Fold the style groups away, and join a divider's parts into one figure

[ZombieHDGaming]

Two things in the Credit Designer, plus a formatting fix the CI asked for.

---

1. The style groups fold away

A StyleEditor is the tallest thing in the section editor — a font, a size, a fill with its stops, an outline, a shadow, an alignment — and a bridged row with subtitles stacks five of them between the placement rows and the entry table. That is exactly the scroll the folding groups were added to end; the style groups were simply left out of that first pass.

All five now fold:

group	was	now
Text Style	QGroupBox	CollapsibleGroup
Separate Style For The Right Column / Subtitle	checkable QGroupBox	checkable CollapsibleGroup
Separate Colour For The Bridge / Artwork	checkable QGroupBox	checkable CollapsibleGroup
Style For The Left Subtitle	QGroupBox	CollapsibleGroup
Style For The Right Subtitle	QGroupBox	CollapsibleGroup

Entries and Divider Pieces are deliberately untouched: they are the tables that already take the pane's leftover height, and folding the thing you came to the pane to edit is not a convenience.

The checkbox CollapsibleGroup was written to avoid

Its original note argued a checkbox on a group reads as switching the group off, which is what the checkable groups in this editor already mean — so it offered only a disclosure triangle. Two of the style groups need both readings at once, so the class now carries both, side by side rather than one standing in for the other:

• the checkbox says whether the style applies — unchanged document semantics, unchanged useSecondaryStyle / useBridgeStyle round trip;
• the triangle beside it says only whether it is on screen.

Neither moves the other. Switched off, a group greys its settings where they are, exactly as the checkable QGroupBox it replaces did. Separate signals (toggled, expandedChanged) mean folding a group never marks the document dirty. setHeaderToolTip came with it: a tooltip set on the group reached only the strip of it no child covered, which on an expanded group is almost nowhere.

---

2. A divider's parts can be joined into one figure

Every join in a Section Divider was a gap floored at zero, so a composed divider read as a row of separate marks however carefully its shapes were picked. Two settings answer that, and they compose:

• The gaps reach below zero.dividerGap and dividerPieceGap stop holding two parts apart and push them into each other instead. Not a second setting with a name of its own — a cap overlapping its rule by six pixels is the same edit as one clearing it by six, and one spin box running through zero says so. A join between two pieces of a stack is bounded at half the narrower of them, so pieces are pushed together but never through each other.
• dividerConnect runs the rule the whole way through. One unbroken arm from the middle of one cap to the middle of the other, passing behind the centre rather than breaking either side of it, so a diamond sits on the line. The middle of the outermost piece rather than the stack's outer edge, because that point is inside every cap's silhouette whatever the cap is — run to the edge and the rule pokes a blunt nose out of a tapering arrowhead.

The new dividers-joined scene shows all three states: held apart, run through, and closed up by hand.

What made it cheap

No new painting machinery. Every tinted part of a divider already goes into one silhouette and is inked once, so an overlap unions seamlessly rather than showing a seam, a doubled outline or a restarted gradient — the mechanism that was already giving you one outline around the whole divider.

Two things did need care:

• Arms are now placed before the stacks rather than after, so a piece is drawn over the rule it sits on. Invisible for tinted artwork, but a custom picture left in its own colours is painted straight to the strip, and a rule across the front of it is not the ornament anybody placed.
• The clamped joins are computed once, in the measure pass, and carried on MeasuredStack::gaps to the place pass. Two passes working the bound out separately would be two chances to disagree about a width the section below is positioned from.

Whatever a join is set to, the arms stay inside the section's own box.

Compatibility

dividerConnect defaults off and is absent from every existing document, so every roll draws exactly as it did. The loader still tells a stored zero apart from a missing key for both gaps, and now bounds them on the negative side at kMaxDividerJoin — the same figure the designer's spin boxes offer.

---

3. A formatting commit, and why it is here

The third commit is nothing but clang-format output. The Check Formatting 🔍 job installs obsproject/tools/clang-format@19 and runs it over every file a pull request changes, and six of the files this branch touches were already non-conforming under 19.1.1 on master:

file	violations on origin/master	added by this branch
src/model/CreditsModel.cpp	5	0
src/render/StripRenderer.cpp	20	0
src/ui/SectionEditor.cpp	25	0
tests/harness/Fixtures.cpp	2	0
tests/suites/Dividers.cpp	3	0
tests/suites/Persistence.cpp	1	0

Because the check only looks at changed files, master never trips it — the failure appears the moment any branch touches one of these. So the drift is cleared here rather than left to fail every future PR that goes near them. Every hunk is the formatter's own output: line wrapping and argument alignment, plus one spot in SectionEditor.cpp where two statements had ended up sharing a line separated by a tab. Nothing changes what any of it does.

It is a separate commit so it can be read (or dropped) independently of the two changes above.

---

Testing

• cmake -S . -B build -DENABLE_TESTS=ON && cmake --build build — plugin and test binary build clean, no new warnings.
• ./build/tests/closing-time-tests — 71 suites, 4558 checks, 0 failures (was 68 / 4498 before), unchanged by the formatting commit.
• Three new render suites and extended persistence coverage:
  • divider_connected_rule — a divider held apart has columns with no ink in them; the same divider joined has none, at the same width and height, and the end gap makes no difference once it is not consulted.
  • divider_negative_join — measured on a centre stack with no arm, where the ink is the stack's width: a positive gap opens both joins by exactly what was asked, and an unbounded overlap takes back exactly one piece width across the two joins rather than however much was typed.
  • divider_join_stays_inside — a join dragged to its far end leaves no ink outside the section's box.
  • Persistence — divider_connect round-trips, absent means off, and a hostile -1e9 gap off a file is bounded on the negative side rather than rejected for being negative.
• clang-format 19.1.1 (the version build-aux/run-clang-format requires) reports every changed file clean.

The section editor itself has no offscreen coverage — it needs a window — so the UI behaviour in part 1 was verified by reading, not by a test. Worth a look in a real OBS build before merge.