intel-hex.js


/**
 * Parser/writer for the "Intel hex" format.
 */

/*
 * A regexp that matches lines in a .hex file.
 *
 * One hexadecimal character is matched by "[0-9A-Fa-f]".
 * Two hex characters are matched by "[0-9A-Fa-f]{2}"
 * Eight or more hex characters are matched by "[0-9A-Fa-f]{8,}"
 * A capture group of two hex characters is "([0-9A-Fa-f]{2})"
 *
 * Record mark         :
 * 8 or more hex chars  ([0-9A-Fa-f]{8,})
 * Checksum                              ([0-9A-Fa-f]{2})
 * Optional newline                                      (?:\r\n|\r|\n|)
 */
const hexLineRegexp = /:([0-9A-Fa-f]{8,})([0-9A-Fa-f]{2})(?:\r\n|\r|\n|)/g;


// Takes a Uint8Array as input,
// Returns an integer in the 0-255 range.
function checksum(bytes) {
    return (-bytes.reduce((sum, v)=>sum + v, 0)) & 0xFF;
}

// Takes two Uint8Arrays as input,
// Returns an integer in the 0-255 range.
function checksumTwo(array1, array2) {
    const partial1 = array1.reduce((sum, v)=>sum + v, 0);
    const partial2 = array2.reduce((sum, v)=>sum + v, 0);
    return -( partial1 + partial2 ) & 0xFF;
}


// Trivial utility. Converts a number to hex and pads with zeroes up to 2 characters.
function hexpad(number) {
    return number.toString(16).toUpperCase().padStart(2, '0');
}


// Polyfill as per https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Number/isInteger
Number.isInteger = Number.isInteger || function(value) {
    return typeof value === 'number' &&
    isFinite(value) &&
    Math.floor(value) === value;
};


/**
 * @class MemoryMap
 *
 * Represents the contents of a memory layout, with main focus into (possibly sparse) blocks of data.
 *<br/>
 * A {@linkcode MemoryMap} acts as a subclass of
 * {@linkcode https://developer.mozilla.org/docs/Web/JavaScript/Reference/Global_Objects/Map|Map}.
 * In every entry of it, the key is the starting address of a data block (an integer number),
 * and the value is the <tt>Uint8Array</tt> with the data for that block.
 *<br/>
 * The main rationale for this is that a .hex file can contain a single block of contiguous
 * data starting at memory address 0 (and it's the common case for simple .hex files),
 * but complex files with several non-contiguous data blocks are also possible, thus
 * the need for a data structure on top of the <tt>Uint8Array</tt>s.
 *<br/>
 * In order to parse <tt>.hex</tt> files, use the {@linkcode MemoryMap.fromHex} <em>static</em> factory
 * method. In order to write <tt>.hex</tt> files, create a new {@linkcode MemoryMap} and call
 * its {@linkcode MemoryMap.asHexString} method.
 *
 * @extends Map
 * @example
 * import MemoryMap from 'nrf-intel-hex';
 *
 * let memMap1 = new MemoryMap();
 * let memMap2 = new MemoryMap([[0, new Uint8Array(1,2,3,4)]]);
 * let memMap3 = new MemoryMap({0: new Uint8Array(1,2,3,4)});
 * let memMap4 = new MemoryMap({0xCF0: new Uint8Array(1,2,3,4)});
 */
class MemoryMap {
    /**
     * @param {Iterable} blocks The initial value for the memory blocks inside this
     * <tt>MemoryMap</tt>. All keys must be numeric, and all values must be instances of
     * <tt>Uint8Array</tt>. Optionally it can also be a plain <tt>Object</tt> with
     * only numeric keys.
     */
    constructor(blocks) {
        this._blocks = new Map();

        if (blocks && typeof blocks[Symbol.iterator] === 'function') {
            for (const tuple of blocks) {
                if (!(tuple instanceof Array) || tuple.length !== 2) {
                    throw new Error('First parameter to MemoryMap constructor must be an iterable of [addr, bytes] or undefined');
                }
                this.set(tuple[0], tuple[1]);
            }
        } else if (typeof blocks === 'object') {
            // Try iterating through the object's keys
            const addrs = Object.keys(blocks);
            for (const addr of addrs) {
                this.set(parseInt(addr), blocks[addr]);
            }

        } else if (blocks !== undefined && blocks !== null) {
            throw new Error('First parameter to MemoryMap constructor must be an iterable of [addr, bytes] or undefined');
        }
    }

    set(addr, value) {
        if (!Number.isInteger(addr)) {
            throw new Error('Address passed to MemoryMap is not an integer');
        }
        if (addr < 0) {
            throw new Error('Address passed to MemoryMap is negative');
        }
        if (!(value instanceof Uint8Array)) {
            throw new Error('Bytes passed to MemoryMap are not an Uint8Array');
        }
        return this._blocks.set(addr, value);
    }
    // Delegate the following to the 'this._blocks' Map:
    get(addr)    { return this._blocks.get(addr);    }
    clear()      { return this._blocks.clear();      }
    delete(addr) { return this._blocks.delete(addr); }
    entries()    { return this._blocks.entries();    }
    forEach(callback, that) { return this._blocks.forEach(callback, that); }
    has(addr)    { return this._blocks.has(addr);    }
    keys()       { return this._blocks.keys();       }
    values()     { return this._blocks.values();     }
    get size()   { return this._blocks.size;         }
    [Symbol.iterator]() { return this._blocks[Symbol.iterator](); }


    /**
     * Parses a string containing data formatted in "Intel HEX" format, and
     * returns an instance of {@linkcode MemoryMap}.
     *<br/>
     * The insertion order of keys in the {@linkcode MemoryMap} is guaranteed to be strictly
     * ascending. In other words, when iterating through the {@linkcode MemoryMap}, the addresses
     * will be ordered in ascending order.
     *<br/>
     * The parser has an opinionated behaviour, and will throw a descriptive error if it
     * encounters some malformed input. Check the project's
     * {@link https://github.com/NordicSemiconductor/nrf-intel-hex#Features|README file} for details.
     *<br/>
     * If <tt>maxBlockSize</tt> is given, any contiguous data block larger than that will
     * be split in several blocks.
     *
     * @param {String} hexText The contents of a .hex file.
     * @param {Number} [maxBlockSize=Infinity] Maximum size of the returned <tt>Uint8Array</tt>s.
     *
     * @return {MemoryMap}
     *
     * @example
     * import MemoryMap from 'nrf-intel-hex';
     *
     * let intelHexString =
     *     ":100000000102030405060708090A0B0C0D0E0F1068\n" +
     *     ":00000001FF";
     *
     * let memMap = MemoryMap.fromHex(intelHexString);
     *
     * for (let [address, dataBlock] of memMap) {
     *     console.log('Data block at ', address, ', bytes: ', dataBlock);
     * }
     */
    static fromHex(hexText, maxBlockSize = Infinity) {
        const blocks = new MemoryMap();

        let lastCharacterParsed = 0;
        let matchResult;
        let recordCount = 0;

        // Upper Linear Base Address, the 16 most significant bits (2 bytes) of
        // the current 32-bit (4-byte) address
        // In practice this is a offset that is summed to the "load offset" of the
        // data records
        let ulba = 0;

        hexLineRegexp.lastIndex = 0; // Reset the regexp, if not it would skip content when called twice

        while ((matchResult = hexLineRegexp.exec(hexText)) !== null) {
            recordCount++;

            // By default, a regexp loop ignores gaps between matches, but
            // we want to be aware of them.
            if (lastCharacterParsed !== matchResult.index) {
                throw new Error(
                    'Malformed hex file: Could not parse between characters ' +
                    lastCharacterParsed +
                    ' and ' +
                    matchResult.index +
                    ' ("' +
                    hexText.substring(lastCharacterParsed, Math.min(matchResult.index, lastCharacterParsed + 16)).trim() +
                    '")');
            }
            lastCharacterParsed = hexLineRegexp.lastIndex;

            // Give pretty names to the match's capture groups
            const [, recordStr, recordChecksum] = matchResult;

