25351425939

Committed 05 May 2026 12:36AM UTC coverage: 58.829% (+0.03%) from 58.796%

Build # 25351425939

Build Type

Pull #3414

github

Committed by

web-flow

Commit Message

Merge 9769e89aa into 9c7af408f

Pull Request Pull Request #3414: feat(processors): auto-detect delimiter for CSV/TSV/DSV files

Coverage Stats

6977 of 14243 branches covered (48.99%)

Branch coverage included in aggregate %.

17 of 17 new or added lines in 2 files covered. (100.0%)

51 existing lines in 1 file now uncovered.

14338 of 21989 relevant lines covered (65.21%)

79.6 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

51.42

/src/processors/src/data-processor.ts

// SPDX-License-Identifier: MIT
// Copyright contributors to the kepler.gl project

import * as arrow from 'apache-arrow';
import {csvParseRows, tsvParseRows, dsvFormat} from 'd3-dsv';
import {DATA_TYPES as AnalyzerDATA_TYPES} from 'type-analyzer';
import normalize from '@mapbox/geojson-normalize';
import {parseSync} from '@loaders.gl/core';
import {ArrowTable} from '@loaders.gl/schema';
import {WKBLoader} from '@loaders.gl/wkt';

import {
  ALL_FIELD_TYPES,
  DATASET_FORMATS,
  GEOARROW_EXTENSIONS,
  GEOARROW_METADATA_KEY,
  GUIDES_FILE_FORMAT_DOC
} from '@kepler.gl/constants';
import {ProcessorResult, Field} from '@kepler.gl/types';
import {
  arrowDataTypeToAnalyzerDataType,
  arrowDataTypeToFieldType,
  hasOwnProperty,
  isPlainObject
} from '@kepler.gl/utils';
import {
  analyzerTypeToFieldType,
  getSampleForTypeAnalyze,
  getSampleForTypeAnalyzeArrow,
  getFieldsFromData,
  h3IsValid,
  notNullorUndefined,
  toArray
} from '@kepler.gl/common-utils';
import {KeplerGlSchema, ParsedDataset, SavedMap, LoadedMap} from '@kepler.gl/schemas';
import {Feature} from '@deck.gl-community/editable-layers';

// if any of these value occurs in csv, parse it to null;
// const CSV_NULLS = ['', 'null', 'NULL', 'Null', 'NaN', '/N'];
// matches empty string
export const CSV_NULLS = /^(null|NULL|Null|NaN|\/N||)$/;

const SUPPORTED_DELIMITERS = [',', '\t', ';', '|'] as const;

/**
 * Detect the delimiter used in a DSV string by checking the first line.
 * Returns the delimiter that produces the most columns (minimum 2).
 * Falls back to comma if no delimiter produces multiple columns.
 */
export function detectDelimiter(rawData: string): string {
  const newlineIdx = rawData.indexOf('\n');
  const firstLine = newlineIdx === -1 ? rawData : rawData.slice(0, newlineIdx);
  if (!firstLine) return ',';

  let bestDelimiter = ',';
  let bestCount = 1;

  for (const delimiter of SUPPORTED_DELIMITERS) {
    const parseRows =
      delimiter === ','
        ? csvParseRows
        : delimiter === '\t'
          ? tsvParseRows
          : dsvFormat(delimiter).parseRows;
    const parsed = parseRows(firstLine);
    const count = parsed[0]?.length || 0;
    if (count > bestCount) {
      bestCount = count;
      bestDelimiter = delimiter;
    }
  }

  return bestDelimiter;
}

function tryParseJsonString(str) {
  try {
    return JSON.parse(str);
  } catch (e) {
    return null;
  }
}

