ContextQMD
Back to Plate

Void-like zero-width IME proofs need the real void spacer structure

docs/solutions/logic-errors/2026-04-04-void-like-zero-width-ime-proofs-need-the-real-void-spacer-structure.md

53.0.62.7 KB
Original Source

Void-like zero-width IME proofs need the real void spacer structure

What happened

After the slate-v2 line-break and inline-edge proofs turned green, the next behavior seam was the FEFF-backed void-like zero-width path.

The first proof said the path was dead in Chromium IME.

That conclusion was wrong.

The proof surface was not mirroring real void rendering.

What didn't work

The first v2 void proof put everything inside one non-editable wrapper:

  • the visible void chrome
  • the zero-width text leaf

That is not how real Slate voids render.

With that fake structure, Chromium never committed text and the root stayed at "\uFEFFvoid".

Treating that as a real policy conclusion would have been bullshit.

What fixed it

The proof surface had to match the real void seam from legacy Slate:

  • a data-slate-void="true" element
  • a non-editable content wrapper for the visible void chrome
  • a separate absolutely positioned data-slate-spacer sibling containing the zero-width text leaf

Once the v2 proof surface matched that structure, the FEFF-backed void-like path passed in Chromium:

  • IME committed すし
  • final Slate selection landed at 0.0:2|0.0:2

Why this works

Void behavior is not just “zero-width leaf next to some UI.”

The spacer is part of the contract. Legacy Slate routes selection and DOM behavior through a separate invisible spacer leaf, not through the visible non-editable void content.

If the proof collapses those roles into one wrapper, it is measuring a fake DOM shape and will happily invent fake policy limits.

Reusable rule

For void-like browser proofs:

  • do not build an ad hoc “looks close enough” wrapper
  • mirror the real void seam: non-editable content wrapper plus separate spacer leaf
  • only draw policy conclusions from proofs that clear that bar

The current honest slate-v2 zero-width split is:

  • line-break path: no FEFF works in Chromium with the DOM-owned interior
  • inline-edge path: FEFF-backed behavior is green in Chromium
  • void-like path: FEFF-backed behavior is also green once the proof uses the real void spacer structure
  • non-linebreak no-FEFF paths: still unproved and therefore still conservative