The how-to-read class scan reads whole classes and skips comments

[johnnyt]

What

The info test "names every mark the hook draws" holds the classes assets/js/statifier_blocks_map.js draws to the how-to-read paragraph. It had two gaps:

• It matched an interpolation only right after the sb-map__ prefix. A class built as a stem plus an interpolated modifier (sb-map__edge--${edge.kind}) passed on its stem alone.
• It read classes out of the hook's comments, so a comment naming a class could fail the gate.

The scan is now one test helper, hook_classes/1:

• it cuts // and /* */ comments (a // after a : is kept, so the SVG namespace URL survives);
• it reads each class whole, with every ${...} in it read as ${} wherever it stands;
• an interpolation whose every string is empty or opens with a space (the block's ${container ? " sb-map__block--container" : ""}) appends other classes rather than extending the name, so it ends the name and the classes inside it are read on their own.

The hook's stem classes (end--${}, mark--${}, slot--${}) are listed in a new @modifiers table with the graph key their value comes from. A new test, "names every value the hook puts after a stem", holds every value the library documents give under each key to the existing variants table. A second new test, "reads no class out of a comment", pins the comment skip on a small source. The caption table's key is now written in the normalized form, ${}-caption.

Test-only: no lib/ or hook change, no changelog fragment (changelog.d/README.md excludes tests).

Sabotage

Each reverted from a copy, byte-equal, and recompiled before the next:

Mutation	Test that went red
hook: sb-map__edge--${edge.kind} added to the plain edge's template	names every mark the hook draws
hook: an unnamed plain class added to the fork's template	names every mark the hook draws
hook: sb-map__${edge.kind}-tag added to the plain edge's template	names every mark the hook draws
mark/1 in lib/statifier_blocks/map.ex answers "hourglass" for an await	names every value the hook puts after a stem (and the existing variants test)
hook_classes/1 with its comment cuts removed	reads no class out of a comment
hook_classes/1 reading every interpolation as part of the name	names every mark the hook draws

The first mutation run against the test as it stands on main left it green, which is the defect this fixes. Two existing sabotage notes that injected into a hook comment now read "template", the shape they were re-run in.

Gate

Full mix quality green on the committed tree: 3,953 of 3,953 tests, 95.3% coverage, dialyzer clean, ADR cites green.

Review

In-turn review: re-read the diff against the bead. The acceptance asks the scan to catch an interpolation anywhere in a class name, each stem covered by the variants table, and to skip comments, each pinned by a sabotage; the table above maps each part to a red test. Every sb-map__ occurrence in the hook was read by hand against the helper's reading: the plain classes it returns are the ones the old scan returned, and the built set is the caption plus the three stems. A slot style that no library document draws (the drafts shelf's tray) is not exercised by the new values test; renaming it left the suite green, so the sabotage row uses the await's mark instead.

Refs: sb-9wvq