export const PARSE_FIELD_VALUE_FROM_STRING = {
  [ALL_FIELD_TYPES.boolean]: {
    valid: (d: unknown): boolean => typeof d === 'boolean',
    parse: (d: unknown): boolean => {
      const s = String(d).toLowerCase();
      return s === 'true' || s === 'yes' || s === '1';
    }
  },
  [ALL_FIELD_TYPES.integer]: {
    // @ts-ignore
    valid: (d: unknown): boolean => parseInt(d, 10) === d,
    // @ts-ignore
    parse: (d: unknown): number => parseInt(d, 10)
  },
  [ALL_FIELD_TYPES.timestamp]: {
    valid: (d: unknown, field: Field): boolean =>
      ['x', 'X'].includes(field.format) ? typeof d === 'number' : typeof d === 'string',
    parse: (d: any, field: Field) => (['x', 'X'].includes(field.format) ? Number(d) : d)
  },
  [ALL_FIELD_TYPES.real]: {
    // @ts-ignore
    valid: (d: unknown): boolean => parseFloat(d) === d,
    // Note this will result in NaN for some string
    parse: parseFloat
  },
  [ALL_FIELD_TYPES.object]: {
    valid: isPlainObject,
    parse: tryParseJsonString
  },

  [ALL_FIELD_TYPES.array]: {
    valid: Array.isArray,
    parse: tryParseJsonString
  },

  [ALL_FIELD_TYPES.h3]: {
    valid: d => h3IsValid(d),
    parse: d => d
  }
};

/**
 * Process csv data, output a data object with `{fields: [], rows: []}`.
 * The data object can be wrapped in a `dataset` and pass to [`addDataToMap`](../actions/actions.md#adddatatomap)
 * @param rawData raw csv string
 * @returns data object `{fields: [], rows: []}` can be passed to addDataToMaps
 * @public
 * @example
 * import {processCsvData} from '@kepler.gl/processors';
 *
 * const testData = `gps_data.utc_timestamp,gps_data.lat,gps_data.lng,gps_data.types,epoch,has_result,id,time,begintrip_ts_utc,begintrip_ts_local,date
 * 2016-09-17 00:09:55,29.9900937,31.2590542,driver_analytics,1472688000000,False,1,2016-09-23T00:00:00.000Z,2016-10-01 09:41:39+00:00,2016-10-01 09:41:39+00:00,2016-09-23
 * 2016-09-17 00:10:56,29.9927699,31.2461142,driver_analytics,1472688000000,False,2,2016-09-23T00:00:00.000Z,2016-10-01 09:46:37+00:00,2016-10-01 16:46:37+00:00,2016-09-23
 * 2016-09-17 00:11:56,29.9907261,31.2312742,driver_analytics,1472688000000,False,3,2016-09-23T00:00:00.000Z,,,2016-09-23
 * 2016-09-17 00:12:58,29.9870074,31.2175827,driver_analytics,1472688000000,False,4,2016-09-23T00:00:00.000Z,,,2016-09-23`
 *
 * const dataset = {
 *  info: {id: 'test_data', label: 'My Csv'},
 *  data: processCsvData(testData)
 * };
 *
 * dispatch(addDataToMap({
 *  datasets: [dataset],
 *  options: {centerMap: true, readOnly: true}
 * }));
 */
export function processCsvData(rawData: unknown[][] | string, header?: string[]): ProcessorResult {
  let rows: unknown[][] | undefined;
  let headerRow: string[] | undefined;

  if (typeof rawData === 'string') {
    const delimiter = detectDelimiter(rawData);
    const parseRows =
      delimiter === ','
        ? csvParseRows
        : delimiter === '\t'
          ? tsvParseRows
          : dsvFormat(delimiter).parseRows;
    const parsedRows: string[][] = parseRows(rawData);

    if (!Array.isArray(parsedRows) || parsedRows.length < 2) {
      // looks like an empty file, throw error to be catch
      throw new Error('process Csv Data Failed: CSV is empty');
    }
    headerRow = parsedRows[0];
    rows = parsedRows.slice(1);
  } else if (Array.isArray(rawData) && rawData.length) {
    rows = rawData;
    headerRow = header;

    if (!Array.isArray(headerRow)) {
      // if data is passed in as array of rows and missing header
      // assume first row is header
      // @ts-ignore
      headerRow = rawData[0];
      rows = rawData.slice(1);
    }
  }

  if (!rows || !headerRow) {
    throw new Error('invalid input passed to processCsvData');
  }

  // here we assume the csv file that people uploaded will have first row
  // as name of the column

  cleanUpFalsyCsvValue(rows);
  // No need to run type detection on every data point
  // here we get a list of none null values to run analyze on
  const sample = getSampleForTypeAnalyze({fields: headerRow, rows});
  const fields = getFieldsFromData(sample, headerRow);
  const parsedRows = parseRowsByFields(rows, fields);

  return {fields, rows: parsedRows};
}

/**
 * Parse rows of csv by analyzed field types. So that `'1'` -> `1`, `'True'` -> `true`
 * @param rows
 * @param fields
 */
export function parseRowsByFields(rows: any[][], fields: Field[]) {
  // Edit rows in place
  const geojsonFieldIdx = fields.findIndex(f => f.name === '_geojson');
  fields.forEach(parseCsvRowsByFieldType.bind(null, rows, geojsonFieldIdx));

  return rows;
}

/**
 * Convert falsy value in csv including `'', 'null', 'NULL', 'Null', 'NaN'` to `null`,
 * so that type-analyzer won't detect it as string
 *
 * @param rows
 */
function cleanUpFalsyCsvValue(rows: unknown[][]): void {
  const re = new RegExp(CSV_NULLS, 'g');
  for (let i = 0; i < rows.length; i++) {
    for (let j = 0; j < rows[i].length; j++) {
      // analyzer will set any fields to 'string' if there are empty values
      // which will be parsed as '' by d3.csv
      // here we parse empty data as null
      // TODO: create warning when deltect `CSV_NULLS` in the data
      if (typeof rows[i][j] === 'string' && (rows[i][j] as string).match(re)) {
        rows[i][j] = null;
      }
    }
  }
}

/**
 * Process uploaded csv file to parse value by field type
 *
 * @param rows
 * @param geoFieldIdx field index
 * @param field
 * @param i
 */
export function parseCsvRowsByFieldType(
  rows: unknown[][],
  geoFieldIdx: number,
  field: Field,
  i: number
): void {
  const parser = PARSE_FIELD_VALUE_FROM_STRING[field.type];
  if (parser) {
    // check first not null value of it's already parsed
    const first = rows.find(r => notNullorUndefined(r[i]));
    if (!first || parser.valid(first[i], field)) {
      return;
    }
    rows.forEach(row => {
      // parse string value based on field type
      if (row[i] !== null) {
        row[i] = parser.parse(row[i], field);
        if (
          geoFieldIdx > -1 &&
          isPlainObject(row[geoFieldIdx]) &&
          // @ts-ignore
          hasOwnProperty(row[geoFieldIdx], 'properties')
        ) {
          // @ts-ignore
          row[geoFieldIdx].properties[field.name] = row[i];
        }
      }
    });
  }
}

/* eslint-enable complexity */

/**
 * Process data where each row is an object, output can be passed to [`addDataToMap`](../actions/actions.md#adddatatomap)
 * NOTE: This function may mutate input.
 * @param rawData an array of row object, each object should have the same number of keys
 * @returns dataset containing `fields` and `rows`
 * @public
 * @example
 * import {addDataToMap} from '@kepler.gl/actions';
 * import {processRowObject} from '@kepler.gl/processors';
 *
 * const data = [
 *  {lat: 31.27, lng: 127.56, value: 3},
 *  {lat: 31.22, lng: 126.26, value: 1}
 * ];
 *
 * dispatch(addDataToMap({
 *  datasets: {
 *    info: {label: 'My Data', id: 'my_data'},
 *    data: processRowObject(data)
 *  }
 * }));
 */
export function processRowObject(rawData: unknown[]): ProcessorResult {
  if (!Array.isArray(rawData)) {
    return null;
  } else if (!rawData.length) {
    // data is empty
    return {
      fields: [],
      rows: []
    };
  }

  const firstRow = rawData[0] as Record<string, unknown>;
  const keys = Object.keys(firstRow); // [lat, lng, value]
  const rows = rawData.map(d => keys.map(key => (d as Record<string, unknown>)[key])); // [[31.27, 127.56, 3]]

  // row object can still contain values like `Null` or `N/A`
  cleanUpFalsyCsvValue(rows);

  return processCsvData(rows, keys);
}

/**
 * Process GeoJSON [`FeatureCollection`](http://wiki.geojson.org/GeoJSON_draft_version_6#FeatureCollection),
 * output a data object with `{fields: [], rows: []}`.
 * The data object can be wrapped in a `dataset` and passed to [`addDataToMap`](../actions/actions.md#adddatatomap)
 * NOTE: This function may mutate input.
 *
 * @param rawData raw geojson feature collection
 * @returns dataset containing `fields` and `rows`
 * @public
 * @example
 * import {addDataToMap} from '@kepler.gl/actions';
 * import {processGeojson} from '@kepler.gl/processors';
 *
 * const geojson = {
 *         "type" : "FeatureCollection",
 *         "features" : [{
 *                 "type" : "Feature",
 *                 "properties" : {
 *                         "capacity" : "10",
 *                         "type" : "U-Rack"
 *                 },
 *                 "geometry" : {
 *                         "type" : "Point",
 *                         "coordinates" : [ -71.073283, 42.417500 ]
 *                 }
 *         }]
 * };
 *
 * dispatch(addDataToMap({
 *  datasets: {
 *    info: {
 *      label: 'Sample Taxi Trips in New York City',
 *      id: 'test_trip_data'
 *    },
 *    data: processGeojson(geojson)
 *  }
 * }));
 */
export function processGeojson(rawData: unknown): ProcessorResult {
  const normalizedGeojson = normalize(rawData);

  if (!normalizedGeojson || !Array.isArray(normalizedGeojson.features)) {
    throw new Error(
      `Read File Failed: File is not a valid GeoJSON. Read more about [supported file format](${GUIDES_FILE_FORMAT_DOC})`
    );
  }

  // getting all feature fields
  const allDataRows: Array<{_geojson: Feature} & keyof Feature> = [];
  for (let i = 0; i < normalizedGeojson.features.length; i++) {
    const f = normalizedGeojson.features[i];
    if (f.geometry) {
      allDataRows.push({
        // add feature to _geojson field
        _geojson: f,
        ...(f.properties || {})
      });
    }
  }
  // get all the field
  const fields = allDataRows.reduce<string[]>((accu, curr) => {
    Object.keys(curr).forEach(key => {
      if (!accu.includes(key)) {
        accu.push(key);
      }
    });
    return accu;
  }, []);

  // make sure each feature has exact same fields
  allDataRows.forEach(d => {
    fields.forEach(f => {
      if (!(f in d)) {
        d[f] = null;
        if (d._geojson.properties) {
          d._geojson.properties[f] = null;
        }
      }
    });
  });

  return processRowObject(allDataRows);
}

/**
 * Process saved kepler.gl json to be pass to [`addDataToMap`](../actions/actions.md#adddatatomap).
 * The json object should contain `datasets` and `config`.
 * @param rawData
 * @param schema
 * @returns datasets and config `{datasets: {}, config: {}}`
 * @public
 * @example
 * import {addDataToMap} from '@kepler.gl/actions';
 * import {processKeplerglJSON} from '@kepler.gl/processors';
 *
 * dispatch(addDataToMap(processKeplerglJSON(keplerGlJson)));
 */
export function processKeplerglJSON(rawData: SavedMap, schema = KeplerGlSchema): LoadedMap | null {
  return rawData ? schema.load(rawData.datasets, rawData.config) : null;
}

/**
 * Parse a single or an array of datasets saved using kepler.gl schema
 * @param rawData
 * @param schema
 */
export function processKeplerglDataset(
  rawData: object | object[],
  schema = KeplerGlSchema
): ParsedDataset | ParsedDataset[] | null {
  if (!rawData) {
    return null;
  }

  const results = schema.parseSavedData(toArray(rawData));
  if (!results) {
    return null;
  }
  return Array.isArray(rawData) ? results : results[0];
}

/**
 * Parse arrow table and return a dataset
 *
 * @param arrowTable ArrowTable to parse, see loaders.gl/schema
 * @returns dataset containing `fields` and `rows` or null
 */
export function processArrowTable(arrowTable: ArrowTable): ProcessorResult | null {
  // @ts-ignore - Unknown data type causing build failures
  return processArrowBatches(arrowTable.data.batches);
}

/**
 * Extracts GeoArrow metadata from an Apache Arrow table schema.
 * For geoparquet files geoarrow metadata isn't present in fields, so extract extra info from schema.
 * @param table The Apache Arrow table to extract metadata from.
 * @returns An object mapping column names to their GeoArrow encoding type.
 * @throws Logs an error message if parsing of metadata fails.
 */
export function getGeoArrowMetadataFromSchema(table: arrow.Table): Record<string, string> {
  const geoArrowMetadata: Record<string, string> = {};
  try {
    const geoString = table.schema.metadata?.get('geo');
    if (geoString) {
      const parsedGeoString = JSON.parse(geoString);
      if (parsedGeoString.columns) {
        Object.keys(parsedGeoString.columns).forEach(columnName => {
          const columnData = parsedGeoString.columns[columnName];
          if (columnData?.encoding === 'WKB') {
            geoArrowMetadata[columnName] = GEOARROW_EXTENSIONS.WKB;
          }
          // TODO potentially there are other types but no datasets to test
        });
      }
    }
  } catch (error) {
    console.error('An error during arrow table schema metadata parsing');
  }
  return geoArrowMetadata;
}

/**
 * Converts an Apache Arrow table schema into an array of Kepler.gl field objects.
 * @param table The Apache Arrow table whose schema needs to be converted.
 * @param fieldTypeSuggestions Optional mapping of field names to suggested field types.
 * @returns An array of field objects suitable for Kepler.gl.
 */
export function arrowSchemaToFields(
  table: arrow.Table,
  fieldTypeSuggestions: Record<string, string> = {}
): Field[] {
  const headerRow = table.schema.fields.map(f => f.name);
  const sample = getSampleForTypeAnalyzeArrow(table, headerRow);
  const keplerFields = getFieldsFromData(sample, headerRow);
  const geoArrowMetadata = getGeoArrowMetadataFromSchema(table);

  return table.schema.fields.map((field: arrow.Field, fieldIndex: number) => {
    let type = arrowDataTypeToFieldType(field.type);
    let analyzerType = arrowDataTypeToAnalyzerDataType(field.type);
    let format = '';

    const fieldTypeSuggestion = fieldTypeSuggestions[field.name];
    const keplerField = keplerFields[fieldIndex];

    // geometry fields produced by DuckDB's st_asgeojson()
    if (fieldTypeSuggestion === 'JSON') {
      type = ALL_FIELD_TYPES.geojson;
      analyzerType = AnalyzerDATA_TYPES.GEOMETRY_FROM_STRING;
    } else if (
      fieldTypeSuggestion === 'GEOMETRY' ||
      field.metadata.get(GEOARROW_METADATA_KEY)?.startsWith('geoarrow')
    ) {
      type = ALL_FIELD_TYPES.geoarrow;
      analyzerType = AnalyzerDATA_TYPES.GEOMETRY;
    } else if (geoArrowMetadata[field.name]) {
      type = ALL_FIELD_TYPES.geoarrow;
      analyzerType = AnalyzerDATA_TYPES.GEOMETRY;
      field.metadata?.set(GEOARROW_METADATA_KEY, geoArrowMetadata[field.name]);
    } else if (fieldTypeSuggestion === 'BLOB') {
      // When arrow wkb column saved to DuckDB as BLOB without any metadata, then queried back
      try {
        const data = table.getChildAt(fieldIndex)?.get(0);
        if (data) {
          const binaryGeo = parseSync(data, WKBLoader);
          if (binaryGeo) {
            type = ALL_FIELD_TYPES.geoarrow;
            analyzerType = AnalyzerDATA_TYPES.GEOMETRY;
            field.metadata?.set(GEOARROW_METADATA_KEY, GEOARROW_EXTENSIONS.WKB);
          }
        }
      } catch (error) {
        // ignore, not WKB
      }
    } else if (
      fieldTypeSuggestion === 'VARCHAR' &&
      (keplerField.analyzerType === AnalyzerDATA_TYPES.GEOMETRY ||
        keplerField.analyzerType === AnalyzerDATA_TYPES.GEOMETRY_FROM_STRING)
    ) {
      // When wkb/wkt was saved as varchar in DuckDB
      type = keplerField.type;
      analyzerType = keplerField.analyzerType;
      format = keplerField.format;
    } else if (fieldTypeSuggestion === 'VARCHAR' && keplerField.type === ALL_FIELD_TYPES.h3) {
      // when kepler detected h3 column using getFieldsFromData(), set type to h3 and analyzerType to H3
      type = ALL_FIELD_TYPES.h3;
      analyzerType = keplerField.analyzerType;
    } else {
      // TODO should we use Kepler getFieldsFromData instead
      // of arrowDataTypeToFieldType for all fields?
      if (keplerField.type === ALL_FIELD_TYPES.timestamp) {
        type = keplerField.type;
        analyzerType = keplerField.analyzerType;
        format = keplerField.format;
      }
    }

    return {
      ...field,
      name: field.name,
      id: field.name,
      displayName: field.name,
      format: format,
      fieldIdx: fieldIndex,
      type,
      analyzerType,
      valueAccessor: (dc: any) => d => {
        return dc.valueAt(d.index, fieldIndex);
      },
      metadata: field.metadata
    };
  });
}

const CAST_BIGINTS = false;

