Kinneyzhang 83352c4571 feat!: add source-clean live buffer layout

Replace the physical ekp-region renderer with ekp-buffer text-property projection backed by shared semantic layout plans and stable live-edit transactions.

Optimize the C-backed resize hot path, and add deterministic performance, ERT, GUI, fuzz, release, and documentation coverage for task017 through task032.

BREAKING CHANGE: require ekp-buffer instead of ekp-region and rename module-owned ekp-region-* settings to ekp-buffer-*.

2026-07-30 01:07:25 +08:00

2.4 KiB

Raw Blame History

Keep Static Edge Policy out of Live Editor State

Context

The source-clean renderer had two interaction regressions after its initial quality gate.

First, every reprojection copied the old mark position and restored it with set-mark. Emacs treats mark position and mark-active as separate state, but set-mark changes both. A showcase width key therefore turned an old inactive mark into a highlighted region.

Second, the live renderer correctly skipped glue justification on the point-containing line, then reused static paragraph-edge cleanup. The KP plan excludes leading/trailing whitespace from its content bounds, so that cleanup assigned display "" to a newly typed edge space. The source character existed immediately, but the user saw it only after typing the next glyph.

Decision

Editor state is restored by its actual owners:

set the existing mark marker's position directly, then restore mark-active independently;
retain static edge hiding for committed lines;
never hide source-edge whitespace on the active live line;
when the first planned line becomes active, clear projection from the span's source start so previously hidden leading whitespace becomes natural too.

No showcase-specific selection workaround and no space self-insert advice is added. Both symptoms are fixed in ekp-buffer.el, where reprojection and live-line presentation are owned.

Why the Initial Tests Missed It

The live typing test inserted " x" as one operation. By assertion time, the space was interior and no longer eligible for edge cleanup. The GUI matrix also tested an intentionally active region, but not an existing inactive mark.

The replacement gates stop at the temporal boundary:

inspect state immediately after one space, before another glyph;
cover leading/trailing space, tab, CJK, deletion-exposed whitespace, consecutive spaces, newline, yank, and undo;
test inactive and active mark states separately;
record the real showcase width command and command-loop space insertion in a clean full-screen GUI run.

Consequences

One space now moves point/cursor immediately and carries no EKP replacing display. Backspace cannot re-hide the exposed whitespace. Width changes preserve point, mark position, and region activation exactly.

Committed static paragraphs still hide stripped edge whitespace, so this correction does not change completed KP layout or the source-clean representation contract.

2.4 KiB Raw Blame History