Pair squeeze across <span> boundaries

punctuation

Every number below is re-measured on change. KLR is JIS plus one vert mapping (： stays upright), so horizontally the two are identical by construction — the vertical columns are where it differs. MOE is the interesting row: its 、。， are centred while ． is corner-attached, which is what trips Chromium's font-wide consistency gate.

Does text-spacing-trim: trim-start still collapse the space between two adjacent full-width punctuation glyphs when an inline element boundary sits between them — i.e. 》。 vs <span>》</span><span>。</span>?

Each row measures the rendered width of the highlighted .probe (the punctuation pair only, kept off the line start by a leading Han). natural = no trim (upper bound). same-node = trim, one text node. split-span = trim, each glyph in its own <span>. If split-span ≈ natural (i.e. ≫ same-node), the span boundary defeats the squeeze.

pair natural (px) same-node (px) split-span (px) inline-block (px) split keeps squeeze?

What kind of split breaks it? (》。)

Same pair, different container markup. The rule is not "inline-block" or "separate elements" — it is whether the two glyphs share an inline formatting context. Splitting them across atomic-inline boxes (inline-block / -flex / -grid) breaks the squeeze; plain inline spans (any nesting), or both glyphs in one box, keep it.

markup width (px) squeeze

Rendered specimens (horizontal)

Vertical (visual)

same-node
あ》。い
split-span
あ》。い
no trim
あ》。い