/**
 * Cast 64-bit integer Arrow columns (Int64, Uint64) to Float64 to avoid BigInt values
 * that are incompatible with d3 scales, sorting, and other numeric operations.
 * Mirrors the DuckDB approach of casting BIGINT/UBIGINT to DOUBLE.
 */
function castBigIntColumnsToFloat64(arrowTable: arrow.Table): arrow.Table {
  if (!CAST_BIGINTS) {
    return arrowTable;
  }

  const needsCast = arrowTable.schema.fields.some(
    f => arrow.DataType.isInt(f.type) && f.type.bitWidth === 64
  );
  if (!needsCast) {
    return arrowTable;
  }

  const newColumns: Record<string, arrow.Vector> = {};
  for (let i = 0; i < arrowTable.numCols; i++) {
    const field = arrowTable.schema.fields[i];
    const col = arrowTable.getChildAt(i)!;
    if (arrow.DataType.isInt(field.type) && field.type.bitWidth === 64) {
      const float64Array = new Float64Array(col.length);
      for (let j = 0; j < col.length; j++) {
        const val = col.get(j);
        float64Array[j] = val === null ? NaN : Number(val);
      }
      newColumns[field.name] = arrow.makeVector(float64Array);
    } else {
      newColumns[field.name] = col;
    }
  }
  return new arrow.Table(newColumns);
}

/**
 * Parse arrow batches returned from parseInBatches()
 *
 * @param arrowTable the arrow table to parse
 * @returns dataset containing `fields` and `rows` or null
 */
export function processArrowBatches(arrowBatches: arrow.RecordBatch[]): ProcessorResult | null {
  if (arrowBatches.length === 0) {
    return null;
  }
  const arrowTable = castBigIntColumnsToFloat64(new arrow.Table(arrowBatches));
  const fields = arrowSchemaToFields(arrowTable);

  const cols = [...Array(arrowTable.numCols).keys()].map(i => arrowTable.getChildAt(i));

  // return empty rows and use raw arrow table to construct column-wise data container
  return {
    fields,
    rows: [],
    cols,
    metadata: arrowTable.schema.metadata,
    // Save original arrow schema, for better ingestion into DuckDB.
    // TODO consider returning arrowTable in cols, not an array of Vectors from arrowTable.
    arrowSchema: arrowTable.schema
  };
}

export const DATASET_HANDLERS = {
  [DATASET_FORMATS.row]: processRowObject,
  [DATASET_FORMATS.geojson]: processGeojson,
  [DATASET_FORMATS.csv]: processCsvData,
  [DATASET_FORMATS.arrow]: processArrowTable,
  [DATASET_FORMATS.keplergl]: processKeplerglDataset
};

export const Processors: {
  processGeojson: typeof processGeojson;
  processCsvData: typeof processCsvData;
  processArrowTable: typeof processArrowTable;
  processArrowBatches: typeof processArrowBatches;
  processRowObject: typeof processRowObject;
  processKeplerglJSON: typeof processKeplerglJSON;
  processKeplerglDataset: typeof processKeplerglDataset;
  analyzerTypeToFieldType: typeof analyzerTypeToFieldType;
  getFieldsFromData: typeof getFieldsFromData;
  parseCsvRowsByFieldType: typeof parseCsvRowsByFieldType;
} = {
  processGeojson,
  processCsvData,
  processArrowTable,
  processArrowBatches,
  processRowObject,
  processKeplerglJSON,
  processKeplerglDataset,
  analyzerTypeToFieldType,
  getFieldsFromData,
  parseCsvRowsByFieldType
};

keplergl / kepler.gl / 25351425939

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous