RE: "It would be massively incompatible with the existing toolchain; not just Maven but Gradle,SBT, static analysis tools, bazel, and everything else that sits on top of the Maven repository system. The cost of introducing this now vastly outweighs any conceivable benefit."
As described above, the intent for this would be that the pom.xml published to the repository system would be entirely element-based (as today), so nothing downstream from the user's project would be affected. Absolutely agree that the Maven repository system is highest priority. I just did some more testing, and it appears that the pom.xml in the user's project is the same as the pom that gets published (e.g. included in the jar generated by package), which would absolutely lead to breakage. That level of breakage is contrary to the whole point of this proposal (allow a user local pom.xml to use attributes, but keep the pom.xml published to repositories the same as today with elements). I thought that the published pom.xml included in the release was generated/modified, but that's apparently not true (and clarifies the purpose of the Build vs Consumer POM proposal ( https://cwiki.apache.org/confluence/display/MAVEN/Build+vs+Consumer+POM). In the past I could have sworn the final pom published was not the same as the pom.xml in the user directory, but that appears to not be the case (at least for an ordinary project). It's possible that some tooling I was using (e.g. my CI tool) was tweaking the pom during publishing, which is where I got the idea that happened by default. To restate more simply: The pom.xml in the user's directory is the *same* as the pom.xml that is bundled into the JAR produced by package. So, *no* changes can be made to the userland pom.xml without also breaking downstream publishing. That's a killer for the proposal, at least if/until Build vs Consumer comes along. In the meantime, the only thing I can think of is for Maven core to add a bit more native support for polyglot to make it easier to work with, but that's an entirely different thing. Thanks for the feedback - I'll include this information in the bug tracker for the issue for future reference and it can be closed. One kind of weird point of clarification: The XSD declares a ton of optional values and also includes documentation, but doesn't appear to actually enforce validation (that's done by Modello generated Java code). This appears to be a very common misconception about the Maven XSD - that it's involved in the validation process. I was certainly a bit surprised to see everything marked as optional in the XSD when I downloaded it and reviewed it. Cheers, -Will On Sat, Dec 12, 2020 at 11:02 AM Hunter C Payne <hunterpayne2...@yahoo.com.invalid> wrote: > So there have been a few comments so far (yea) so I'm going to try to > address them here: > 1) choice of formatAny format that specifies a POM should have validation > (which JSON, HOCON and XML do). YAML should be a non-starter as it has no > validation (or types and it depends on invisible characters for > formatting). But Unbound only translates between formats and so you can > still write your 1000s of lines of XML if you want but it allows the rest > of us to have 10 line POM files. > But I've noticed that none of you have mentioned HOCON (which has > comments) so far which leads me to think that none of you know this format > or understand what it enables. Its an extension of JSON so it picks up all > the validation functionality. But it also has an include primitive which > allows abstracting of POMs in a really useful way. I can write a block of > HOCON and host it on a server somewhere in my infrastructure and then every > project in the org can add an include line to use that block of POM. > That's so much nicer than the way we abstract POMs currently. > Multi-project POMs are often a nightmare to do mainly because the > inheritance of POMs is so complex. > HOCON spec: https://github.com/lightbend/config/blob/master/HOCON.md > > 2) why we are doing thisIf you don't have problems getting teammates to > use Maven, then you are the lucky 1%. The rest of us have to work with > folks under the age of 35 and its pretty impossible to get them to use or > learn Maven due to XML. Also, my HOCON POMs are a small fraction as long > as the XML ones but with the same functionality. > > I'm just not understanding the resistance here. I feel like those that > are resisting don't really understand the current developer community. SBT > is a slow and very poor build tool but people use it because they dislike > XML that much more. Not sure why folks on this list don't understand > that. XML is pretty universally hated at this point. Perhaps best to just > accept that instead of talking about features (XSD) that almost nobody ever > used. > > 3) ease of integrationI'm only asking to have Unbound included, not to > change the POM format, or