Give every fragment's content root its canonical id

[ctgnz]

Closes #80.

6,211 fragments across the two trees now hold exactly one content element: a <g> carrying the name its family goes by — main, mod1, mod2, frame, frame_overlay, echelon, oca, eng_bar, amplifier, hqtffd. All lowercase, so there is no exception to remember. 649 files changed here, 717 in hallux.

Injecting fragments into the generated classes needs one rule for finding a fragment's content. With canonical ids that rule is "the element with the expected id"; without them it falls back on guessing from position, which fails outright for the fragments whose content was a loose <path> at the root.

Four kinds of change

	files
renames	~190	mostly an editor's duplicate suffix (main_1_, mod1_2_, HQTFFD_1_, main-7), plus case (Main, Symbol, Mod1, Octagon), two plain typos (mian, ocatgon), and 24 that had drifted to another family's name — mod2 fragments whose group was called mod1 or main
wraps	445	the name was on a bare <rect>, <path>, <text> or <line>; each is now inside a group, so one rule covers every family rather than one for OCA and Engagement and another for everything else
merges	70	content in more than one root element: a duplicate main-7 group, a <path> drawn beside an empty main, or an anonymous group that was plainly the octagon guide and had lost its id
the guide	56	Octagon / ocatgon spelled one way

Two decisions taken along the way, both yours: amplifier lowercase (52 files went from Amplifier rather than the 8 going the other way, because every other id is lowercase), and hqtffd losing its capitals for the same reason.

Redundant nested groups

A second commit takes out 564 groups across 492 files — every one a <g> with no attributes at all and no siblings, so a container around exactly one thing, styling nothing and transforming nothing. Editor leftovers from a copy or an ungroup; Dismounted 27110107 wrapped its single polygon in five of them.

That one cannot change the drawing by construction: an element's effective attributes are its own plus every ancestor's, and an ancestor with no attributes contributes nothing; with no siblings there is no paint order to disturb either. The rendered comparison agrees — 234 of 234 identical — and every measured bound comes back the same.

Nothing is drawn differently — which took three goes

Worth recording, because each was a way to silently lose a drawing:

• Hoisting a group's children drops what it was imposing on them. A main-7 group with transform="translate(0,0.4)" lost the shift. A group carrying anything but an id now moves intact.
• Paint order is document order. Appending everything to the end of the canonical group put a Frame that was drawn behind an icon in front of it. What sat before the content now goes in at the top, what sat after goes in at the bottom.
• "Empty" means an empty <g>, never a leaf. <line id="main"> has no children, was read as an empty container, and was dropped — taking the drawing with it.

The first two showed up only in the rendered comparison, not in any count.

Verification

Three independent checks, since SvgFingerprint can't guard structural moves:

• Every drawing element compared before and after with its ancestors' attributes folded in and display:none subtrees left out, so a move between groups carrying nothing but an id is invisible to the comparison. 649 of 649 identical here, 717 of 717 in hallux (a further 71 hallux files differ from its HEAD because of the uncommitted Echelon amplifiers are half text and half geometry - make them all geometry #68/Control Measure fragments: separate the drawn content from the construction scaffolding #78/Frame fragments: give each a single content root #79 work).
• FragmentMeasurer re-run over both models: every measured bound identical, no diff at all.
• The 2,972 shared fragments are still byte-identical across the two trees.

Normaliser clean, mvn verify green.

Left alone

13 fragments, listed with the reasoning on the issue: five that draw nothing at all, and eight where merging would impose a transform, style or display on the content being moved — a change to the drawing rather than to the structure, so the pass refused rather than guess.

Next

FragmentShapeChecker currently covers free canvas fragments only, and its javadoc says this is where it grows. With these names settled it can cover every family.

The matching changes to jmsfx-hallux are in the hallux working tree, uncommitted.

🤖 Generated with Claude Code