Beagle SCM

DIS-029: MARK renders bare-indent divs as raw <p> text and merges across the depth boundary

The StrictMark block model is (INDENT|QUOTE)* LIST? LEAF? and "absent a leaf, a paragraph is implied" (StrictMark §Block markup); a bare 4-space indent "opens a generic container (a div)" (§Block markup), and "block-stack depth changes drive nesting". The MARK HTML renderer honors none of this for plain indented text. In mark_blocks (mark/MARK.c:281) the paragraph branch (:393-404) computes depth = MKDTIndentDepth(linef) (:360) and the marker end mend (:362) but then ignores both: it never emits a <div> wrapper, and it feeds the FULL line linec — leading spaces included — to mark_inline, so the indent reaches HTML as literal gutter spaces inside a paragraph. Worse, in_para persists across a depth drop, so an indented line and the following non-indented line collapse into ONE <p>, erasing the structural boundary the depth change is supposed to enforce. depth/mend are consumed only for ULIST/QUOTE markers (:364-391), never for the INDENT (div) container. See DIS-001 (the other open StrictMark/store-layout gap), GET.

Issues

The renderer treats indentation as cosmetic whitespace, not as the div container the grammar defines.

• Repro (build/bin/mark on a 3-line .mkd):

    #   Header
        Indented
    Non indented

• Current output — no <div>, and the two body lines share one <p>:

    <h1>Header</h1>
    <p>
        Indented
    Non indented
    </p>

• Spec-correct output — indent = div, depth drop closes it:

    <h1>Header</h1>
    <div><p>Indented</p></div>
    <p>Non indented</p>

• Root cause 1 (no container): mark/MARK.c:393-404 never opens/closes a <div> per depth; MKDTIndentDepth is read (:360) then dropped on the floor for the paragraph case.
• Root cause 2 (leaked indent): mark_inline(c, linec) (:403) passes the whole line incl. the 4-space prefix; it should start at mend / strip depth*4 like the heading branch strips its prefix (:336-340).
• Root cause 3 (cross-depth merge): in_para (:286, :396-398) is not closed when depth changes, so a depth-1 line and a depth-0 line fold into one paragraph; a depth change must close the open leaf, same as a blank line does (:312-317).
• The block lexer MKDTLexer already carves the indent off as an R structural prefix (dog/tok/MKDT.c:316-328); only the MARK renderer discards that meaning, so the fix is renderer-local.

Blockers

None. Disjoint from the wire/store DIS tickets; the change is confined to mark/MARK.c block rendering.

Planned

Make MARK render a bare 4-space indent as a <div> nesting level and treat a depth change as a leaf boundary; keep list/quote handling unchanged.

• Repro first (CLAUDE §17): add a table case to mark/test/MARK.c asserting the 3-line input above yields <h1> + <div><p>Indented</p></div> + <p>Non indented</p> — FAILS pre-fix on all three counts. Add nested-depth ( → <div><div>…) and depth-restore cases.
• Track an open-div depth in mark_blocks alongside in_para/in_list/in_quote: when depth rises, emit N <div>; when it falls (incl. to 0), close the open leaf and the surplus <div>s before rendering the line. Mirror the close on blank lines and at EOF (:407-410).
• Strip the indent before inline: feed mark_inline the slice from mend (or linec advanced by depth*4), not the raw linec, so gutter spaces never reach HTML.
• Reconcile the StrictMark §Headers "header may span lines" wording with §Block markup depth rules so the div reading is unambiguous (a deeper line opens a div; header-spanning is same-depth) — case-by-case per strictmark-wiki-reorg.
• Verify no regression on the existing wiki pages (make the wiki, diff rendered HTML); run the full mark suite; update mark/INDEX.md if behavior notes change.