Naming convention: drop BEM entirely, or adopt it consistently? #33

Two reviewers have pointed in opposite directions on CSS class naming, and we're currently inconsistent in a way neither would endorse. The #16 lint series is landing the fixes as suggested in review — this issue is just to settle the convention so we stop rediscovering the question.

@aduth in #2, on "selector-class-pattern": null in .stylelintrc.json:

I assume this is overriding the BEM naming rule enforcement? Personally I think that could still be valuable to uphold.

@joedolson in #18:

I think we should change the remaining BEM style classes… The naming needs to be self-consistent.

Two things that aren't obvious

1. The Gutenberg convention isn't textbook BEM. It uses __ for child elements, but modifiers are not -- — they're separate is-/has- classes (.components-panel.is-opened). So badge--experiment is wrong under both readings; under Gutenberg it'd be .accessibility-lab-badge.is-experiment. We already do this correctly for .accessibility-lab-row.is-dependent.

2. Our linter currently enforces the opposite of Gutenberg. We extend @wordpress/stylelint-config, whose selector-class-pattern is hyphens-only and rejects __. Gutenberg uses a different package, @wordpress/stylelint-tools, which explicitly allows __ and -- and ships a suppressions file for grandfathering. Upholding the Gutenberg rule is a package switch, not a config tweak.

Options

A — Drop BEM, kebab-case everywhere. Matches @joedolson, already enforced by our linter for free, no tooling change. But diverges from the Gutenberg convention, and names get long with no block/element boundary (accessibility-lab-media-view-config-field-description).

B — Adopt Gutenberg properly: __ for descendants, is-/has- for modifiers, switch to @wordpress/stylelint-tools. Matches @aduth and upstream expectations. But it's a tooling change, and it means revisiting the src/settings classes a second time.

(@aduth's suppressions idea is a migration mechanism for either option, not a third answer.)

Scope

~30 occurrences either way, so cost isn't the deciding factor — the question is really whether we follow upstream Gutenberg or our current linter.

Whichever we pick, it should go in CONTRIBUTING.md, which documents PHP naming in detail but says nothing about CSS classes.