## Summary `Gitlab::Dangerfiles::TitleLinting::DRAFT_REGEX` is there to remove the not-ready markers a subject starts with (`Draft:`, `[draft]`, `(draft)`, `draft - `, `WIP:`, `[WIP]`) and the trailing `WIP` or `draft` that the "when used as suffix" examples in `base_linter_spec.rb` cover. The pattern starts with `\A*`, and the `*` makes the anchor optional. Because `remove_draft_flag` calls `gsub`, the prefix forms are removed wherever they appear in the subject, not only at the start. `BaseLinter#subject` lints the stripped text, so the commit linter and the merge request title linter check a different subject from the one the author wrote: - `Reject draft: titles` is linted as `Reject titles` and gets the `subject_too_short` warning, while `Reject ready titles` passes. - `Explain why a draft: prefix is stripped before the merge request is linted` is 74 characters long, is measured as 67, and passes `subject_too_long`. The same subject with `drafty` in place of `draft:` is reported. `subject_too_long` is one of the problems `warn_or_fail_commits` can report as a failure. ## Steps to reproduce On master at 1153e0d5 (4.13.0), with Ruby 3.2.9 from `.tool-versions`, save this as `repro.rb` and run `ruby -Ilib repro.rb` from the repository root: ```ruby require "gitlab/dangerfiles/commit_linter" title_linting = Gitlab::Dangerfiles::TitleLinting p title_linting.remove_draft_flag("Fix the draft: parser") # => "Fix the parser" p title_linting.remove_draft_flag("Show (draft) badge") # => "Show badge" p title_linting.remove_draft_flag("Handle a WIP: prefix") # => "Handle a prefix" p title_linting.remove_draft_flag("Use draft - mode") # => "Use mode" commit = Struct.new(:message, :sha, :diff_parent) p Gitlab::Dangerfiles::CommitLinter.new(commit.new("Reject draft: titles\n\nBody.", "abc", [])).lint.problems # => {:subject_too_short=>"The commit subject must contain at least 3 words"} p Gitlab::Dangerfiles::CommitLinter.new(commit.new("Reject ready titles\n\nBody.", "abc", [])).lint.problems # => {} ``` `Gitlab::Dangerfiles::MergeRequestLinter` (used by `commit_messages.lint_mr_title`) reports the same `subject_too_short` for an MR titled `Reject draft: titles`. ## What is the current bug behavior Any `Draft:`, `[draft]`, `(draft)`, `draft - `, `WIP:` or `[WIP]` in the middle of a commit subject or MR title is removed before the subject is linted, so the word count and the length are measured on the wrong text. The deprecated `has_draft_flag?` also returns `true` for these subjects. ## What is the expected correct behavior The prefix markers are removed only at the start of the subject, including several in a row (`Draft: [WIP] My title` becomes `My title`, which works today because `gsub` removes each one). The trailing `WIP` or `draft` keeps being removed, as `base_linter_spec.rb` expects. A marker in the middle of a subject is part of the text and is linted with it. The trailing forms also match an ordinary last word (`Save the draft` is linted as `Save the `). The specs pin that behaviour, so I have left it out of this issue. ## Relevant code - [`title_linting.rb#L6`](https://gitlab.com/gitlab-org/ruby/gems/gitlab-dangerfiles/-/blob/1153e0d5836c9cb08c10e021862246134331b0de/lib/gitlab/dangerfiles/title_linting.rb#L6): `DRAFT_REGEX = /\A*#{Regexp.union(...)}+\s*/i` - [`title_linting.rb#L17-19`](https://gitlab.com/gitlab-org/ruby/gems/gitlab-dangerfiles/-/blob/1153e0d5836c9cb08c10e021862246134331b0de/lib/gitlab/dangerfiles/title_linting.rb#L17-19): `remove_draft_flag` uses `title.gsub(DRAFT_REGEX, "")` - [`base_linter.rb#L70-72`](https://gitlab.com/gitlab-org/ruby/gems/gitlab-dangerfiles/-/blob/1153e0d5836c9cb08c10e021862246134331b0de/lib/gitlab/dangerfiles/base_linter.rb#L70-72): `subject` is the stripped first line - [`commit_messages.rb#L128-141`](https://gitlab.com/gitlab-org/ruby/gems/gitlab-dangerfiles/-/blob/1153e0d5836c9cb08c10e021862246134331b0de/lib/danger/plugins/commit_messages.rb#L128-141): `subject_too_short` is always a warning, `subject_too_long` can fail the job - [`base_linter_spec.rb#L117-130`](https://gitlab.com/gitlab-org/ruby/gems/gitlab-dangerfiles/-/blob/1153e0d5836c9cb08c10e021862246134331b0de/spec/gitlab/dangerfiles/base_linter_spec.rb#L117-130): the suffix examples The pattern was brought over from gitlab-org/gitlab in 34580735. GitLab's own `MergeRequest::DRAFT_REGEX` keeps the `\A*` shape, but the pattern it wraps starts with `\A` ([`merge_requests.rb#L10-12`](https://gitlab.com/gitlab-org/gitlab/-/blob/f967fc7d877b2be5ed289151f624752e6bdcfbd6/lib/gitlab/regex/merge_requests.rb#L10-12)), so there it only matches at the start of a title. ## Possible fixes Anchor the prefix forms at `\A`, allowing whitespace between repeated markers, and anchor the suffix forms at `\z`: ```ruby DRAFT_REGEX = /\A(?:(?:\[WIP\]|WIP:|\[draft\]|\(draft\)|draft:|draft\s-\s)\s*)+|(?:WIP|draft)\z/i.freeze ``` With this pattern the examples above keep their markers, `Reject draft: titles` passes and the 74-character subject gets `subject_too_long`. The existing suite still passes (626 examples, 0 failures on Ruby 3.2.9) and RuboCop reports no offenses on the file. Anchoring the whole pattern at `\A` is not enough: it stops removing the suffix and fails the two "when used as suffix" examples. The fix would also add examples for markers in the middle of a subject to `title_linting_spec.rb` and `base_linter_spec.rb`. I'd like to work on this.