-component-api/pom.xml: >> > >> > <!-- Make it easy for components that wish to log - They don't have >> to explicitly import SLF4J --> >> > <dependency> >> > <groupId>org.slf4j</groupId> >> > <artifactId>slf4j-api</artifactId> >> > </dependency> >> > >> > * In the same manner we have a dependency on javax.inject in >> xwiki-commons-component-api/pom.xml: >> > >> > <!-- We add this dependency here so that users of the Component API >> just need to depend on this artifact and >> > don't have to explicitly add a dependency on >> javax.inject:java.inject. --> >> > <dependency> >> > <groupId>javax.inject</groupId> >> > <artifactId>javax.inject</artifactId> >> > <version>1</version> >> > </dependency> >> > >> > So the question is: do we want to force each module depending on >> xwiki-commons-component-api to have to declare an explicit dep on >> javax.inject and org.slf4j? >> > >> > I'm not so sure about that… >> >> I'm -0 and near -1 to list this kind of dependencies, using slf4j or >> javax.inject are what you HAVE TO use when you write an XWiki >> component so it's redundant from my POV. >> >> > >> > WDYT? >> > >> > Thanks >> > -Vincent >> > >> >>> Note that the "Unused declared dependencies found:" doesn't always >> generate correct results as is the case here. This is mostly because it's a >> static byte code check so any dep used at runtime will be considered unused. >> >>> See >>
http://www.sonatype.com/books/mvnex-book/reference/optimizing-sect-dependenc...
>> >> >> >> Some of these dependencies are not used directly by us, but are needed >> >> transitively by another library. For example, slf4j needs logback, which >> >> we never use directly (although we don't really declare it in every >> >> module that does logging). Hibernate needs us to pick a cache, a >> >> connection pool, validator, and a bytecode manipulation utility. So yes, >> >> we can safely ignore most of these false negatives, but we should still >> >> try to remove those that are really wrongfully declared as dependencies. >> >> >> >>> Thanks >> >>> -Vincent >> > >> > _______________________________________________ >> > devs mailing list >> > [email protected] >> > http://lists.xwiki.org/mailman/listinfo/devs >> > >> >> >> >> -- >> Thomas Mortagne >> _______________________________________________ >> devs mailing list >> [email protected] >> http://lists.xwiki.org/mailman/listinfo/devs >> > > > > -- > Denis Gervalle > SOFTEC sa - CEO > _______________________________________________ > devs mailing list > [email protected] > http://lists.xwiki.org/mailman/listinfo/devs
-- Thomas Mortagne _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Denis Gervalle SOFTEC sa - CEO _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Thomas Mortagne _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Denis Gervalle SOFTEC sa - CEO _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
_______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Denis Gervalle SOFTEC sa - CEO _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs