@graphty/graph-io 0.3.40 → 0.3.42

package/dist/csv.js

@@ -1,8 +1,8 @@
-import { b as plural, bU as DICT_SAMPLE_ROWS, ac as DictHeuristic, R as ROLE_TAKEN_CODE$1, I as ID_MERGED_CODE$1, m as DUPLICATE_EDGE_ID_CODE$1, n as DUPLICATE_NODE_CODE$1, t as MISSING_ID_CODE$1, s as MISSING_ENDPOINT_CODE$1, E as EMPTY_INPUT_CODE$1, al as resolveImportOptions, a8 as ImportReportBuilder, ao as reportSinkOptions, an as reportUnusedOptions, aq as DirectionResolver, ar as IdCoercer, am as throwIfAborted, O as OPTION_IGNORED_CODE, aK as parseWeightText, a0 as uniqueColumnName, ay as RENAMED_CODE, a1 as declareResolved, c as capabilities, L as LOSS, a as checkCapabilities, r as resolveExportOptions, p as pairFolding, bD as countMixedEdges, $ as agree, e as explicitWeights, b_ as otherFormatDialect, bS as joinListText, aG as formatInteger, aF as formatDecimal, aH as canonicalId, M as MIXED_DIRECTION_CODE, D as DIRECTION_FORCED_CODE, d as DIRECTION_REFUSED_CODE, S as SINK_OPTION_CODE, C as COLUMN_RENAMED_CODE, w as UNKNOWN_ENCODING_CODE, x as ENCODING_FALLBACK_CODE, y as INVALID_ENCODING_CODE, z as INVALID_UTF8_CODE, P as PRECISION_CODE, F as ELEMENT_ISSUE, G as INPUT_ISSUE } from "./chunks/export-B-aVCwyG.js";
-import { i as inferTextDtype, T as TextCellWriter, p as parseTextCell, W as WIDENING_UNSUPPORTED_CODE } from "./chunks/text-Bjfa1xYL.js";
+import { b as plural, bU as DICT_SAMPLE_ROWS, ac as DictHeuristic, R as ROLE_TAKEN_CODE$1, I as ID_MERGED_CODE$1, m as DUPLICATE_EDGE_ID_CODE$1, n as DUPLICATE_NODE_CODE$1, t as MISSING_ID_CODE$1, s as MISSING_ENDPOINT_CODE$1, E as EMPTY_INPUT_CODE$1, al as resolveImportOptions, a8 as ImportReportBuilder, ao as reportSinkOptions, an as reportUnusedOptions, aq as DirectionResolver, ar as IdCoercer, am as throwIfAborted, O as OPTION_IGNORED_CODE, aK as parseWeightText, a0 as uniqueColumnName, ay as RENAMED_CODE, a1 as declareResolved, c as capabilities, L as LOSS, a as checkCapabilities, r as resolveExportOptions, p as pairFolding, $ as agree, e as explicitWeights, b_ as otherFormatDialect, bD as countMixedEdges, bS as joinListText, aG as formatInteger, aF as formatDecimal, aH as canonicalId, M as MIXED_DIRECTION_CODE, D as DIRECTION_FORCED_CODE, d as DIRECTION_REFUSED_CODE, S as SINK_OPTION_CODE, C as COLUMN_RENAMED_CODE, w as UNKNOWN_ENCODING_CODE, x as ENCODING_FALLBACK_CODE, y as INVALID_ENCODING_CODE, z as INVALID_UTF8_CODE, P as PRECISION_CODE, F as ELEMENT_ISSUE, G as INPUT_ISSUE } from "./chunks/export-rKLeGQMo.js";
+import { i as inferTextDtype, T as TextCellWriter, p as parseTextCell, W as WIDENING_UNSUPPORTED_CODE } from "./chunks/text-DHOJXnED.js";
 import { GraphFormatError, INVALID_INDEX } from "@graphty/graph-format";
-import { s as stripLeadingComments, a as sniffNewline, b as sniffDelimiter, D as DELIMITER_CANDIDATES, c as checkRecordSyntax, C as CsvRecordReader, d as splitRecords, B as BAD_QUOTE_CODE, U as UNCLOSED_QUOTE_CODE } from "./chunks/records-r3SFHz11.js";
-import { o as quoteCsvCell } from "./chunks/escape-BJRLd3vJ.js";
+import { s as stripLeadingComments, a as sniffNewline, b as sniffDelimiter, D as DELIMITER_CANDIDATES, C as CsvRecordReader, d as splitRecords, c as checkRecordSyntax, B as BAD_QUOTE_CODE, U as UNCLOSED_QUOTE_CODE } from "./chunks/records-Cu4bxYOm.js";
+import { o as quoteCsvCell } from "./chunks/escape-B62ZGM_a.js";
 import { j as joinText, e as encodeChunks } from "./chunks/writer-DxjgHwSe.js";
 const SOURCE_NAMES = Object.freeze([
   "source",
@@ -336,6 +336,38 @@ function isPadded(text, quoted) {
 }
 function resolveCsvOptions(options) {
   const o = options ?? {};
+  checkDialectOptions(o);
+  for (const name of ["sourceColumn", "targetColumn", "idColumn", "labelColumn"]) {
+    checkColumnRef(name, o[name]);
+  }
+  if (o.typeColumn !== null) {
+    checkColumnRef("typeColumn", o.typeColumn);
+  }
+  if (o.table === "adjacency") {
+    checkAdjacencyOptions(o);
+  }
+  if (o.rowNumberIds !== void 0 && typeof o.rowNumberIds !== "boolean") {
+    throw new GraphFormatError("E_UNSUPPORTED", "option rowNumberIds: expected a boolean", {
+      option: "rowNumberIds",
+      found: o.rowNumberIds
+    });
+  }
+  return {
+    delimiter: o.delimiter ?? null,
+    decimalComma: o.decimal === ",",
+    header: o.header ?? "auto",
+    table: o.table ?? "auto",
+    sourceColumn: o.sourceColumn ?? null,
+    targetColumn: o.targetColumn ?? null,
+    typeColumn: o.typeColumn,
+    idColumn: o.idColumn ?? null,
+    labelColumn: o.labelColumn ?? null,
+    nodes: nodesInput(o.nodes),
+    rowNumberIds: o.rowNumberIds ?? false,
+    explicitDirected: typeof options?.defaultDirected === "boolean" ? options.defaultDirected : null
+  };
+}
+function checkDialectOptions(o) {
   if (o.delimiter !== void 0 && (typeof o.delimiter !== "string" || o.delimiter.length === 0)) {
     throw new GraphFormatError("E_UNSUPPORTED", "option delimiter: expected a non-empty string", {
       option: "delimiter",
@@ -369,48 +401,22 @@ function resolveCsvOptions(options) {
       found: o.table
     });
   }
-  for (const name of ["sourceColumn", "targetColumn", "idColumn", "labelColumn"]) {
-    checkColumnRef(name, o[name]);
-  }
-  if (o.typeColumn !== null) {
-    checkColumnRef("typeColumn", o.typeColumn);
-  }
-  if (o.table === "adjacency") {
-    for (const name of ["sourceColumn", "targetColumn", "typeColumn", "idColumn", "labelColumn"]) {
-      if (o[name] !== void 0) {
-        throw new GraphFormatError("E_UNSUPPORTED", `option ${name}: an adjacency table has no columns`, {
-          option: name,
-          found: o[name]
-        });
-      }
-    }
-    if (o.rowNumberIds === true) {
-      throw new GraphFormatError("E_UNSUPPORTED", "option rowNumberIds: an adjacency table names its nodes", {
-        option: "rowNumberIds",
-        found: o.rowNumberIds
+}
+function checkAdjacencyOptions(o) {
+  for (const name of ["sourceColumn", "targetColumn", "typeColumn", "idColumn", "labelColumn"]) {
+    if (o[name] !== void 0) {
+      throw new GraphFormatError("E_UNSUPPORTED", `option ${name}: an adjacency table has no columns`, {
+        option: name,
+        found: o[name]
       });
     }
   }
-  if (o.rowNumberIds !== void 0 && typeof o.rowNumberIds !== "boolean") {
-    throw new GraphFormatError("E_UNSUPPORTED", "option rowNumberIds: expected a boolean", {
+  if (o.rowNumberIds === true) {
+    throw new GraphFormatError("E_UNSUPPORTED", "option rowNumberIds: an adjacency table names its nodes", {
       option: "rowNumberIds",
       found: o.rowNumberIds
     });
   }
-  return {
-    delimiter: o.delimiter ?? null,
-    decimalComma: o.decimal === ",",
-    header: o.header ?? "auto",
-    table: o.table ?? "auto",
-    sourceColumn: o.sourceColumn ?? null,
-    targetColumn: o.targetColumn ?? null,
-    typeColumn: o.typeColumn,
-    idColumn: o.idColumn ?? null,
-    labelColumn: o.labelColumn ?? null,
-    nodes: nodesInput(o.nodes),
-    rowNumberIds: o.rowNumberIds ?? false,
-    explicitDirected: typeof options?.defaultDirected === "boolean" ? options.defaultDirected : null
-  };
 }
 function checkColumnRef(name, value) {
   if (value === void 0) {
@@ -433,6 +439,15 @@ function isBlank(text) {
 function isUnset(text, quoted) {
   return quoted !== true && isBlank(text);
 }
+function setNeighborCells(row, quoted) {
+  let neighbors = 0;
+  for (let k = 1; k < row.length; k++) {
+    if (!isUnset(row[k], quoted[k])) {
+      neighbors++;
+    }
+  }
+  return neighbors;
+}
 const ABORT_CHECK_INTERVAL = 64;
 function uniqueNames(names, report, line) {
   const seen = /* @__PURE__ */ new Set();
@@ -564,39 +579,11 @@ class TableReader {
     const firstLine = this.reader.line;
     const firstQuoted = this.reader.quoted.slice(0, firstRow.length);
     const pending = [];
-    let header;
-    let mode = this.state.csv.header;
-    if (mode === "auto" && this.kind === "adjacency") {
-      mode = false;
-    } else if (mode === "auto" && this.rowNumbers) {
-      mode = true;
-    }
-    if (mode === "auto") {
-      const second = await iterator.next();
-      const secondRow = second.done ? null : second.value;
-      const secondLine = this.reader.line;
-      const secondQuoted = this.reader.quoted.slice(0, secondRow?.length ?? 0);
-      header = looksLikeHeader(firstRow, secondRow) || this.namesExplicitColumn(firstRow);
-      if (!header) {
-        pending.push({ row: firstRow, quoted: firstQuoted, line: firstLine });
-      }
-      if (secondRow !== null) {
-        pending.push({ row: secondRow, quoted: secondQuoted, line: secondLine });
-      }
-    } else {
-      header = mode;
-      if (!header) {
-        pending.push({ row: firstRow, quoted: firstQuoted, line: firstLine });
-      }
-    }
-    if (this.rowNumbers && !header) {
-      report.warning(
-        "unsupported",
-        OPTION_IGNORED_CODE,
-        "option rowNumberIds ignored: a headerless node table takes its ids from column 0",
-        { line: firstLine, element: "rowNumberIds" }
-      );
-    }
+    const header = await this.decideHeader(
+      iterator,
+      { row: firstRow, quoted: firstQuoted, line: firstLine },
+      pending
+    );
     const names = header ? uniqueNames(headerNames(firstRow), report, firstLine) : positionalNames(firstRow.length);
     const last = firstRow.length - 1;
     const optionalLast = header && last > 0 && isUnset(firstRow[last], firstQuoted[last]);
@@ -626,6 +613,70 @@ class TableReader {
     for (const { row, quoted, line } of pending) {
       this.processRow(row, quoted, line);
     }
+    await this.readRemainingRows(iterator);
+    if (this.dataRows === 0 && header) {
+      report.warning("missing-value", NO_DATA_ROWS_CODE, "the table has a header and no data rows", {
+        line: firstLine
+      });
+    }
+    this.reportReaderFindings();
+  }
+  /**
+   * Decide whether the first row is the header, reading the second row when the header option is
+   * "auto", and queue the rows read so far that are data.
+   * @param iterator - the record iterator, past the first row
+   * @param first - the first row
+   * @param pending - receives the data rows read so far, in order
+   * @returns whether the first row is the header
+   */
+  async decideHeader(iterator, first, pending) {
+    const mode = this.headerMode();
+    let second = null;
+    let header;
+    if (mode === "auto") {
+      const next = await iterator.next();
+      second = next.done ? null : { row: next.value, quoted: this.reader.quoted.slice(0, next.value.length), line: this.reader.line };
+      header = looksLikeHeader(first.row, second?.row ?? null) || this.namesExplicitColumn(first.row);
+    } else {
+      header = mode;
+    }
+    if (!header) {
+      pending.push(first);
+    }
+    if (second !== null) {
+      pending.push(second);
+    }
+    if (this.rowNumbers && !header) {
+      this.state.report.warning(
+        "unsupported",
+        OPTION_IGNORED_CODE,
+        "option rowNumberIds ignored: a headerless node table takes its ids from column 0",
+        { line: first.line, element: "rowNumberIds" }
+      );
+    }
+    return header;
+  }
+  /**
+   * The header option as this table reads it: an adjacency table has no header unless the caller
+   * says so (its rows vary in width); a table numbered by rowNumberIds has one, since a headerless
+   * table takes its ids from column 0.
+   * @returns true, false or "auto"
+   */
+  headerMode() {
+    const mode = this.state.csv.header;
+    if (mode === "auto" && this.kind === "adjacency") {
+      return false;
+    }
+    if (mode === "auto" && this.rowNumbers) {
+      return true;
+    }
+    return mode;
+  }
+  /**
+   * Push every row after the header and the rows read to decide it, then finish the column writers.
+   * @param iterator - the record iterator
+   */
+  async readRemainingRows(iterator) {
     const { signal } = this.state.common;
     let sinceCheck = 0;
     for (; ; ) {
@@ -642,12 +693,6 @@ class TableReader {
     for (const writer of this.writers) {
       writer?.finish();
     }
-    if (this.dataRows === 0 && header) {
-      report.warning("missing-value", NO_DATA_ROWS_CODE, "the table has a header and no data rows", {
-        line: firstLine
-      });
-    }
-    this.reportReaderFindings();
   }
   /**
    * Whether a row is Excel's delimiter directive (`sep=;`) under a delimiter the caller gave (a
@@ -1199,12 +1244,7 @@ class TableReader {
     const { where } = this;
     where.line = line;
     where.element = null;
-    let neighbors = 0;
-    for (let k = 1; k < row.length; k++) {
-      if (!isUnset(row[k], quoted[k])) {
-        neighbors++;
-      }
-    }
+    const neighbors = setNeighborCells(row, quoted);
     if (isUnset(row[0], quoted[0])) {
       report.error("missing-value", MISSING_ID_CODE, `blank node cell`, { line });
       counts.skippedNodes++;
@@ -1926,19 +1966,23 @@ function nodeTableNote(names, table, notes) {
     );
   }
 }
+function idTextCollisions(ids) {
+  const seen = /* @__PURE__ */ new Set();
+  let collisions = 0;
+  for (let i = 0; i < ids.size; i++) {
+    const text = String(ids.idOf(i));
+    if (seen.has(text)) {
+      collisions++;
+    } else {
+      seen.add(text);
+    }
+  }
+  return collisions;
+}
 function idNotes(snapshot, values, note) {
   const { ids } = snapshot;
   if (ids.kind === "mixed") {
-    const seen = /* @__PURE__ */ new Set();
-    let collisions = 0;
-    for (let i = 0; i < ids.size; i++) {
-      const text = String(ids.idOf(i));
-      if (seen.has(text)) {
-        collisions++;
-      } else {
-        seen.add(text);
-      }
-    }
+    const collisions = idTextCollisions(ids);
     if (collisions > 0) {
       note(
         CSV_LOSS.ID_TEXT_COLLISION,
@@ -1983,6 +2027,44 @@ function planExport(snapshot, options, values) {
       edgeRows.push(e);
     }
   }
+  directionNotes(snapshot, csv, edgeRows.length, note);
+  if (folding.mutualCount > 0 && csv.table !== "nodes") {
+    note(
+      CSV_LOSS.MUTUAL_EXPANDED,
+      `${folding.mutualCount} mutual pair${plural(folding.mutualCount)} ${agree(folding.mutualCount, "is", "are")} written as two directed rows; the mutual mark is lost`,
+      null,
+      folding.mutualCount
+    );
+  }
+  if (csv.table === "edges") {
+    noteNodeCoverage(snapshot, edgeRows, note);
+  }
+  const weights = explicitWeights(snapshot);
+  const { edgeId, edgeLabel, edgeColumns } = planEdgeColumns(snapshot, csv, edgeRows, values, note);
+  const { nodeLabel, nodeColumns } = planNodeColumns(snapshot, csv, values, note);
+  if (csv.table !== "nodes") {
+    nodeTableNote(
+      [...nodeLabel === null ? [] : [nodeLabel.meta.name], ...nodeColumns.map((c) => c.column.meta.name)],
+      csv.table,
+      notes
+    );
+  }
+  return {
+    csv,
+    common,
+    notes,
+    edgeRows,
+    folding,
+    edgeId,
+    edgeLabel,
+    weights,
+    edgeColumns,
+    nodeLabel,
+    nodeColumns,
+    idText
+  };
+}
+function directionNotes(snapshot, csv, edgeRowCount, note) {
   if (csv.table !== "nodes" && (csv.dialect.type === null || csv.table === "adjacency")) {
     const where = csv.table === "adjacency" ? "an adjacency table" : "the generic dialect";
     const mixed = countMixedEdges(snapshot);
@@ -2001,7 +2083,7 @@ function planExport(snapshot, options, values) {
         mixed
       );
     }
-  } else if (csv.table === "edges" && !snapshot.directed && edgeRows.length === 0) {
+  } else if (csv.table === "edges" && !snapshot.directed && edgeRowCount === 0) {
     note(
       CSV_LOSS.DIRECTION_DROPPED,
       "the direction lives in the Type cell of each edge row; an undirected graph without edges reads back as directed",
@@ -2009,18 +2091,8 @@ function planExport(snapshot, options, values) {
       0
     );
   }
-  if (folding.mutualCount > 0 && csv.table !== "nodes") {
-    note(
-      CSV_LOSS.MUTUAL_EXPANDED,
-      `${folding.mutualCount} mutual pair${plural(folding.mutualCount)} ${agree(folding.mutualCount, "is", "are")} written as two directed rows; the mutual mark is lost`,
-      null,
-      folding.mutualCount
-    );
-  }
-  if (csv.table === "edges") {
-    noteNodeCoverage(snapshot, edgeRows, note);
-  }
-  const weights = explicitWeights(snapshot);
+}
+function planEdgeColumns(snapshot, csv, edgeRows, values, note) {
   const edgeId = snapshot.edges.byRole("id");
   const edgeLabel = snapshot.edges.byRole("label");
   const edgeReserved = new Set([csv.dialect.source, csv.dialect.target, csv.dialect.weight].map(lower));
@@ -2053,6 +2125,9 @@ function planExport(snapshot, options, values) {
       checkValues(column, "edge", edgeRows, note);
     }
   }
+  return { edgeId, edgeLabel, edgeColumns };
+}
+function planNodeColumns(snapshot, csv, values, note) {
   const nodeLabel = snapshot.nodes.byRole("label");
   const nodeReserved = /* @__PURE__ */ new Set([lower(csv.dialect.id)]);
   const nodeRows = [];
@@ -2069,34 +2144,13 @@ function planExport(snapshot, options, values) {
     nodeLabel,
     csv.table === "nodes" ? note : null
   );
-  if (csv.table === "nodes") {
-    if (values) {
-      checkRoleColumn(nodeLabel, "label", LABEL_NAMES, "node", nodeRows, note);
-      for (const { column } of nodeColumns) {
-        checkValues(column, "node", nodeRows, note);
-      }
+  if (csv.table === "nodes" && values) {
+    checkRoleColumn(nodeLabel, "label", LABEL_NAMES, "node", nodeRows, note);
+    for (const { column } of nodeColumns) {
+      checkValues(column, "node", nodeRows, note);
     }
-  } else {
-    nodeTableNote(
-      [...nodeLabel === null ? [] : [nodeLabel.meta.name], ...nodeColumns.map((c) => c.column.meta.name)],
-      csv.table,
-      notes
-    );
   }
-  return {
-    csv,
-    common,
-    notes,
-    edgeRows,
-    folding,
-    edgeId,
-    edgeLabel,
-    weights,
-    edgeColumns,
-    nodeLabel,
-    nodeColumns,
-    idText
-  };
+  return { nodeLabel, nodeColumns };
 }
 function noteNodeCoverage(snapshot, edgeRows, note) {
   const { nodeCount } = snapshot;
@@ -2152,39 +2206,44 @@ function attributeColumns(table, domain, reserved, idColumn, labelColumn, note)
     if (role !== null && SKIPPED_ROLES.has(role)) {
       continue;
     }
-    const low = lower(name);
     const set = column.length - column.nullCount;
-    const isIdName = domain === "edge" ? findColumn([name], EDGE_ID_NAMES) >= 0 : false;
-    if (reserved.has(low) || isIdName || idColumn !== null && lower(idColumn.meta.name) === low) {
-      note?.(
-        CSV_LOSS.RESERVED_NAME,
-        `${domain} column "${name}" is not written: the name is reserved for the ${isIdName ? "edge id" : low} column`,
-        name,
-        set
-      );
+    if (isReservedName(name, domain, reserved, idColumn, set, note)) {
       continue;
     }
-    if (findColumn([name], LABEL_NAMES) >= 0) {
-      if (labelColumn !== null) {
-        note?.(
-          CSV_LOSS.RESERVED_NAME,
-          `${domain} column "${name}" is not written: the name is reserved for the label column`,
-          name,
-          set
-        );
-        continue;
-      }
-      note?.(
-        CSV_LOSS.ROLE_ASSUMED,
-        `${domain} column "${name}" reads back with the label role (string)`,
-        name,
-        set
-      );
+    if (findColumn([name], LABEL_NAMES) >= 0 && !labelNameWritten(name, domain, labelColumn, set, note)) {
+      continue;
     }
     out.push({ column, header: name });
   }
   return out;
 }
+function isReservedName(name, domain, reserved, idColumn, set, note) {
+  const low = lower(name);
+  const isIdName = domain === "edge" ? findColumn([name], EDGE_ID_NAMES) >= 0 : false;
+  if (reserved.has(low) || isIdName || idColumn !== null && lower(idColumn.meta.name) === low) {
+    note?.(
+      CSV_LOSS.RESERVED_NAME,
+      `${domain} column "${name}" is not written: the name is reserved for the ${isIdName ? "edge id" : low} column`,
+      name,
+      set
+    );
+    return true;
+  }
+  return false;
+}
+function labelNameWritten(name, domain, labelColumn, set, note) {
+  if (labelColumn !== null) {
+    note?.(
+      CSV_LOSS.RESERVED_NAME,
+      `${domain} column "${name}" is not written: the name is reserved for the label column`,
+      name,
+      set
+    );
+    return false;
+  }
+  note?.(CSV_LOSS.ROLE_ASSUMED, `${domain} column "${name}" reads back with the label role (string)`, name, set);
+  return true;
+}
 function checkRoleColumn(column, role, names, domain, rows, note) {
   if (column === null) {
     return;
@@ -2223,20 +2282,7 @@ function checkValues(column, domain, rows, note, roleColumn = false) {
   const { name, dtype } = column.meta;
   const label = `${domain} column "${name}"`;
   if (column.dtype === "f32" || column.dtype === "f64") {
-    let nonFinite = 0;
-    const { components } = column.meta;
-    const { data } = column;
-    for (const row of rows) {
-      if (!column.isSet(row)) {
-        continue;
-      }
-      for (let k = 0; k < components; k++) {
-        if (!Number.isFinite(data[row * components + k])) {
-          nonFinite++;
-          break;
-        }
-      }
-    }
+    const nonFinite = countNonFinite(column, rows);
     if (nonFinite > 0) {
       note(
         CSV_LOSS.NONFINITE,
@@ -2250,15 +2296,50 @@ function checkValues(column, domain, rows, note, roleColumn = false) {
   if (dtype !== "string" && dtype !== "dict") {
     return;
   }
+  const scan = scanText(column, rows);
+  if (roleColumn || scan.widest < 0) {
+    return;
+  }
+  const readsAsDict = scan.candidate && scan.heuristic.decide() === "dict";
+  if (scan.widest < WIDENING.string && !readsAsDict) {
+    note(CSV_LOSS.TEXT_INFERRED, `${label}: every value reads back as ${TEXT_DTYPES[scan.widest]}`, name, scan.set);
+    return;
+  }
+  if (dtype === "dict" && !readsAsDict) {
+    note(
+      CSV_LOSS.STORAGE_CLASS_CHANGED,
+      `${label}: reads back as string (cardinality too high for a dict)`,
+      name,
+      null
+    );
+  } else if (dtype === "string" && readsAsDict) {
+    note(CSV_LOSS.STORAGE_CLASS_CHANGED, `${label}: reads back as dict (low cardinality)`, name, null);
+  }
+}
+function countNonFinite(column, rows) {
+  let nonFinite = 0;
+  const { components } = column.meta;
+  const { data } = column;
+  for (const row of rows) {
+    if (!column.isSet(row)) {
+      continue;
+    }
+    for (let k = 0; k < components; k++) {
+      if (!Number.isFinite(data[row * components + k])) {
+        nonFinite++;
+        break;
+      }
+    }
+  }
+  return nonFinite;
+}
+function scanText(column, rows) {
   let widest = -1;
   let candidate = true;
   let set = 0;
   const heuristic = new DictHeuristic(DICT_SAMPLE_ROWS);
   for (const row of rows) {
-    if (!column.isSet(row)) {
-      continue;
-    }
-    const text = column.value(row);
+    const text = column.isSet(row) ? column.value(row) : null;
     if (typeof text !== "string") {
       continue;
     }
@@ -2266,30 +2347,11 @@ function checkValues(column, domain, rows, note, roleColumn = false) {
     const kind = inferTextDtype(text);
     widest = Math.max(widest, WIDENING[kind]);
     if (!heuristic.decided) {
-      if (kind !== "string") {
-        candidate = false;
-      }
+      candidate &&= kind === "string";
       heuristic.observe(text);
     }
   }
-  if (roleColumn || widest < 0) {
-    return;
-  }
-  const readsAsDict = candidate && heuristic.decide() === "dict";
-  if (widest < WIDENING.string && !readsAsDict) {
-    note(CSV_LOSS.TEXT_INFERRED, `${label}: every value reads back as ${TEXT_DTYPES[widest]}`, name, set);
-    return;
-  }
-  if (dtype === "dict" && !readsAsDict) {
-    note(
-      CSV_LOSS.STORAGE_CLASS_CHANGED,
-      `${label}: reads back as string (cardinality too high for a dict)`,
-      name,
-      null
-    );
-  } else if (dtype === "string" && readsAsDict) {
-    note(CSV_LOSS.STORAGE_CLASS_CHANGED, `${label}: reads back as dict (low cardinality)`, name, null);
-  }
+  return { widest, candidate, set, heuristic };
 }
 function typeText(snapshot, plan, e) {
   if (!snapshot.directed || !plan.folding.sourceDirected(e)) {
@@ -2303,6 +2365,14 @@ function neighbourCell(id, weight) {
   }
   return id.includes(":") ? `${id}:` : id;
 }
+function sourceRunEnd(src, edgeRows, r) {
+  const source = src[edgeRows[r]];
+  let end = r;
+  while (end < edgeRows.length && src[edgeRows[end]] === source) {
+    end++;
+  }
+  return end;
+}
 function* adjacencyLines(snapshot, plan) {
   const { delimiter, newline } = plan.csv;
   const forced = (text) => `"${text.replaceAll('"', '""')}"`;
@@ -2329,10 +2399,7 @@ function* adjacencyLines(snapshot, plan) {
   let r = 0;
   while (r < edgeRows.length) {
     const source = list.src[edgeRows[r]];
-    let end = r;
-    while (end < edgeRows.length && list.src[edgeRows[end]] === source) {
-      end++;
-    }
+    const end = sourceRunEnd(list.src, edgeRows, r);
     const fresh = [];
     let highest = -1;
     let inOrder = true;
@@ -2366,59 +2433,54 @@ function* adjacencyLines(snapshot, plan) {
   yield* singles(snapshot.nodeCount - 1);
 }
 function* lines(snapshot, plan) {
-  const { csv } = plan;
-  if (csv.table === "adjacency") {
+  if (plan.csv.table === "adjacency") {
     yield* adjacencyLines(snapshot, plan);
-    return;
+  } else if (plan.csv.table === "nodes") {
+    yield* nodeLines(snapshot, plan);
+  } else {
+    yield* edgeLines(snapshot, plan);
   }
-  const { delimiter, newline } = csv;
+}
+function cellWriters(delimiter) {
   const quote = (text) => quoteCsvCell(text, delimiter);
   const cell = (column, row) => {
     const text = cellText(column, row);
     return text === null ? "" : quote(text);
   };
-  if (csv.table === "nodes") {
-    const headers2 = [csv.dialect.id];
-    if (plan.nodeLabel !== null) {
-      headers2.push(plan.nodeLabel.meta.name);
-    }
-    for (const { header } of plan.nodeColumns) {
-      headers2.push(header);
-    }
-    if (csv.header) {
-      yield headers2.map(quote).join(delimiter) + newline;
-    }
-    for (let i = 0; i < snapshot.nodeCount; i++) {
-      const cells = [quote(plan.idText(i))];
-      if (plan.nodeLabel !== null) {
-        cells.push(cell(plan.nodeLabel, i));
-      }
-      for (const { column } of plan.nodeColumns) {
-        cells.push(cell(column, i));
-      }
-      yield cells.join(delimiter) + newline;
-    }
-    return;
-  }
-  const headers = [csv.dialect.source, csv.dialect.target];
-  if (csv.dialect.type !== null) {
-    headers.push(csv.dialect.type);
-  }
-  if (plan.edgeId !== null) {
-    headers.push(plan.edgeId.meta.name);
-  }
-  if (plan.edgeLabel !== null) {
-    headers.push(plan.edgeLabel.meta.name);
-  }
-  if (plan.weights.weighted) {
-    headers.push(csv.dialect.weight);
+  return { quote, cell };
+}
+function* nodeLines(snapshot, plan) {
+  const { csv } = plan;
+  const { delimiter, newline } = csv;
+  const { quote, cell } = cellWriters(delimiter);
+  const headers = [csv.dialect.id];
+  if (plan.nodeLabel !== null) {
+    headers.push(plan.nodeLabel.meta.name);
   }
-  for (const { header } of plan.edgeColumns) {
+  for (const { header } of plan.nodeColumns) {
     headers.push(header);
   }
   if (csv.header) {
     yield headers.map(quote).join(delimiter) + newline;
   }
+  for (let i = 0; i < snapshot.nodeCount; i++) {
+    const cells = [quote(plan.idText(i))];
+    if (plan.nodeLabel !== null) {
+      cells.push(cell(plan.nodeLabel, i));
+    }
+    for (const { column } of plan.nodeColumns) {
+      cells.push(cell(column, i));
+    }
+    yield cells.join(delimiter) + newline;
+  }
+}
+function* edgeLines(snapshot, plan) {
+  const { csv } = plan;
+  const { delimiter, newline } = csv;
+  const { quote, cell } = cellWriters(delimiter);
+  if (csv.header) {
+    yield edgeHeaders(plan).map(quote).join(delimiter) + newline;
+  }
   const list = snapshot.edgeList();
   for (const e of plan.edgeRows) {
     const cells = [quote(plan.idText(list.src[e])), quote(plan.idText(list.dst[e]))];
@@ -2440,6 +2502,26 @@ function* lines(snapshot, plan) {
     yield cells.join(delimiter) + newline;
   }
 }
+function edgeHeaders(plan) {
+  const { dialect } = plan.csv;
+  const headers = [dialect.source, dialect.target];
+  if (dialect.type !== null) {
+    headers.push(dialect.type);
+  }
+  if (plan.edgeId !== null) {
+    headers.push(plan.edgeId.meta.name);
+  }
+  if (plan.edgeLabel !== null) {
+    headers.push(plan.edgeLabel.meta.name);
+  }
+  if (plan.weights.weighted) {
+    headers.push(dialect.weight);
+  }
+  for (const { header } of plan.edgeColumns) {
+    headers.push(header);
+  }
+  return headers;
+}
 function prepare(snapshot, options) {
   const plan = planExport(snapshot, options, false);
   const collision = plan.notes.find((n) => n.code === CSV_LOSS.ID_TEXT_COLLISION);