Le sam. 14 août 2021 à 18:56, Enno Thieleke <enno.thiel...@holisticon.de> a écrit :
> You're right. Currently, if I want to control my dependencies, I must > resort to excludes and define the dependencies I really want myself. And > putting something like that in one or two POMs and aggregating them in a > BOM is very dirty in my opinion. Not only is it noisy/verbose when it comes > to writing XML, it also adds nodes to the dependency graph where they > shouldn't be - not to mention unnecessary indirections. > The need for overrides is becoming bigger and bigger. As the Maven > ecosystem grows, artifacts with practically the same content but with > different coordinates are becoming more frequent. > I disagree, your solution is 1-1 with existing one, just somewhere else which is one pitfall as mentionned. It is also not new and solved since at least ten years so not sure what is specific for you here and what makes the common solution less practical or verbose. > Whether my proposal is about rewriting a POM dynamically is unclear at > this point. I came here to propose the feature of overrides and provided a > simple PoC, not the final solution. What I consider to be clear at this > point is that overrides need to exist in Maven as a core feature, not as an > extension, because this way the Maven Resolver can be changed as well to > honor the feature which would enable overrides to work transitively (all > part of the PoC). > It is already there, definition is deterministic and fully defined. Verbosity is a thing but then you want g:a:v definition too if you want to tackle it so I assume it is out of topic. > To your point: Yes, you can do all the stuff we're talking about today > already. Just not in a convenient way with a clean result - the effective > result is just "good enough". Once again, 1-1 with your proposal as of today since you just meta define deps in it. This brings me back to writing POMs and BOMs for "overriding" dependencies. > That's a lot of work. If one snippet of XML - a different grammar - does > all that, then I'd gladly take it. > Isnt it xlst ;), can be another way to impl it with mvn 4 but still not a core feature IMHO. Brings more drawbacks and no new feature. ________________________________ > From: Romain Manni-Bucau <rmannibu...@gmail.com> > Sent: Saturday, August 14, 2021 5:08 PM > To: Maven Developers List <dev@maven.apache.org> > Subject: Re: Request for Enhancement: Dependency Overrides > > Le sam. 14 août 2021 à 16:52, Enno Thieleke <enno.thiel...@holisticon.de> > a > écrit : > > > You are right, the feature is mainly useful for transitive dependencies. > > That's my main motivation anyway. > > > > Consider the following: > > You have a dependency to something that depends on > > org.hibernate.javax.persistence:hibernate-jpa-2.2-api. You yourself or > > another thing you depend on depends on > > jakarta.persistence:jakarta.persistence-api. Both artifacts have pretty > > much the same content. As for your classpath, you only want one of both > > (transitive) dependencies. But Maven will resolve both, because their > > coordinates are different. Here Maven's superpower of selecting the right > > version can't come into play. > > > > As for overrides inside the reactor (no matter the depth), this is mainly > > for providing a consistent result. If overrides affect transitive > > dependencies, they must also affect dependencies inside the reactor, but > > for such overrides a warning should be produced, since they are under > your > > control. > > > > I think my example even shows both cases: depending on something you > > override yourself and transitively. > > > > Think the question is: do you want to control your dependencies or not. If > you want the only option is an exclude (and you can create a pom > excluding/adding the one you want, kind of superbom for convenience). Using > dependency management is also doable very easily (what we tend to use as of > today in project I think). > > Your proposal is about rewriting a pom dynamically which is really an > extension IMHO and proven random with all the spec gav out there > (javax->jakarta migrations prove you must not rely on such behavior, it > leads to conflict if not explicitly done as mentionned which is a "less > worse" solution IMHO). > > So back to the point it is already doable and built in, you just propose > another grammar IMHO. > > > ________________________________ > > From: Bernd Eckenfels <e...@zusammenkunft.net> > > Sent: Saturday, August 14, 2021 4:31 PM > > To: Maven Developers List <dev@maven.apache.org> > > Subject: Re: Request for Enhancement: Dependency Overrides > > > > I don't really see it convenient at all. If you have to change the source > > anyway, you can also fix the POM, that's much less confusing than having > a > > project using technology x but declaring to use technology y. > > > > The only time I see dependency overwrites as somewhat useful if you want > > to replace transitive dependencies with fixed minor updates. This is > > however seldomly needed since maven adjusts version conflicts to the > > closest version anyway. > > > > Gruss > > Bernd > > -- > > http://bernd.eckenfels.net > > ________________________________ > > Von: Enno Thieleke <enno.thiel...@holisticon.de> > > Gesendet: Saturday, August 14, 2021 3:38:53 PM > > An: Maven Developers List <dev@maven.apache.org> > > Betreff: Re: Request for Enhancement: Dependency Overrides > > > > Hi Bernd, > > > > hopefully I understand you correctly: I'd expect the artifact (and its > > source and javadoc attachment) to change, nothing more. The source would > be > > left untouched. It is not the responsibility of the dependency resolution > > mechanism/engine to do anything else. > > > > Just to make sure that we have the same understanding: If I replaced one > > dependency with another one, I would effectively alter the POM to contain > > that other dependency instead of the original. After that Maven would do > > its thing as usual. > > > > Obviously, such a feature should be used carefully: If I replaced > > commons-logging:commons-logging with log4j:log4j, any code referencing > code > > from commons-logging would not compile anymore. It's the responsibility > of > > the developer to provide a working replacement. But this is fine, if you > > ask me, because if I replaced an artifact by exclusion and added a wrong > > dependency, the issue would be the same - and this is possible today > > already. One could say that an override would be a very, very convenient > > way the exclude and add a dependency. > > > > Kind regards, > > Enno > > > > ________________________________ > > From: Bernd Eckenfels <e...@zusammenkunft.net> > > Sent: Saturday, August 14, 2021 2:28 PM > > To: Maven Developers List <dev@maven.apache.org> > > Subject: Re: Request for Enhancement: Dependency Overrides > > > > Would you expect to only replace artifacts or also change the java source > > and class files to actually change the used classes? Your cases are no > > simple drop-in replace compatibility. Therefore I doubt a maven feature > to > > do so is any more useful than just blocking the deprecated artifacts. > > > > Gruss > > Bernd > > -- > > http://bernd.eckenfels.net > > ________________________________ > > Von: Enno Thieleke <enno.thiel...@holisticon.de> > > Gesendet: Saturday, August 14, 2021 1:24:26 PM > > An: Maven Developers List <dev@maven.apache.org> > > Betreff: Re: Request for Enhancement: Dependency Overrides > > > > So, you're saying that I can replace artifact coordinates with > > dependencyManagement.dependencies already? I'm not talking about > versions, > > scopes and exclusions, I'm talking about groupIds and artifactIds. My > idea > > is, simply put, if artifact commons-logging:commons-logging is found, > > org.slf4j:jcl-over-slf4j is used instead. Or if artifact > > org.hibernate.javax.persistence:hibernate-jpa-2.2-api is found, > > jakarta.persistence:jakarta.persistence-api is used instead. I don't > think > > that this is possible with the current version of Maven. If it is, please > > share a link to the docs covering this feature. > > > > Feel free to look at the example I provided. I feel it makes my idea more > > comprehensible. > > https://github.com/strohmattenverleger/maven-resolver/tree/MNG-4530 > > https://github.com/strohmattenverleger/maven/tree/MNG-4530 (requires the > > changes I made to maven-resolver) > > https://github.com/strohmattenverleger/maven-MNG-4530-example (the > > "magic" happens in the parent/root POM) > > > > Also, I don't think implementing this as an extension is the right thing > > to do. If it were to be fully implemented, it'd have to be a core feature > > of Maven and the Maven Resolver. This way the entire user base of Maven > > would benefit from it. > > ________________________________ > > From: Romain Manni-Bucau <rmannibu...@gmail.com> > > Sent: Saturday, August 14, 2021 8:55 AM > > To: Maven Developers List <dev@maven.apache.org> > > Subject: Re: Request for Enhancement: Dependency Overrides > > > > Hmm, > > > > dependencyManagement.dependencies already does that today. > > My point is mainly we shouldnt have two ways to do the same and even less > > two chained ways to do it. > > > > If you dont like mvn dep graph definition you can already write a mvn > > extension to write it differently so can also be an option, but as of > today > > i dont see anything not already smoothly doable. > > > > Le sam. 14 août 2021 à 02:01, Enno Thieleke <enno.thiel...