Pants doesn't complain when goal doesn't exist for target

[sid-kap]

Hi,

I have a pants project which contains a hadoop_binary target. (It actually contains two hadoop_binary targets.) My goal was to compile the project and create a bundle of one of the binary targets (in the dist/ directory).

Unfortunately, I forgot to include the target in my pants command. So, instead of

./pants bundle path/to/project:my_binary_name

I typed

./pants bundle path/to/project

As a result, pants did not produce a bundle.

Eventually, I realized that the bundle was not created, and that I was typing the wrong command. However, I wish that pants would have given me an error when I typed this command.

Do you think this would be a useful change? If so, I might try to submit a PR.

[kwlzn]

path/to/project implicitly resolves to the address path/to/project:project (based on inferring the target name from a dirname of the BUILD file path). the fact that it did not throw an error in this instance indicates that this target actually exists - but that the targets type resulted in a no-op for the bundle goal.

beyond bundle itself, I think it would be helpful to error out when a goal is specified that does not apply to any of the specified targets - tho it'd probably be good to socialize the idea on pants-devel@ etc to ensure there aren't use cases that this behavior needs to be permitted for. at a minimum, you'd probably want to put this checking behind a feature flag that can be toggled on/off via the repo-specific pants.ini.

my gut feeling is that this is probably somewhat non-trivial due to the way pants works - ymmv tho.

[sid-kap]

Thanks for your reply!

What you said is correct -- there is a target called path/to/project:project, but it is a scala_library target. As you said, the goal bundle doesn't apply to scala_library targets, which is why it no-oped.

I'll post this on pants-devel@ to solicit more opinions.

[Eric-Arellano]

Closing as V2 goals now have a mechanism for doing this by requesting Addresses or Targets in the rule signature, then erroring if it's == 0. Currently, we only error in certain cases if too many targets are given, rather than too little, but it's trivial to implement.