Add a global --tag option to filter targets based on their tags.
Review Request #2362 — Created June 13, 2015 and submitted — Latest diff uploaded
--tag=foo,bar includes only targets that have at least one of those tags. --tag=-foo,bar includes only targets that have none of those tags. --tag=foo,bar --tag=-baz includes only targets that have at least one of the tags foo or bar, and do not have the tag baz. The filtering is on target roots, of course, similar to spec_excludes and exclude_target_regexp. Dependencies are included in the targets passed to Task.execute() regardless of their tags. We already had similar filtering functionality in the Filter task. So I refactored some of that into a new filtering.py file under util. While doing so, I noticed that the help text on Filter's options was wrong: In --tag=-foo,bar the '-' prefix applies to both foo and bar, and --tag=-foo,+bar is not sensible. While fixing up the help strings I realized that it would be more succinct to move the common help text (the part explaining about the prefix and how multiple filters are specified) into the goal description. Then I realized that it would be pretty verbose to put that in the with_description() text in register.py, so I added a little thing that uses the task's docstring as the description, if one isn't provided explicitly. Note that Filter still has its own --tag option, which we can think about deprecating in a future change. IMPORTANT: While testing, I noticed that we had a longstanding bug, where multiple filters of the same type wouldn't work, because in the loop the filter() function was capturing the same references in its closure on each iteration. See the test I added in test_filter.py - that test fails with no other changes at HEAD. Also note that the test marked as xfail is behaving as intended: according to the implementation, separate specifications of the same option are ANDed (while comma-separated values in a single specification are ORed). If we want the behavior to be different then we can discuss, but for now at least that xfail was inappropriate and misleading.
CI passes: https://travis-ci.org/pantsbuild/pants/builds/66695055
Loading file attachments...