dirty-water report of Spoon: analysis and mitigation

[monperrus]

thanks a lot @randomicecube for preparing the dirty-water report of Spoon
https://gist.github.com/monperrus/34663084981de3c56f3120f932e0a4b7

let's fix everything

where do the 4 404 github_url come from? (extracted? computed)

index	package_name	github_url	github_exists
1	org.slf4j:slf4j-api@2.0.16	https://github.com/qos-ch/slf4j/slf4j-parent/slf4j-api	False
2	org.assertj:assertj-core@3.26.3	https://github.com/assertj/assertj/assertj-parent/assertj-core	False
3	com.google.guava:guava@33.3.1-jre	https://github.com/google/guava/guava	False
4	ch.qos.logback:logback-classic@1.5.10	https://github.com/qos-ch/logback/logback-classic	False

[randomicecube]

@monperrus the URLs are computed from each dependency, using the following command:

mvn help:evaluate -Dexpression=project.scm.url -Dartifact={group_id}:{artifact_id}:{version} -q -DforceStdout

I believe that when researching, I got the idea to use project.scm.url from here: https://maven.apache.org/plugins/maven-project-info-reports-plugin/scm-mojo.html

As an example, if you have this call for the first package in this table, you get the following interaction:

❯ mvn help:evaluate -Dexpression=project.scm.url -Dartifact=org.slf4j:slf4j-api:2.0.16 -q -DforceStdout
https://github.com/qos-ch/slf4j/slf4j-parent/slf4j-api

You can also see that in maven's central repository for this dependency, the source control URL on the tab at the bottom right of the page leads to a URL which doesn't exist: https://central.sonatype.com/artifact/org.slf4j/slf4j-api/2.0.16

I'm not quite sure how this URL is actually acquired, since the pom for this dependency doesn't actually have an SCM tag, only its supermodule does (and that one has a correct URL) -- perhaps this is dynamically acquired from the supermodule? And if they (hypothetically) just append the project's name to the URL, it's bound for these errors to happen, I'd say.

Overall, it seems like this issue is probably related with maintainers not writing the SCM tag for submodules of a supermodule. What do you think?

[monperrus]

Interesting. There are probably two mitigations:

1. PR those projects to add project.scm.url.
2. We probably found a bug Maven itself on the dynamic generation of project.scm.url. We can PR Maven itself to fix it.

Option 1 is the easiest, could you give it a try?

Thanks!

[randomicecube]

@monperrus regarding the second point, I found a thread which might be useful for us here: https://stackoverflow.com/questions/50321827/which-urls-does-maven-resolve-with-a-subpath-in-child-poms

The URL to the project's browsable SCM repository, such as ViewVC or Fisheye.
Default value is: parent value [+ path adjustment] + artifactId
So, the default value for all of these elements includes both path adjustment and artifactId.

So essentially, this default value is the root cause here, since the SCM tag is not defined in some child modules, as with slf4j-api. Being a default value, it might even be the case that this is intended behavior (?) but I could investigate further.
Regardless, I'll try to open PRs for these projects soon. I'll be away from my computer until Monday morning, I'll come back to this by then.

[monperrus]

I'll try to open PRs for these projects soon.

great, thanks @randomicecube

[monperrus]

citing the Stackoverflow post:

As it turns out, project.directory overrides artifactId for all three URLs. This is unexpected and is not covered in the documentation.

This is exactly the bug we're talking about. It would be great to firt report it and then fix it.

[monperrus]

@MartinWitt suggests that there is a bug because we have way more than 17 dependencies in the supply chain.

@randomicecube could you have a look? thanks!

[randomicecube]

@MartinWitt are you sure? I'm saying this because looking strictly at spoon's pom.xml file, it looks like every dependency is being correctly parsed. Each sub-project (e.g., spoon-visualization) has a further pom.xml, with more dependencies, of course, but those aren't to be directly parsed here, I think.

One think which would be fair, though, is that I don't think the dependencies would be, for now, correctly parsed for the child projects -- that is, for spoon-visualization, for instance, we'd parse that child project's pom.xml, but the parent's dependencies would be ignored (and they shouldn't, I think).

Do you agree?

cc @monperrus

[I-Al-Istannen]

Spoon has 17 direct dependencies, but also inherits some from its parent. Even if your tool only analyzes direct dependencies, it should be 23:

❯ mvn dependency:tree -DoutputType=json -DoutputFile=/dev/stdout | grep -v "\[INFO" | jq '.children | length'
23

But the first line in your readme states (emphasis mine)

Dirty-waters automatically finds software supply chain issues in software projects by analyzing the available metadata of all dependencies, transitively

If you include those it gets closer to 45:

❯ mvn dependency:tree -DoutputType=json -DoutputFile=/dev/stdout | grep -v "\[INFO" | jq '[.children | .. | {groupId, artifactId}?] | length'
45

And then you also have omitted dependencies:

❯ mvn dependency:tree -Dverbose -DoutputType=json -DoutputFile=/dev/stdout | grep -v "\[INFO" | jq '[.children | .. | {version, groupId, artifactId}?] | unique | length'
53

[randomicecube]

Thanks for the detailed feedback, I'll take a look!

[I-Al-Istannen]

Martin Witt also rightly complained that this does not include plugins and their transitive dependencies. I omitted those as there isn't a good and easy way to fetch them. But this should come close (though it overcounts because it includes conflicts):

❯ mvn dependency:resolve-plugins | sed -n '/The following plugins/,$p' | tail +2 | head -n -8 | sed 's/\[INFO\] *//' | sort | uniq | wc -l
311

Quite a few more, but not all of them necessarily have much input in the final build.

[monperrus]

improved report at https://gist.github.com/monperrus/34663084981de3c56f3120f932e0a4b7

now 352 dependencies in the Spoon supply chain