@markup-carve/pandoc-carve 0.1.6 → 0.1.7

package/README.md

@@ -20,7 +20,7 @@ npm install @markup-carve/pandoc-carve
 ```
 
 The CLI requires Pandoc 3.x on `PATH` for formats other than JSON. CI currently
-tests Pandoc 3.11. Pandoc is not bundled.
+tests Pandoc 3.11 and 3.10.2. Pandoc is not bundled.
 
 ## CLI
 
@@ -105,6 +105,13 @@ them. Highlights include:
 - raw target content only survives in matching writers;
 - comments and attribute placement require round-trip metadata.
 
+These limits are measured rather than asserted.
+[pandoc-format-fidelity](https://github.com/markup-carve/pandoc-format-fidelity) runs Carve
+source through this bridge and back for every writer it lists, and publishes the verdicts per
+format: `results/carve-rt.json` holds the Carve round-trip lane, and
+[the report](https://github.com/markup-carve/pandoc-format-fidelity/blob/main/docs/index.html)
+covers every pandoc writer and reader.
+
 Dangerous URL schemes such as `javascript:` are blanked and reported. The
 [URL policy](docs/reference.md#dangerous-url-schemes-are-blanked) documents the
 exact behavior. The [mapping and limitations reference](docs/reference.md#what-maps-to-what)

package/dist/diagnostics.js

@@ -10,7 +10,7 @@ const RULES = [
     { test: /^block: pandoc node/, code: 'unsupported-pandoc-block', class: 'unsupported' },
     { test: /^citation: .*typed locator/, code: 'citation-locator-flattened', class: 'normalized' },
     { test: /^citation: .*suppresses its author/, code: 'citation-mode-normalized', class: 'normalized' },
-    { test: /^Cite mixes /, code: 'citation-mode-normalized', class: 'normalized' },
+    { test: /^Cite mixes /, code: 'citation-mode-normalized', class: 'degraded' },
     { test: /^Cite mapped /, code: 'citation-bibliography-not-emitted', class: 'degraded' },
     { test: /^SmallCaps /, code: 'smallcaps-degraded', class: 'degraded' },
     { test: /^short caption:/, code: 'short-caption-source-unavailable', class: 'unsupported' },
@@ -40,16 +40,19 @@ const RULES = [
     { test: /^list-table: structure/, code: 'list-table-structure-degraded', class: 'degraded' },
     { test: /^list-table: the short caption/, code: 'list-table-short-caption-dropped', class: 'lossy' },
     { test: /^list-table: a body group's attributes/, code: 'list-table-body-attributes-dropped', class: 'lossy' },
-    { test: /^list-table: the table's .* body groups/, code: 'list-table-body-groups-merged', class: 'normalized' },
-    { test: /^list-table: the body groups disagree/, code: 'list-table-row-heads-normalized', class: 'normalized' },
+    { test: /^list-table: a row's attributes/, code: 'list-table-row-attributes-dropped', class: 'lossy' },
+    { test: /^list-table: the table's .* body groups/, code: 'list-table-body-groups-merged', class: 'degraded' },
+    { test: /^list-table: the body groups disagree/, code: 'list-table-row-heads-normalized', class: 'degraded' },
     { test: /^list-table: rowspan/, code: 'list-table-rowspan-clipped', class: 'lossy' },
     { test: /^table: a row header outside/, code: 'table-row-head-outside-leading-run', class: 'lossy' },
     { test: /^table: attributes on/, code: 'table-continuation-attributes-dropped', class: 'lossy' },
     { test: /^table: colspan continuation/, code: 'table-colspan-origin-missing', class: 'degraded' },
     { test: /^table: rowspan crossing/, code: 'table-rowspan-clipped', class: 'lossy' },
     { test: /^table: rowspan continuation/, code: 'table-rowspan-origin-missing', class: 'degraded' },
-    { test: /^table: .*padded with empty cells/, code: 'table-groups-normalized', class: 'normalized' },
+    { test: /^table: .*padded with empty cells/, code: 'table-groups-normalized', class: 'degraded' },
     { test: /^table: a cell holds block content/, code: 'table-groups-normalized', class: 'normalized' },
+    { test: /^table: the merged id/, code: 'table-wrapper-id-dropped', class: 'lossy' },
+    { test: /^table: a blank row has no pipe-table source spelling/, code: 'table-blank-row-spelled', class: 'normalized' },
     { test: /^table: .*converted with the implicit head\/body split/, code: 'table-row-groups-invalid', class: 'degraded' },
     { test: /^table: .*preserved in the Carve AST as `rowGroups`/, code: 'table-groups-flattened', class: 'degraded' },
     { test: /^table: a foot row's row header is dropped/, code: 'table-foot-row-header-dropped', class: 'lossy' },

package/dist/reverse.js

@@ -983,9 +983,8 @@ function table(ctx, c, captionOverride, shortCaptionOverride = null) {
     }));
     // Row-head columns no longer send a table to the list-table form: the pipe
     // table marks those cells directly (see `rowHeadCols` above), which is
-    // readable, needs no extension, and round-trips. Block cells remain the one
-    // reason to leave, because a Carve `table_cell` holds INLINES and there is
-    // no pipe form for a cell holding blocks at all.
+    // readable, needs no extension, and round-trips. Block cells need the
+    // list-table form, as do blank rows without attributes or alignment.
     //
     // A FOOT USED TO BE A SECOND REASON, AND IT WAS A REAL ONE AT THE TIME.
     // The clause read `|| (ctx.target === 'source' && footRaw.length > 0)`.
@@ -1005,7 +1004,25 @@ function table(ctx, c, captionOverride, shortCaptionOverride = null) {
     // (corpus 376, spec f074372), so at the time the pipe form genuinely could
     // not carry a foot and the list-table detour was the only way to keep it.
     // The limitation is gone, so the detour is too.
-    const useListTable = ctx.target === 'source' && hasBlockCells;
+    const occupied = allRaw.map(() => []);
+    const hasBlankRows = allRaw.some((row, r) => {
+        let col = 0;
+        let blank = nCols > 0;
+        for (const raw of row[1]) {
+            while (occupied[r][col])
+                col++;
+            const [attr, align, height, width, content] = raw;
+            if (width > 1 || height > 1 || content.length || fromAttr(attr) || align.t !== 'AlignDefault')
+                blank = false;
+            for (let down = 1; down < height && r + down < allRaw.length; down++) {
+                for (let across = 0; across < width; across++)
+                    occupied[r + down][col + across] = true;
+            }
+            col += width;
+        }
+        return blank && !occupied[r].some(Boolean);
+    });
+    const useListTable = ctx.target === 'source' && (hasBlockCells || hasBlankRows);
     const pending = allRaw.map(() => Array(nCols));
     const rows = [];
     // The pandoc blocks that sat at each grid position, kept alongside the
@@ -1093,6 +1110,7 @@ function table(ctx, c, captionOverride, shortCaptionOverride = null) {
             attrs: fromAttr(a),
             colAligns,
             colWidths: columns.map((column) => column.width),
+            blankRows: !hasBlockCells && hasBlankRows,
         });
     }
     const node = { type: 'table', rows };
@@ -1299,14 +1317,15 @@ function isInlineShaped(cellBlocks) {
  */
 function listTable(ctx, input) {
     const { rows, cellBlocksAt, headRows, footRows, bodies, caption, attrs, colAligns, colWidths } = input;
-    // One reason is left to be here, so the diagnostic states it rather than
-    // selecting between three strings two of which nothing could reach: row-head
-    // columns are marked on the cells and a foot is stated on the attribute
-    // line, both in the pipe form.
-    warn(ctx, 'table: a cell holds block content, which a pipe table cannot spell - emitted as a `::: list-table` (structure preserved)');
+    warn(ctx, input.blankRows
+        ? 'table: a blank row has no pipe-table source spelling - emitted as a `::: list-table` (structure preserved)'
+        : 'table: a cell holds block content, which a pipe table cannot spell - emitted as a `::: list-table` (structure preserved)');
     if (bodies.some((b) => b.attrs)) {
         warn(ctx, 'list-table: a body group\'s attributes are dropped - the extension has no body to hang them on');
     }
+    if (rows.some((row) => row.attrs)) {
+        warn(ctx, 'list-table: a row\'s attributes are dropped - the extension has no row-attribute spelling');
+    }
     if (bodies.slice(1).some((b) => b.headRows === 0)) {
         warn(ctx, `list-table: the table's ${bodies.length} body groups merge where a later body has no header row to mark its boundary`);
     }
@@ -1471,27 +1490,6 @@ function figure(ctx, c) {
             return [node];
         }
     }
-    if (single?.t === 'Table') {
-        // The wrapper and the Table collapse into ONE Carve node, so their
-        // attrs merge rather than the inner one silently winning: pandoc's
-        // readers put the label on the Figure, not on the Table it wraps, and
-        // dropping it took the id a `</#id>` resolves against with it. The
-        // outer id wins, classes union, key/values merge with the outer taking
-        // precedence.
-        //
-        // A table is the only host that collapses this way, because it is the
-        // only one Carve gives a caption of its own. Every other host keeps
-        // both nodes - the wrapper stays a `figure` and the host's own attrs
-        // ride on the target - which is what the Div arm below sorts out.
-        //
-        // It matters most for a §4c table PANEL, whose id is what resolves as
-        // the group's number plus a letter.
-        const node = table(ctx, single.c, caption, shortCaption);
-        const outer = fromAttr(a);
-        if (outer)
-            node.attrs = mergeCAttrs(node.attrs, outer);
-        return [node];
-    }
     // A HOST THAT CARRIES ITS OWN ATTRIBUTES ARRIVES INSIDE A DIV. Pandoc's
     // BlockQuote, Para and CodeBlock have no Attr slot, so `block()` in the
     // forward direction wraps an attributed one in a Div. A `div` is not a
@@ -1515,6 +1513,21 @@ function figure(ctx, c) {
             ]);
         }
     }
+    if (host?.t === 'Table') {
+        // A table caption lives on the table, including when a reader puts
+        // an alignment Div between it and the Figure. Outer attributes win.
+        const node = table(ctx, host.c, caption, shortCaption);
+        const outer = fromAttr(a);
+        for (const attrs of [hostAttrs, outer]) {
+            const current = node.attrs;
+            if (current?.id && attrs?.id && current.id !== attrs.id) {
+                warn(ctx, `table: the merged id "${current.id}" is dropped in favor of "${attrs.id}" - one Carve node has one id`);
+            }
+            if (attrs)
+                node.attrs = mergeCAttrs(current, attrs);
+        }
+        return [node];
+    }
     if (host && FIGURE_HOSTS.has(host.t)) {
         // A single-host `Figure` is an ordinary `figure` around that host -
         // the generic captioned wrapper of PART 9 §4b, which is what the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@markup-carve/pandoc-carve",
-  "version": "0.1.6",
+  "version": "0.1.7",
   "description": "Carve to Pandoc bridge: convert Carve markup to Pandoc's JSON AST and reach every pandoc output format (LaTeX, Typst, DOCX, PDF, ...)",
   "keywords": [
     "carve",