@loaders.gl/parquet 3.3.0 → 3.4.0-alpha.1

package/dist/parquetjs/parser/parquet-envelope-reader.js

@@ -0,0 +1,136 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ParquetEnvelopeReader = void 0;
+const constants_1 = require("../../constants");
+const parquet_thrift_1 = require("../parquet-thrift");
+const read_utils_1 = require("../utils/read-utils");
+const decoders_1 = require("./decoders");
+const DEFAULT_DICTIONARY_SIZE = 1e6;
+/**
+ * The parquet envelope reader allows direct, unbuffered access to the individual
+ * sections of the parquet file, namely the header, footer and the row groups.
+ * This class is intended for advanced/internal users; if you just want to retrieve
+ * rows from a parquet file use the ParquetReader instead
+ */
+class ParquetEnvelopeReader {
+    static async openBuffer(buffer) {
+        const readFn = (position, length) => Promise.resolve(buffer.slice(position, position + length));
+        const closeFn = () => Promise.resolve();
+        return new ParquetEnvelopeReader(readFn, closeFn, buffer.length);
+    }
+    constructor(read, close, fileSize, options) {
+        this.read = read;
+        this.close = close;
+        this.fileSize = fileSize;
+        this.defaultDictionarySize = options?.defaultDictionarySize || DEFAULT_DICTIONARY_SIZE;
+    }
+    async readHeader() {
+        const buffer = await this.read(0, constants_1.PARQUET_MAGIC.length);
+        const magic = buffer.toString();
+        switch (magic) {
+            case constants_1.PARQUET_MAGIC:
+                break;
+            case constants_1.PARQUET_MAGIC_ENCRYPTED:
+                throw new Error('Encrypted parquet file not supported');
+            default:
+                throw new Error(`Invalid parquet file (magic=${magic})`);
+        }
+    }
+    async readRowGroup(schema, rowGroup, columnList) {
+        const buffer = {
+            rowCount: Number(rowGroup.num_rows),
+            columnData: {}
+        };
+        for (const colChunk of rowGroup.columns) {
+            const colMetadata = colChunk.meta_data;
+            const colKey = colMetadata?.path_in_schema;
+            if (columnList.length > 0 && (0, read_utils_1.fieldIndexOf)(columnList, colKey) < 0) {
+                continue; // eslint-disable-line no-continue
+            }
+            buffer.columnData[colKey.join()] = await this.readColumnChunk(schema, colChunk);
+        }
+        return buffer;
+    }
+    /**
+     * Do reading of parquet file's column chunk
+     * @param schema
+     * @param colChunk
+     */
+    async readColumnChunk(schema, colChunk) {
+        if (colChunk.file_path !== undefined && colChunk.file_path !== null) {
+            throw new Error('external references are not supported');
+        }
+        const field = schema.findField(colChunk.meta_data?.path_in_schema);
+        const type = (0, read_utils_1.getThriftEnum)(parquet_thrift_1.Type, colChunk.meta_data?.type);
+        if (type !== field.primitiveType) {
+            throw new Error(`chunk type not matching schema: ${type}`);
+        }
+        const compression = (0, read_utils_1.getThriftEnum)(parquet_thrift_1.CompressionCodec, colChunk.meta_data?.codec);
+        const pagesOffset = Number(colChunk.meta_data?.data_page_offset);
+        let pagesSize = Number(colChunk.meta_data?.total_compressed_size);
+        if (!colChunk.file_path) {
+            pagesSize = Math.min(this.fileSize - pagesOffset, Number(colChunk.meta_data?.total_compressed_size));
+        }
+        const options = {
+            type,
+            rLevelMax: field.rLevelMax,
+            dLevelMax: field.dLevelMax,
+            compression,
+            column: field,
+            numValues: colChunk.meta_data?.num_values,
+            dictionary: []
+        };
+        let dictionary;
+        const dictionaryPageOffset = colChunk?.meta_data?.dictionary_page_offset;
+        if (dictionaryPageOffset) {
+            const dictionaryOffset = Number(dictionaryPageOffset);
+            // Getting dictionary from column chunk to iterate all over indexes to get dataPage values.
+            dictionary = await this.getDictionary(dictionaryOffset, options, pagesOffset);
+        }
+        dictionary = options.dictionary?.length ? options.dictionary : dictionary;
+        const pagesBuf = await this.read(pagesOffset, pagesSize);
+        return await (0, decoders_1.decodeDataPages)(pagesBuf, { ...options, dictionary });
+    }
+    /**
+     * Getting dictionary for allows to flatten values by indices.
+     * @param dictionaryPageOffset
+     * @param options
+     * @param pagesOffset
+     * @returns
+     */
+    async getDictionary(dictionaryPageOffset, options, pagesOffset) {
+        if (dictionaryPageOffset === 0) {
+            // dictionarySize = Math.min(this.fileSize - pagesOffset, this.defaultDictionarySize);
+            // pagesBuf = await this.read(pagesOffset, dictionarySize);
+            // In this case we are working with parquet-mr files format. Problem is described below:
+            // https://stackoverflow.com/questions/55225108/why-is-dictionary-page-offset-0-for-plain-dictionary-encoding
+            // We need to get dictionary page from column chunk if it exists.
+            // Now if we use code commented above we don't get DICTIONARY_PAGE we get DATA_PAGE instead.
+            return [];
+        }
+        const dictionarySize = Math.min(this.fileSize - dictionaryPageOffset, this.defaultDictionarySize);
+        const pagesBuf = await this.read(dictionaryPageOffset, dictionarySize);
+        const cursor = { buffer: pagesBuf, offset: 0, size: pagesBuf.length };
+        const decodedPage = await (0, decoders_1.decodePage)(cursor, options);
+        return decodedPage.dictionary;
+    }
+    async readFooter() {
+        const trailerLen = constants_1.PARQUET_MAGIC.length + 4;
+        const trailerBuf = await this.read(this.fileSize - trailerLen, trailerLen);
+        const magic = trailerBuf.slice(4).toString();
+        if (magic !== constants_1.PARQUET_MAGIC) {
+            throw new Error(`Not a valid parquet file (magic="${magic})`);
+        }
+        const metadataSize = trailerBuf.readUInt32LE(0);
+        const metadataOffset = this.fileSize - metadataSize - trailerLen;
+        if (metadataOffset < constants_1.PARQUET_MAGIC.length) {
+            throw new Error(`Invalid metadata size ${metadataOffset}`);
+        }
+        const metadataBuf = await this.read(metadataOffset, metadataSize);
+        // let metadata = new parquet_thrift.FileMetaData();
+        // parquet_util.decodeThrift(metadata, metadataBuf);
+        const { metadata } = (0, read_utils_1.decodeFileMetadata)(metadataBuf);
+        return metadata;
+    }
+}
+exports.ParquetEnvelopeReader = ParquetEnvelopeReader;

package/dist/parquetjs/parser/parquet-reader.d.ts

@@ -1,58 +1,68 @@
-import type { ReadableFile } from '@loaders.gl/loader-utils';
+/// <reference types="node" />
+import { ParquetEnvelopeReader } from './parquet-envelope-reader';
+import { FileMetaData } from '../parquet-thrift';
 import { ParquetSchema } from '../schema/schema';
-import { ColumnChunk, FileMetaData, RowGroup } from '../parquet-thrift';
-import { ParquetBuffer, ParquetData, ParquetOptions } from '../schema/declare';
-export type ParquetReaderProps = {
-    defaultDictionarySize?: number;
-};
-/** Properties for initializing a ParquetRowGroupReader */
-export type ParquetIterationProps = {
-    /** Filter allowing some columns to be dropped */
-    columnList?: string[] | string[][];
-};
+import { ParquetCursor } from './parquet-cursor';
 /**
- * The parquet envelope reader allows direct, unbuffered access to the individual
- * sections of the parquet file, namely the header, footer and the row groups.
- * This class is intended for advanced/internal users; if you just want to retrieve
- * rows from a parquet file use the ParquetReader instead
+ * A parquet reader allows retrieving the rows from a parquet file in order.
+ * The basic usage is to create a reader and then retrieve a cursor/iterator
+ * which allows you to consume row after row until all rows have been read. It is
+ * important that you call close() after you are finished reading the file to
+ * avoid leaking file descriptors.
  */
-export declare class ParquetReader {
-    props: Required<ParquetReaderProps>;
-    file: ReadableFile;
-    metadata: Promise<FileMetaData> | null;
-    constructor(file: ReadableFile, props?: ParquetReaderProps);
-    close(): void;
-    /** Yield one row at a time */
-    rowIterator(props?: ParquetIterationProps): AsyncGenerator<import("../schema/declare").ParquetRecord, void, unknown>;
-    /** Yield one batch of rows at a time */
-    rowBatchIterator(props?: ParquetIterationProps): AsyncGenerator<import("../schema/declare").ParquetRecord[], void, unknown>;
-    /** Iterate over the raw row groups */
-    rowGroupIterator(props?: ParquetIterationProps): AsyncGenerator<ParquetBuffer, void, unknown>;
-    getRowCount(): Promise<number>;
-    getSchema(): Promise<ParquetSchema>;
+export declare class ParquetReader<T> implements AsyncIterable<T> {
     /**
-     * Returns the user (key/value) metadata for this file
-     * In parquet this is not stored on the schema like it is in arrow
+     * return a new parquet reader initialized with a read function
+     */
+    static openBlob<T>(blob: Blob): Promise<ParquetReader<T>>;
+    /**
+     * return a new parquet reader initialized with a read function
+     */
+    static openArrayBuffer<T>(arrayBuffer: ArrayBuffer): Promise<ParquetReader<T>>;
+    static openBuffer<T>(buffer: Buffer): Promise<ParquetReader<T>>;
+    metadata: FileMetaData;
+    envelopeReader: ParquetEnvelopeReader;
+    schema: ParquetSchema;
+    /**
+     * Create a new parquet reader from the file metadata and an envelope reader.
+     * It is not recommended to call this constructor directly except for advanced
+     * and internal use cases. Consider using one of the open{File,Buffer} methods
+     * instead
+     */
+    constructor(metadata: FileMetaData, envelopeReader: ParquetEnvelopeReader);
+    /**
+     * Close this parquet reader. You MUST call this method once you're finished
+     * reading rows
      */
-    getSchemaMetadata(): Promise<Record<string, string>>;
-    getFileMetadata(): Promise<FileMetaData>;
-    /** Metadata is stored in the footer */
-    readHeader(): Promise<void>;
-    /** Metadata is stored in the footer */
-    readFooter(): Promise<FileMetaData>;
-    /** Data is stored in row groups (similar to Apache Arrow record batches) */
-    readRowGroup(schema: ParquetSchema, rowGroup: RowGroup, columnList: string[][]): Promise<ParquetBuffer>;
+    close(): Promise<void>;
     /**
-     * Each row group contains column chunks for all the columns.
+     * Return a cursor to the file. You may open more than one cursor and use
+     * them concurrently. All cursors become invalid once close() is called on
+     * the reader object.
+     *
+     * The required_columns parameter controls which columns are actually read
+     * from disk. An empty array or no value implies all columns. A list of column
+     * names means that only those columns should be loaded from disk.
+     */
+    getCursor(): ParquetCursor<T>;
+    getCursor<K extends keyof T>(columnList: (K | K[])[]): ParquetCursor<Pick<T, K>>;
+    getCursor(columnList: (string | string[])[]): ParquetCursor<Partial<T>>;
+    /**
+     * Return the number of rows in this file. Note that the number of rows is
+     * not neccessarily equal to the number of rows in each column.
+     */
+    getRowCount(): number;
+    /**
+     * Returns the ParquetSchema for this file
+     */
+    getSchema(): ParquetSchema;
+    /**
+     * Returns the user (key/value) metadata for this file
      */
-    readColumnChunk(schema: ParquetSchema, colChunk: ColumnChunk): Promise<ParquetData>;
+    getMetadata(): Record<string, string>;
     /**
-     * Getting dictionary for allows to flatten values by indices.
-     * @param dictionaryPageOffset
-     * @param options
-     * @param pagesOffset
-     * @returns
+     * Implement AsyncIterable
      */
-    getDictionary(dictionaryPageOffset: number, options: ParquetOptions, pagesOffset: number): Promise<string[]>;
+    [Symbol.asyncIterator](): AsyncIterator<T>;
 }
 //# sourceMappingURL=parquet-reader.d.ts.map

package/dist/parquetjs/parser/parquet-reader.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"parquet-reader.d.ts","sourceRoot":"","sources":["../../../src/parquetjs/parser/parquet-reader.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAC,YAAY,EAAC,MAAM,0BAA0B,CAAC;AAE3D,OAAO,EAAC,aAAa,EAAC,MAAM,kBAAkB,CAAC;AAK/C,OAAO,EAAC,WAAW,EAAoB,YAAY,EAAE,QAAQ,EAAO,MAAM,mBAAmB,CAAC;AAC9F,OAAO,EACL,aAAa,EAEb,WAAW,EAEX,cAAc,EACf,MAAM,mBAAmB,CAAC;AAI3B,MAAM,MAAM,kBAAkB,GAAG;IAC/B,qBAAqB,CAAC,EAAE,MAAM,CAAC;CAChC,CAAC;AAEF,0DAA0D;AAC1D,MAAM,MAAM,qBAAqB,GAAG;IAClC,iDAAiD;IACjD,UAAU,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,EAAE,EAAE,CAAC;CACpC,CAAC;AAMF;;;;;GAKG;AACH,qBAAa,aAAa;IACxB,KAAK,EAAE,QAAQ,CAAC,kBAAkB,CAAC,CAAC;IACpC,IAAI,EAAE,YAAY,CAAC;IACnB,QAAQ,EAAE,OAAO,CAAC,YAAY,CAAC,GAAG,IAAI,CAAQ;gBAElC,IAAI,EAAE,YAAY,EAAE,KAAK,CAAC,EAAE,kBAAkB;IAK1D,KAAK,IAAI,IAAI;IAOb,8BAA8B;IACvB,WAAW,CAAC,KAAK,CAAC,EAAE,qBAAqB;IAShD,wCAAwC;IACjC,gBAAgB,CAAC,KAAK,CAAC,EAAE,qBAAqB;IAOrD,sCAAsC;IAC/B,gBAAgB,CAAC,KAAK,CAAC,EAAE,qBAAqB;IAqB/C,WAAW,IAAI,OAAO,CAAC,MAAM,CAAC;IAK9B,SAAS,IAAI,OAAO,CAAC,aAAa,CAAC;IAQzC;;;OAGG;IACG,iBAAiB,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IASpD,eAAe,IAAI,OAAO,CAAC,YAAY,CAAC;IAU9C,uCAAuC;IACjC,UAAU,IAAI,OAAO,CAAC,IAAI,CAAC;IAajC,uCAAuC;IACjC,UAAU,IAAI,OAAO,CAAC,YAAY,CAAC;IAsBzC,4EAA4E;IACtE,YAAY,CAChB,MAAM,EAAE,aAAa,EACrB,QAAQ,EAAE,QAAQ,EAClB,UAAU,EAAE,MAAM,EAAE,EAAE,GACrB,OAAO,CAAC,aAAa,CAAC;IAgBzB;;OAEG;IACG,eAAe,CAAC,MAAM,EAAE,aAAa,EAAE,QAAQ,EAAE,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;IAoDzF;;;;;;OAMG;IACG,aAAa,CACjB,oBAAoB,EAAE,MAAM,EAC5B,OAAO,EAAE,cAAc,EACvB,WAAW,EAAE,MAAM,GAClB,OAAO,CAAC,MAAM,EAAE,CAAC;CAuBrB"}
+{"version":3,"file":"parquet-reader.d.ts","sourceRoot":"","sources":["../../../src/parquetjs/parser/parquet-reader.ts"],"names":[],"mappings":";AACA,OAAO,EAAC,qBAAqB,EAAC,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAC,YAAY,EAAC,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAC,aAAa,EAAC,MAAM,kBAAkB,CAAC;AAC/C,OAAO,EAAC,aAAa,EAAC,MAAM,kBAAkB,CAAC;AAI/C;;;;;;GAMG;AACH,qBAAa,aAAa,CAAC,CAAC,CAAE,YAAW,aAAa,CAAC,CAAC,CAAC;IACvD;;OAEG;WACU,QAAQ,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;IAkB/D;;OAEG;WACU,eAAe,CAAC,CAAC,EAAE,WAAW,EAAE,WAAW,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;WAevE,UAAU,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;IAY9D,QAAQ,EAAE,YAAY,CAAC;IACvB,cAAc,EAAE,qBAAqB,CAAC;IACtC,MAAM,EAAE,aAAa,CAAC;IAE7B;;;;;OAKG;gBACS,QAAQ,EAAE,YAAY,EAAE,cAAc,EAAE,qBAAqB;IAYzE;;;OAGG;IACG,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAM5B;;;;;;;;OAQG;IACH,SAAS,IAAI,aAAa,CAAC,CAAC,CAAC;IAE7B,SAAS,CAAC,CAAC,SAAS,MAAM,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAChF,SAAS,CAAC,UAAU,EAAE,CAAC,MAAM,GAAG,MAAM,EAAE,CAAC,EAAE,GAAG,aAAa,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAkBvE;;;OAGG;IACH,WAAW,IAAI,MAAM;IAIrB;;OAEG;IACH,SAAS,IAAI,aAAa;IAI1B;;OAEG;IACH,WAAW,IAAI,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC;IAQrC;;OAEG;IAEH,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,aAAa,CAAC,CAAC,CAAC;CAG3C"}

package/dist/parquetjs/parser/parquet-reader.js

@@ -1,200 +1,134 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.ParquetReader = void 0;
+// Forked from https://github.com/kbajalc/parquets under MIT license (Copyright (c) 2017 ironSource Ltd.)
+const parquet_envelope_reader_1 = require("./parquet-envelope-reader");
 const schema_1 = require("../schema/schema");
-const decoders_1 = require("./decoders");
-const shred_1 = require("../schema/shred");
+const parquet_cursor_1 = require("./parquet-cursor");
 const constants_1 = require("../../constants");
-const parquet_thrift_1 = require("../parquet-thrift");
-const read_utils_1 = require("../utils/read-utils");
-const decoders_2 = require("./decoders");
-const DEFAULT_PROPS = {
-    defaultDictionarySize: 1e6
-};
+const decoders_1 = require("./decoders");
 /**
- * The parquet envelope reader allows direct, unbuffered access to the individual
- * sections of the parquet file, namely the header, footer and the row groups.
- * This class is intended for advanced/internal users; if you just want to retrieve
- * rows from a parquet file use the ParquetReader instead
+ * A parquet reader allows retrieving the rows from a parquet file in order.
+ * The basic usage is to create a reader and then retrieve a cursor/iterator
+ * which allows you to consume row after row until all rows have been read. It is
+ * important that you call close() after you are finished reading the file to
+ * avoid leaking file descriptors.
  */
 class ParquetReader {
-    constructor(file, props) {
-        this.metadata = null;
-        this.file = file;
-        this.props = { ...DEFAULT_PROPS, ...props };
-    }
-    close() {
-        // eslint-disable-next-line @typescript-eslint/no-floating-promises
-        this.file.close();
-    }
-    // HIGH LEVEL METHODS
-    /** Yield one row at a time */
-    async *rowIterator(props) {
-        for await (const rows of this.rowBatchIterator(props)) {
-            // yield *rows
-            for (const row of rows) {
-                yield row;
-            }
+    /**
+     * return a new parquet reader initialized with a read function
+     */
+    static async openBlob(blob) {
+        const readFn = async (start, length) => {
+            const arrayBuffer = await blob.slice(start, start + length).arrayBuffer();
+            return Buffer.from(arrayBuffer);
+        };
+        const closeFn = async () => { };
+        const size = blob.size;
+        const envelopeReader = new parquet_envelope_reader_1.ParquetEnvelopeReader(readFn, closeFn, size);
+        try {
+            await envelopeReader.readHeader();
+            const metadata = await envelopeReader.readFooter();
+            return new ParquetReader(metadata, envelopeReader);
         }
-    }
-    /** Yield one batch of rows at a time */
-    async *rowBatchIterator(props) {
-        const schema = await this.getSchema();
-        for await (const rowGroup of this.rowGroupIterator(props)) {
-            yield (0, shred_1.materializeRecords)(schema, rowGroup);
+        catch (err) {
+            await envelopeReader.close();
+            throw err;
         }
     }
-    /** Iterate over the raw row groups */
-    async *rowGroupIterator(props) {
-        // Ensure strings are nested in arrays
-        const columnList = (props?.columnList || []).map((x) => Array.isArray(x) ? x : [x]);
-        const metadata = await this.getFileMetadata();
-        const schema = await this.getSchema();
-        const rowGroupCount = metadata?.row_groups.length || 0;
-        for (let rowGroupIndex = 0; rowGroupIndex < rowGroupCount; rowGroupIndex++) {
-            const rowGroup = await this.readRowGroup(schema, metadata.row_groups[rowGroupIndex], columnList);
-            yield rowGroup;
+    /**
+     * return a new parquet reader initialized with a read function
+     */
+    static async openArrayBuffer(arrayBuffer) {
+        const readFn = async (start, length) => Buffer.from(arrayBuffer, start, length);
+        const closeFn = async () => { };
+        const size = arrayBuffer.byteLength;
+        const envelopeReader = new parquet_envelope_reader_1.ParquetEnvelopeReader(readFn, closeFn, size);
+        try {
+            await envelopeReader.readHeader();
+            const metadata = await envelopeReader.readFooter();
+            return new ParquetReader(metadata, envelopeReader);
+        }
+        catch (err) {
+            await envelopeReader.close();
+            throw err;
         }
     }
-    async getRowCount() {
-        const metadata = await this.getFileMetadata();
-        return Number(metadata.num_rows);
-    }
-    async getSchema() {
-        const metadata = await this.getFileMetadata();
-        const root = metadata.schema[0];
-        const { schema: schemaDefinition } = (0, decoders_1.decodeSchema)(metadata.schema, 1, root.num_children);
-        const schema = new schema_1.ParquetSchema(schemaDefinition);
-        return schema;
+    static async openBuffer(buffer) {
+        const envelopeReader = await parquet_envelope_reader_1.ParquetEnvelopeReader.openBuffer(buffer);
+        try {
+            await envelopeReader.readHeader();
+            const metadata = await envelopeReader.readFooter();
+            return new ParquetReader(metadata, envelopeReader);
+        }
+        catch (err) {
+            await envelopeReader.close();
+            throw err;
+        }
     }
     /**
-     * Returns the user (key/value) metadata for this file
-     * In parquet this is not stored on the schema like it is in arrow
+     * Create a new parquet reader from the file metadata and an envelope reader.
+     * It is not recommended to call this constructor directly except for advanced
+     * and internal use cases. Consider using one of the open{File,Buffer} methods
+     * instead
      */
-    async getSchemaMetadata() {
-        const metadata = await this.getFileMetadata();
-        const md = {};
-        for (const kv of metadata.key_value_metadata) {
-            md[kv.key] = kv.value;
+    constructor(metadata, envelopeReader) {
+        if (metadata.version !== constants_1.PARQUET_VERSION) {
+            throw new Error('invalid parquet version');
         }
-        return md;
+        this.metadata = metadata;
+        this.envelopeReader = envelopeReader;
+        const root = this.metadata.schema[0];
+        const { schema } = (0, decoders_1.decodeSchema)(this.metadata.schema, 1, root.num_children);
+        this.schema = new schema_1.ParquetSchema(schema);
     }
-    async getFileMetadata() {
-        if (!this.metadata) {
-            await this.readHeader();
-            this.metadata = this.readFooter();
-        }
-        return this.metadata;
+    /**
+     * Close this parquet reader. You MUST call this method once you're finished
+     * reading rows
+     */
+    async close() {
+        await this.envelopeReader.close();
+        // this.envelopeReader = null;
+        // this.metadata = null;
     }
-    // LOW LEVEL METHODS
-    /** Metadata is stored in the footer */
-    async readHeader() {
-        const buffer = await this.file.read(0, constants_1.PARQUET_MAGIC.length);
-        const magic = buffer.toString();
-        switch (magic) {
-            case constants_1.PARQUET_MAGIC:
-                break;
-            case constants_1.PARQUET_MAGIC_ENCRYPTED:
-                throw new Error('Encrypted parquet file not supported');
-            default:
-                throw new Error(`Invalid parquet file (magic=${magic})`);
+    getCursor(columnList) {
+        if (!columnList) {
+            // tslint:disable-next-line:no-parameter-reassignment
+            columnList = [];
         }
+        // tslint:disable-next-line:no-parameter-reassignment
+        columnList = columnList.map((x) => (Array.isArray(x) ? x : [x]));
+        return new parquet_cursor_1.ParquetCursor(this.metadata, this.envelopeReader, this.schema, columnList);
     }
-    /** Metadata is stored in the footer */
-    async readFooter() {
-        const trailerLen = constants_1.PARQUET_MAGIC.length + 4;
-        const trailerBuf = await this.file.read(this.file.size - trailerLen, trailerLen);
-        const magic = trailerBuf.slice(4).toString();
-        if (magic !== constants_1.PARQUET_MAGIC) {
-            throw new Error(`Not a valid parquet file (magic="${magic})`);
-        }
-        const metadataSize = trailerBuf.readUInt32LE(0);
-        const metadataOffset = this.file.size - metadataSize - trailerLen;
-        if (metadataOffset < constants_1.PARQUET_MAGIC.length) {
-            throw new Error(`Invalid metadata size ${metadataOffset}`);
-        }
-        const metadataBuf = await this.file.read(metadataOffset, metadataSize);
-        // let metadata = new parquet_thrift.FileMetaData();
-        // parquet_util.decodeThrift(metadata, metadataBuf);
-        const { metadata } = (0, read_utils_1.decodeFileMetadata)(metadataBuf);
-        return metadata;
+    /**
+     * Return the number of rows in this file. Note that the number of rows is
+     * not neccessarily equal to the number of rows in each column.
+     */
+    getRowCount() {
+        return Number(this.metadata.num_rows);
     }
-    /** Data is stored in row groups (similar to Apache Arrow record batches) */
-    async readRowGroup(schema, rowGroup, columnList) {
-        const buffer = {
-            rowCount: Number(rowGroup.num_rows),
-            columnData: {}
-        };
-        for (const colChunk of rowGroup.columns) {
-            const colMetadata = colChunk.meta_data;
-            const colKey = colMetadata?.path_in_schema;
-            if (columnList.length > 0 && (0, read_utils_1.fieldIndexOf)(columnList, colKey) < 0) {
-                continue; // eslint-disable-line no-continue
-            }
-            buffer.columnData[colKey.join()] = await this.readColumnChunk(schema, colChunk);
-        }
-        return buffer;
+    /**
+     * Returns the ParquetSchema for this file
+     */
+    getSchema() {
+        return this.schema;
     }
     /**
-     * Each row group contains column chunks for all the columns.
+     * Returns the user (key/value) metadata for this file
      */
-    async readColumnChunk(schema, colChunk) {
-        if (colChunk.file_path !== undefined && colChunk.file_path !== null) {
-            throw new Error('external references are not supported');
-        }
-        const field = schema.findField(colChunk.meta_data?.path_in_schema);
-        const type = (0, read_utils_1.getThriftEnum)(parquet_thrift_1.Type, colChunk.meta_data?.type);
-        if (type !== field.primitiveType) {
-            throw new Error(`chunk type not matching schema: ${type}`);
-        }
-        const compression = (0, read_utils_1.getThriftEnum)(parquet_thrift_1.CompressionCodec, colChunk.meta_data?.codec);
-        const pagesOffset = Number(colChunk.meta_data?.data_page_offset);
-        let pagesSize = Number(colChunk.meta_data?.total_compressed_size);
-        if (!colChunk.file_path) {
-            pagesSize = Math.min(this.file.size - pagesOffset, Number(colChunk.meta_data?.total_compressed_size));
-        }
-        const options = {
-            type,
-            rLevelMax: field.rLevelMax,
-            dLevelMax: field.dLevelMax,
-            compression,
-            column: field,
-            numValues: colChunk.meta_data?.num_values,
-            dictionary: []
-        };
-        let dictionary;
-        const dictionaryPageOffset = colChunk?.meta_data?.dictionary_page_offset;
-        if (dictionaryPageOffset) {
-            const dictionaryOffset = Number(dictionaryPageOffset);
-            // Getting dictionary from column chunk to iterate all over indexes to get dataPage values.
-            dictionary = await this.getDictionary(dictionaryOffset, options, pagesOffset);
+    getMetadata() {
+        const md = {};
+        for (const kv of this.metadata.key_value_metadata) {
+            md[kv.key] = kv.value;
         }
-        dictionary = options.dictionary?.length ? options.dictionary : dictionary;
-        const pagesBuf = await this.file.read(pagesOffset, pagesSize);
-        return await (0, decoders_2.decodeDataPages)(pagesBuf, { ...options, dictionary });
+        return md;
     }
     /**
-     * Getting dictionary for allows to flatten values by indices.
-     * @param dictionaryPageOffset
-     * @param options
-     * @param pagesOffset
-     * @returns
+     * Implement AsyncIterable
      */
-    async getDictionary(dictionaryPageOffset, options, pagesOffset) {
-        if (dictionaryPageOffset === 0) {
-            // dictionarySize = Math.min(this.fileSize - pagesOffset, this.defaultDictionarySize);
-            // pagesBuf = await this.read(pagesOffset, dictionarySize);
-            // In this case we are working with parquet-mr files format. Problem is described below:
-            // https://stackoverflow.com/questions/55225108/why-is-dictionary-page-offset-0-for-plain-dictionary-encoding
-            // We need to get dictionary page from column chunk if it exists.
-            // Now if we use code commented above we don't get DICTIONARY_PAGE we get DATA_PAGE instead.
-            return [];
-        }
-        const dictionarySize = Math.min(this.file.size - dictionaryPageOffset, this.props.defaultDictionarySize);
-        const pagesBuf = await this.file.read(dictionaryPageOffset, dictionarySize);
-        const cursor = { buffer: pagesBuf, offset: 0, size: pagesBuf.length };
-        const decodedPage = await (0, decoders_2.decodePage)(cursor, options);
-        return decodedPage.dictionary;
+    // tslint:disable-next-line:function-name
+    [Symbol.asyncIterator]() {
+        return this.getCursor()[Symbol.asyncIterator]();
     }
 }
 exports.ParquetReader = ParquetReader;

package/dist/parquetjs/schema/declare.d.ts

@@ -45,7 +45,6 @@ export interface ParquetField {
     fieldCount?: number;
     fields?: Record<string, ParquetField>;
 }
-/** @todo better name, this is an internal type? */
 export interface ParquetOptions {
     type: ParquetType;
     rLevelMax: number;
@@ -55,33 +54,27 @@ export interface ParquetOptions {
     numValues?: Int64;
     dictionary?: ParquetDictionary;
 }
+export interface ParquetData {
+    dlevels: number[];
+    rlevels: number[];
+    values: any[];
+    count: number;
+    pageHeaders: PageHeader[];
+}
 export interface ParquetPageData {
     dlevels: number[];
     rlevels: number[];
-    /** Actual column chunks */
     values: any[];
     count: number;
     dictionary?: ParquetDictionary;
-    /** The "raw" page header from the file */
     pageHeader: PageHeader;
 }
 export interface ParquetRecord {
     [key: string]: any;
 }
-/** @
- * Holds data for one row group (column chunks) */
 export declare class ParquetBuffer {
-    /** Number of rows in this page */
     rowCount: number;
     columnData: Record<string, ParquetData>;
     constructor(rowCount?: number, columnData?: Record<string, ParquetData>);
 }
-/** Holds the data for one column chunk */
-export interface ParquetData {
-    dlevels: number[];
-    rlevels: number[];
-    values: any[];
-    count: number;
-    pageHeaders: PageHeader[];
-}
 //# sourceMappingURL=declare.d.ts.map

package/dist/parquetjs/schema/declare.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"declare.d.ts","sourceRoot":"","sources":["../../../src/parquetjs/schema/declare.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,MAAM,YAAY,CAAC;AAC/B,OAAO,KAAK,EAAC,UAAU,EAAC,MAAM,mBAAmB,CAAC;AAElD,MAAM,MAAM,YAAY,GAAG,OAAO,GAAG,KAAK,GAAG,kBAAkB,CAAC;AAChE,MAAM,MAAM,kBAAkB,GAC1B,cAAc,GACd,MAAM,GACN,QAAQ,GACR,KAAK,GACL,QAAQ,GACR,KAAK,GACL,SAAS,GACT,MAAM,CAAC;AACX,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG,UAAU,GAAG,UAAU,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,aAAa,GAAG,YAAY,CAAC;AAEvD;;GAEG;AACH,MAAM,MAAM,aAAa,GAErB,SAAS,GACT,OAAO,GACP,OAAO,GACP,OAAO,GACP,OAAO,GACP,QAAQ,GACR,YAAY,GACZ,sBAAsB,CAAC;AAE3B;;GAEG;AACH,MAAM,MAAM,YAAY,GAEpB,MAAM,GAMN,eAAe,GACf,eAAe,GACf,oBAAoB,GACpB,8BAA8B,GAC9B,MAAM,GACN,aAAa,GACb,aAAa,GACb,kBAAkB,GAClB,kBAAkB,GAClB,QAAQ,GACR,SAAS,GACT,SAAS,GACT,SAAS,GACT,OAAO,GACP,QAAQ,GACR,QAAQ,GACR,QAAQ,GACR,MAAM,GACN,MAAM,GACN,UAAU,CAAC;AAEf,MAAM,MAAM,iBAAiB,GAAG,MAAM,EAAE,CAAC;AAEzC,MAAM,WAAW,gBAAgB;IAC/B,CAAC,MAAM,EAAE,MAAM,GAAG,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,WAAW,CAAC,EAAE,kBAAkB,CAAC;IACjC,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,MAAM,CAAC,EAAE,gBAAgB,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,YAAY,CAAC,EAAE,YAAY,CAAC;IAC5B,cAAc,EAAE,cAAc,CAAC;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,WAAW,CAAC,EAAE,kBAAkB,CAAC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;CACvC;AAED,mDAAmD;AACnD,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,kBAAkB,CAAC;IAChC,MAAM,EAAE,YAAY,CAAC;IACrB,SAAS,CAAC,EAAE,KAAK,CAAC;IAClB,UAAU,CAAC,EAAE,iBAAiB,CAAC;CAChC;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,2BAA2B;IAC3B,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,iBAAiB,CAAC;IAC/B,0CAA0C;IAC1C,UAAU,EAAE,UAAU,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B,CAAC,GAAG,EAAE,MAAM,GAAG,GAAG,CAAC;CACpB;AAED;kDACkD;AAClD,qBAAa,aAAa;IACxB,kCAAkC;IAClC,QAAQ,EAAE,MAAM,CAAC;IAEjB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;gBAC5B,QAAQ,GAAE,MAAU,EAAE,UAAU,GAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAM;CAI/E;AAED,0CAA0C;AAC1C,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,UAAU,EAAE,CAAC;CAC3B"}
+{"version":3,"file":"declare.d.ts","sourceRoot":"","sources":["../../../src/parquetjs/schema/declare.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,MAAM,YAAY,CAAC;AAC/B,OAAO,KAAK,EAAC,UAAU,EAAC,MAAM,mBAAmB,CAAC;AAElD,MAAM,MAAM,YAAY,GAAG,OAAO,GAAG,KAAK,GAAG,kBAAkB,CAAC;AAChE,MAAM,MAAM,kBAAkB,GAC1B,cAAc,GACd,MAAM,GACN,QAAQ,GACR,KAAK,GACL,QAAQ,GACR,KAAK,GACL,SAAS,GACT,MAAM,CAAC;AACX,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG,UAAU,GAAG,UAAU,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,aAAa,GAAG,YAAY,CAAC;AAEvD;;GAEG;AACH,MAAM,MAAM,aAAa,GAErB,SAAS,GACT,OAAO,GACP,OAAO,GACP,OAAO,GACP,OAAO,GACP,QAAQ,GACR,YAAY,GACZ,sBAAsB,CAAC;AAE3B;;GAEG;AACH,MAAM,MAAM,YAAY,GAEpB,MAAM,GAMN,eAAe,GACf,eAAe,GACf,oBAAoB,GACpB,8BAA8B,GAC9B,MAAM,GACN,aAAa,GACb,aAAa,GACb,kBAAkB,GAClB,kBAAkB,GAClB,QAAQ,GACR,SAAS,GACT,SAAS,GACT,SAAS,GACT,OAAO,GACP,QAAQ,GACR,QAAQ,GACR,QAAQ,GACR,MAAM,GACN,MAAM,GACN,UAAU,CAAC;AAEf,MAAM,MAAM,iBAAiB,GAAG,MAAM,EAAE,CAAC;AAEzC,MAAM,WAAW,gBAAgB;IAC/B,CAAC,MAAM,EAAE,MAAM,GAAG,eAAe,CAAC;CACnC;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,CAAC,EAAE,WAAW,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,WAAW,CAAC,EAAE,kBAAkB,CAAC;IACjC,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,MAAM,CAAC,EAAE,gBAAgB,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,YAAY,CAAC,EAAE,YAAY,CAAC;IAC5B,cAAc,EAAE,cAAc,CAAC;IAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,WAAW,CAAC,EAAE,kBAAkB,CAAC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;CACvC;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE,WAAW,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,kBAAkB,CAAC;IAChC,MAAM,EAAE,YAAY,CAAC;IACrB,SAAS,CAAC,EAAE,KAAK,CAAC;IAClB,UAAU,CAAC,EAAE,iBAAiB,CAAC;CAChC;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,UAAU,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,EAAE,GAAG,EAAE,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,iBAAiB,CAAC;IAC/B,UAAU,EAAE,UAAU,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B,CAAC,GAAG,EAAE,MAAM,GAAG,GAAG,CAAC;CACpB;AAED,qBAAa,aAAa;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;gBAC5B,QAAQ,GAAE,MAAU,EAAE,UAAU,GAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAM;CAI/E"}

package/dist/parquetjs/schema/declare.js

@@ -1,8 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.ParquetBuffer = void 0;
-/** @
- * Holds data for one row group (column chunks) */
 class ParquetBuffer {
     constructor(rowCount = 0, columnData = {}) {
         this.rowCount = rowCount;