require significant changes in Maven itself. I'm > not sure why you would want this as the tooling should still use XML for > reasons of legacy and practicality. But forcing devs to write POMs in XML > by hand is currently required and pretty undesirable. > 4) what is wrong with XMLFor one the way lists and maps are represented > are very verbose and unwieldy. But the bigger thing for me is how > inheritance is done in Maven. Mutli-module projects are way too hard to > make work. A simpler block level inheritance that HOCON enables is far > more preferable. Consider this example: > https://github.com/hunterpayne/maven-unbound/blob/master/examples/rpm.conf > This block of code can be included by adding this to your HOCON POM: > include file("examples/rpm.conf") > > And now my pom can build RPMs in a single line. Pretty nifty and so much > better than repeating 80 or so lines of XML. > Hunter > > > > > On Saturday, December 12, 2020, 8:20:08 AM PST, Gary Gregory < > garydgreg...@gmail.com> wrote: > > On Sat, Dec 12, 2020 at 6:53 AM Robert Scholte <rfscho...@apache.org> > wrote: > > > Here's my unpopular response: I'm not going to invest in attribute > support > > for Maven. > > If the verbosity of the pom.xml is the first thing people complain about, > > well, then Maven is doing a pretty good job (if the build itself had > > issues, one would complain about that first, right?). > > By having elements only it is much easier to maintain Maven. > > It'll remove discussion as to: what would be an attribute and what should > > be an element? Can it be based on required versus optional? Allowing both > > for the same "fields" is probably a recipe for disaster. > > > > That's simple IMO and what we usually do at my day job: XML elements and > attributes are like Types (Classes) and its attributes (instance > variables). It does not need to be more complicated than that. FWIW, I'm > baffled at the suggestion that optional vs. required has anything to do > with this. > > Gary > > I'll leave it up to tools like polyglot to do some the transformation from > > your favorite language to the XML as expected by Maven. > > This is a clear separation, and it will give the Maven team the > > opportunity to focus on the real issues. > > > > So please join your forces and spend your energy on improving polyglot! > > > > thanks, > > Robert > > On 12-12-2020 11:04:33, Markus KARG <mar...@headcrashing.eu> wrote: > > Wouldn't it be a more modern and even more effective approach to add JSON > > support for POMs? We could keep POM.xml for legacy reasons but add > support > > for POM.json files. > > -Markus > > > > -----Ursprüngliche Nachricht----- > > Von: Will Iverson [mailto:wiver...@gmail.com] > > Gesendet: Freitag, 11. Dezember 2020 23:40 > > An: dev@maven.apache.org > > Betreff: [DISCUSS] Allow attributes shorthand in pom.xml > > > > One of the biggest complaints about Maven has long been the verbosity of > > the XML format. The verbosity is due in large part to the exclusive > > reliance on XML elements in Maven. > > > > Proposal: Allow Maven pom.xml to treat attributes as a short-hand for > > declaring configuration elements. > > > > Example: One of the most verbose sections of the pom for most projects is > > dependencies. A typical example: > > > > > > commons-io > > commons-io > > 2.8.0 > > > > > > Here is the same declaration expressed with attribute shortcuts: > > > > > > That's an 80% reduction in LoC, and would make Maven comparable with > other > > popular build tools (e.g. compare and contrast with other build tools at > > https://mvnrepository.com/artifact/commons-io/commons-io/2.8.0) > > > > REQUEST: Feedback on if this is something to pursue. I've done some > > research, happy to submit patches, but don't want to pursue if there is > > either a) technical reason[s] not to proceed I'm not aware of or b) a > lack > > of enthusiasm for the entire idea from the community. > > > > Basically, I'm looking for some feedback along the lines of a) love it - > > please submit patches so we can check it out, b) huh, maybe, willing to > > look at it, or c) this is a terrible idea, because X. Effectively, a > > totally non-binding vote on if this is worth exploring. > > > > I've discussed this with others online and done some research, so are a > few > > answers to objections/Qs as I currently understand. I may be > > wrong/uninformed about certain aspects, which would be very helpful > > feedback. > > > > Q: Won't this require a new Maven XSD to be generated? > > A: No. The current Maven XSD declares many elements, but is not actually > > involved in validation. While the current XSD is valuable for tools and > > documentation, it does not actually perform validation. > > > > Q: Wait, so what actually does the validation? > > A: It's all done in Java code generated by Modello. The maven-model > project > > (https://github.com/apache/maven/tree/maven-3.6.3/maven-model) relies on > > the Modello Maven Plugin ( > > http://codehaus-plexus.github.io/modello/modello-maven-plugin/) which in > > turn relies on Modello core (http://codehaus-plexus.github.io/modello/) > to > > generate the Java code that processes the pom.xml > > > > The proposal is to submit a patch for Modello that would allow the > > generated source to accept an attribute as an alias for input. If it's a > > valid element per the Maven maven.mdo file ( > > > > > https://github.com/apache/maven/blob/maven-3.6.3/maven-model/src/main/mdo/maven.mdo > > ) > > it will now accept an attribute as a shortcut. > > > > Q: Wouldn't this break, like, everything? > > A: It would only affect pom.xml files that are read at runtime. All > emitted > > pom.xml files would remain exactly the same. > > > > Q: Does this involve changing or rewriting the user's pom.xml? Isn't that > > the thing that's making it hard to support alternative formats for > pom.xml > > like polyglot poms, etc? > > A: Nope, the pom.xml on disk is still the pom.xml. A > > X.X.X would be the only flag > > recommended to declare that a pom.xml uses attributes for shorthand. > > > > Q: How much work is this to actually implement? > > A: It starts with a few lines added to the Modello code generation to > allow > > for attribute aliasing with a feature flag. Testing those changes through > > the rest of the dependency chain. Adding test cases throughout. > > Documentation. although as "now you can use attributes" is conceptually > > simple it's not too bad. Tools in the Maven ecosystem would be able to > > indicate they have been updated to support this by referring to the > simple > > term, "attribute shortcuts". Because nothing else changes, the only real > > documentation change would be "things that were elements can also be > > declared as attributes." The trickiest part is probably sorting out how > to > > manage the feature flag across the various components. I'm sure there's > > more with a huge ecosystem like this, but the actual changes to the > Modello > > code gen appear to be surprisingly minor. > > > > Q: What about tooling, like IDEs, publishing to Maven Central & Maven > > repositories, etc? > > A: Many IDEs appear to have implemented validation logic on top of Maven > > that currently will flag attributes as errors in a pom.xml. Those IDEs > and > > other tools would require updates to this validation logic. Because the > > rendered pom.xml output remains the same publishing tool chains and Maven > > repositories should be completely unaffected. > > > > Q: Any big issues you've identified? > > A: Many sub-elements are not actually processed by Modello or Maven > Model, > > but are instead passed along to the plugin. For example, > > elements. It would be up to each of these projects to eventually allow > for > > attribute aliasing (or not). Maven projects that rely on Modello would > have > > the choice to adopt the new version and turn on the feature flag (or > not). > > It's possible that this would be confusing for some users - i.e. "why > can I > > declare dependencies with attributes but not configuration values"? That > > said, I think it's manageable and would allow the ecosystem to slowly > > update. > > Q: Shouldn't we wait for Maven 5 to tackle this? > > A: There's an issue going back to 2008 about the verbosity of pom.xml - > > https://issues.apache.org/jira/browse/MNG-3397 - so... that's 12 years. > > While writing this email, I just realized I commented on that issue back > in > > 2014. Any proposal to dramatically change the pom is going to be a *huge* > > effort and is not at all what I'm proposing. This is literally the > simplest > > possible change I can think of that accomplishes the goal (dramatically > > reducing the verbosity of the pom.xml) with the least possible impact to > > the ecosystem. It's been twelve years. Maven 5 is years away. > > > > I know there is a voting system for changes to Maven, and this would be a > > huge userland change. If there is even a soft exploratory "yes" I'm happy > > to submit patches. Even better would be the assistance of an existing > Maven > > committer willing to help me navigate Apache requirements. If the > feedback > > is generally negative, that's fine too - I'll just go ahead and close the > > issue. What I don't want to do is submit patches and then have everyone > > yell at me. The Internet can be rough, you know. :) > > > > I know this is a long email - thanks for reading, and looking forward to > > feedback. > > > > Thanks, > > -Will > > > > P.S. I've been tracking my research on this approach with this issue > > https://issues.apache.org/jira/browse/MNG-7044, in case you are curious > > about the research/additional links. > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: dev-unsubscr...@maven.apache.org > > For additional commands, e-mail: dev-h...@maven.apache.org > > > >
