Ergonomic, hermetic, user-friendly software build system for Python, Go, Java, Kotlin, Protobuf, Scala, Shell. Pants lets you fearlessly scale up your codebase!

Pants

The way it is has a bit of redundancy ```            "globs": {
                "globs": [
                    "examples/src/java/org/pantsbuild/example/antlr4/ExampleAntlr4.java"
                ]
            },``` two `globs` are used, first to represent the `protected Globs globs` in `TargetAddressInfo` class, then the second `globs` refers to `protected List&lt;String&gt; globs`in `Globs` class.

`TargetAddressInfo.globs` expects an object and cannot parse an array

the correct way to represent empty globs is ```            "globs": {
                "globs": []
            },```

otherwise the change in the plugin would be non-trivial

<@U0AALHEDR>: <https://rbcommons.com/s/twitter/r/3614/>

Does anyone have a plugin (or way to confirm with a sequence of pants command lines) that all targets of type `xxx_library()` are referenced by targets of type `xxx_tests()`?  We are thinking of writing a small plugin to analyze our graph and not changes to pass CI unless this is satisfied.

Question #2: Is this something that folks would welcome as a contribution to pants proper?

<@U053H6LQS>: no... we have something to confirm that tests don't depend on tests

<@U053H6LQS>:  foursquare have said that they have a generic framework though? it might be in fsqio

Yeah that's something that graph validation could do

It just provides a product that a dummy task in gen/compile/test depend on, and then it inspects the graph and validates whatever it feels like

It has a helper task upstream of it that first tags the graph according to rules, to aid in validation

I think ours is going to end up being a bit custom too. we want to insure that all `java_library()` targets that gets committed has a  test target with a particular tag: `junit_tests(... tags=['ci'])`

it's simple, but also easy to do wrong, and hard to do efficiently

Yeah.  In our experience this was all too ad hoc to bother with some sort of generic selector based rule system

It's true that it's tough to do efficiently

doing too many walks... ie, one per check

I will try and move that into fsqio this weekend - then anyone can take use it and use it as a starting point if they want

it should be pretty straightforward, I think.

release bump ready for review: <https://rbcommons.com/s/twitter/r/3616/>

Heads up, the fix for the ivy lock seems to have made things worse for us