@holisticon.de > > > > a > > écrit : > > > > > Hi Romain, > > > > > > the dependencyOverrides element is an addition. While it shares some > > > characteristics, it does not duplicate the dependencyManagement > element. > > In > > > fact, it's a part of it. Come to think of it, it may even be ok to > > > locate/place it "next to" the dependencyManagement element, thus making > > it > > > a "standalone thing", but I consider this feature to be a > > > dependencyManagement aspect, which is why I made a sub element of it. > > > > > > I'm not sure if I understand your question correctly: how to override > in > > > the override block? > > > If you mean how would a dependencyOverrides element of one POM be > merged > > > with an element of another POM, pretty much like a dependencyManagement > > > element - I made sure remove redundant sub elements. > > > If you mean how would an override override another override? Currently > > > processing overrides is just a sequential operation. Let's say the > order > > of > > > overrides is as follows (the notation "->" is to be read as "is > > overridden > > > with" or "is replaced by"): > > > > > > * X -> Y > > > * Y -> Z > > > > > > Here having a dependency X will ultimately result in having Z. And now > > for > > > a different order: > > > > > > * Y -> Z > > > * X -> Y > > > > > > Here having a dependency X will result in having Y. > > > > > > I think a helpful implementation would raise an error in both cases, > > since > > > having some sort of hyper edge here is questionable. > > > > > > Where I'm definitely not following you is why one would drop the > > > transitiveness of an override block/element. On the contrary: I'd not > > touch > > > transitiveness in any way. If an override brings transitive > dependencies > > > where an original didn't, then so be it. That'd actually be desired in > my > > > opinion. > > > > > > The final goal of overrides (if you ask me) is to manipulate a > dependency > > > graph in a way that is transparent to developers and plugins. If done > > > right, one wouldn't be able to tell the difference between a graph, > which > > > is clean from the start, and a graph, which has been altered using > > > overrides. > > > > > > Looking forward to your reply. > > > > > > Kind regards, > > > Enno > > > > > > > > > ________________________________ > > > From: Romain Manni-Bucau <rmannibu...@gmail.com> > > > Sent: Friday, August 13, 2021 2:41 PM > > > To: Maven Developers List <dev@maven.apache.org> > > > Subject: Re: Request for Enhancement: Dependency Overrides > > > > > > Hmm, I probably miss a thing but it looks like you just redid > > > dependencyManagement but duplicating it you created a new issue: how to > > > override in the override block? > > > Only neat way I see to handle that is to drop the transitiveness of the > > > override block which would defeat a lot dependency usage so I'm quite > > mixed > > > about this proposal without a more concrete use case. > > > > > > Romain Manni-Bucau > > > @rmannibucau <https://twitter.com/rmannibucau> | Blog > > > <https://rmannibucau.metawerx.net/> | Old Blog > > > <http://rmannibucau.wordpress.com> | Github < > > > https://github.com/rmannibucau> | > > > LinkedIn <https://www.linkedin.com/in/rmannibucau> | Book > > > < > > > > > > https://www.packtpub.com/application-development/java-ee-8-high-performance > > > > > > > > > > > > > Le ven. 13 août 2021 à 12:56, Enno Thieleke < > enno.thiel...@holisticon.de > > > > > > a > > > écrit : > > > > > > > Hi Michael, > > > > > > > > thank you for your answer. > > > > > > > > An interesting thought. It never occurred to me that a global > exclusion > > > > would help in a similar way, but then again, global excludes aren't a > > > > thing, yet (are they?). I think excluding globally and including a > > > specific > > > > dependency as a replacement is not as clear as having an override > > element > > > > in a POM. The former can simply be read as: "Aha, a global exclusion > > > here" > > > > and "Huh, a runtime dependency there." A good developer would surely > > add > > > a > > > > comment or two to make his/her intention clear. The latter implies > the > > > > intention in the very construct that is used. > > > > > > > > Also with global exclusions, when analyzing a (verbose) dependency > > graph, > > > > it wouldn't show excluded dependencies. Sure, this is the expected > > > > behavior, but again, it's not what a proper graph should look like if > > > > overrides/replacements were a thing. If A depends on B and B has been > > > > replaced with C, then I'd expect to see C instead of B or better yet, > > see > > > > that B has been omitted in favor of C. > > > > > > > > While reading the comments of MNG-1977 I found > > > > https://issues.apache.org/jira/browse/MNG-5652 which is a bit more > > like > > > > what I have in mind. > > > > > > > > Since even my proposal is in a very early stage (not to mention the > > code > > > > I've written), I neglected to mention that I also want to make > > overrides > > > be > > > > importable via the "import" scope. This way one can simply import a > BOM > > > as > > > > usual and benefit from any dependency cleanups done by others. My > > > > assumption is that providing overrides in a BOM is a somewhat > > "frameworky > > > > thing" (e.g. Spring, Java/Jakarta EE). One example for a use case is > my > > > > all-time favorite org.wildfly.bom:wildfly-jakartaee8:20.0.1.Final. It > > > > includes a managed dependency for Hibernate with some exclusions to > it. > > > > Now, when using this managed dependency, I must manually add the > > > exclusions > > > > (with the correct coordinates), which are actually missing, hence not > > > > provided via some other coordinate in the graph of said managed > > > Hibernate, > > > > which is somewhat painful. > > > > > > > > In essence, I think global excludes are definitely a good tool to > > have, I > > > > just don't think that they are the right tool when it comes to > > replacing > > > > dependencies in a concise way. > > > > > > > > Please let me know what you think. > > > > > > > > Kind regards, > > > > Enno > > > > > > > > > > > > > > > > ________________________________ > > > > From: Michael Osipov <micha...@apache.org> > > > > Sent: Wednesday, August 11, 2021 4:31 PM > > > > To: dev@maven.apache.org <dev@maven.apache.org> > > > > Subject: Re: Request for Enhancement: Dependency Overrides > > > > > > > > Am 2021-08-06 um 17:55 schrieb Enno Thieleke: > > > > > Hello, > > > > > > > > > > I've been using Maven for 10+ years now. It's great and I still > > prefer > > > > it over similar tools. Recently I encountered a scenario where it > would > > > > have been nice to be able to simply "override" a dependency. For > > example > > > by > > > > specifying _somehow_ that (in my case) commons-logging should be > > > > jcl-over-slf4j in the entire dependency graph of my project. > > > > > > > > > > Now you might think: "That is what exclusions are for". And I agree > > to > > > > some extent. But if I exclude a dependency (maybe just once, maybe > many > > > > times) and add a dependency to my project to replace what I have > > > excluded, > > > > this newly added dependency violates the convention of only having > > > (direct) > > > > dependencies you want to program against. But in this case, it is not > > my > > > > intent to use commons-logging and since it is not used in my code, > some > > > > other developer or plug-in might consider it redundant or obsolete. > > > > > > > > > > I encountered other dependencies I'd like to override too. > Generally > > > > speaking: Whenever some artifact needs to be replaced by a different > > one, > > > > which has a different groupId or artifactId, overrides might come in > > > handy. > > > > Consider the artifactId jboss-jaxrs-api_2.0_spec and its newer > version > > > > jboss-jaxrs-api_2.1_spec. It has version information encoded into it. > > > This > > > > is an issue the Maven resolver is not able to handle nor is it > designed > > > to > > > > handle it. > > > > > > > > > > Long story short, I've found a ticket for something like this: > > > > https://issues.apache.org/jira/browse/MNG-4530 > > > > > I also started to implement this feature (albeit differently from > the > > > > suggested approach in the ticket): > > > > > > https://github.com/strohmattenverleger/maven-resolver/tree/MNG-4530 > > > > > https://github.com/strohmattenverleger/maven/tree/MNG-4530 > (requires > > > > the changes I made to maven-resolver) > > > > > https://github.com/strohmattenverleger/maven-MNG-4530-example > > > > > > > > > > If you find the time, please look at what I've implemented so far. > In > > > > the example project you will find the maven-dependency-plugin > > configured > > > to > > > > copy the project dependencies. I've spoken to some of my colleagues > and > > > > they wanted me to also look the results of "mvn dependency:tree" > which > > > > turned out to be as expected (commons-logging has been replaced > > > entirely). > > > > Please keep in mind, that the feature is not finalized - it's more > of a > > > PoC. > > > > > > > > > > I would like your opinion on whether this feature would be nice to > > have > > > > and, if so, if the approach is the right one. I'm open to suggestions > > and > > > > I'd be happy to answer your questions. If you think this feature has > > what > > > > it takes to be merged into Maven once it's finished, I'd be willing > to > > > > invest more time into it. > > > > > > > > Hi Enno, > > > > > > > > I think you request could be/is related to > > > > https://issues.apache.org/jira/browse/MNG-1977. > > > > > > > > You brought up the most common example for this: Commons Logging. > > > > We obviously not only need global excludes, but also global > overrides. > > > > But I think global overrides can be achieved somewhat by global > > excludes > > > > and a runtime dependency with a comment above. It is not as elegant > as > > > > your proposal, but would ultimately solve the problem with a > different > > > > dependency tree. > > > > > > > > M > > > > > > > > --------------------------------------------------------------------- > > > > To unsubscribe, e-mail: dev-unsubscr...@maven.apache.org > > > > For additional commands, e-mail: dev-h...@maven.apache.org > > > > > > > > > > > > > >