            // String to Uint8Array - https://stackoverflow.com/questions/43131242/how-to-convert-a-hexademical-string-of-data-to-an-arraybuffer-in-javascript
            const recordBytes = new Uint8Array(recordStr.match(/[\da-f]{2}/gi).map((h)=>parseInt(h, 16)));

            const recordLength = recordBytes[0];
            if (recordLength + 4 !== recordBytes.length) {
                throw new Error('Mismatched record length at record ' + recordCount + ' (' + matchResult[0].trim() + '), expected ' + (recordLength) + ' data bytes but actual length is ' + (recordBytes.length - 4));
            }

            const cs = checksum(recordBytes);
            if (parseInt(recordChecksum, 16) !== cs) {
                throw new Error('Checksum failed at record ' + recordCount + ' (' + matchResult[0].trim() + '), should be ' + cs.toString(16) );
            }

            const offset = (recordBytes[1] << 8) + recordBytes[2];
            const recordType = recordBytes[3];
            const data = recordBytes.subarray(4);

            if (recordType === 0) {
                // Data record, contains data
                // Create a new block, at (upper linear base address + offset)
                if (blocks.has(ulba + offset)) {
                    throw new Error('Duplicated data at record ' + recordCount + ' (' + matchResult[0].trim() + ')');
                }
                if (offset + data.length > 0x10000) {
                    throw new Error(
                        'Data at record ' +
                        recordCount +
                        ' (' +
                        matchResult[0].trim() +
                        ') wraps over 0xFFFF. This would trigger ambiguous behaviour. Please restructure your data so that for every record the data offset plus the data length do not exceed 0xFFFF.');
                }

                blocks.set( ulba + offset, data );

            } else {

                // All non-data records must have a data offset of zero
                if (offset !== 0) {
                    throw new Error('Record ' + recordCount + ' (' + matchResult[0].trim() + ') must have 0000 as data offset.');
                }

                switch (recordType) {
                case 1: // EOF
                    if (lastCharacterParsed !== hexText.length) {
                        // This record should be at the very end of the string
                        throw new Error('There is data after an EOF record at record ' + recordCount);
                    }

                    return blocks.join(maxBlockSize);

                case 2: // Extended Segment Address Record
                    // Sets the 16 most significant bits of the 20-bit Segment Base
                    // Address for the subsequent data.
                    ulba = ((data[0] << 8) + data[1]) << 4;
                    break;

                case 3: // Start Segment Address Record
                    // Do nothing. Record type 3 only applies to 16-bit Intel CPUs,
                    // where it should reset the program counter (CS+IP CPU registers)
                    break;

                case 4: // Extended Linear Address Record
                    // Sets the 16 most significant (upper) bits of the 32-bit Linear Address
                    // for the subsequent data
                    ulba = ((data[0] << 8) + data[1]) << 16;
                    break;

                case 5: // Start Linear Address Record
                    // Do nothing. Record type 5 only applies to 32-bit Intel CPUs,
                    // where it should reset the program counter (EIP CPU register)
                    // It might have meaning for other CPU architectures
                    // (see http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka9903.html )
                    // but will be ignored nonetheless.
                    break;
                default:
                    throw new Error('Invalid record type 0x' + hexpad(recordType) + ' at record ' + recordCount + ' (should be between 0x00 and 0x05)');
                }
            }
        }

        if (recordCount) {
            throw new Error('No EOF record at end of file');
        } else {
            throw new Error('Malformed .hex file, could not parse any registers');
        }
    }


    /**
     * Returns a <strong>new</strong> instance of {@linkcode MemoryMap}, containing
     * the same data, but concatenating together those memory blocks that are adjacent.
     *<br/>
     * The insertion order of keys in the {@linkcode MemoryMap} is guaranteed to be strictly
     * ascending. In other words, when iterating through the {@linkcode MemoryMap}, the addresses
     * will be ordered in ascending order.
     *<br/>
     * If <tt>maxBlockSize</tt> is given, blocks will be concatenated together only
     * until the joined block reaches this size in bytes. This means that the output
     * {@linkcode MemoryMap} might have more entries than the input one.
     *<br/>
     * If there is any overlap between blocks, an error will be thrown.
     *<br/>
     * The returned {@linkcode MemoryMap} will use newly allocated memory.
     *
     * @param {Number} [maxBlockSize=Infinity] Maximum size of the <tt>Uint8Array</tt>s in the
     * returned {@linkcode MemoryMap}.
     *
     * @return {MemoryMap}
     */
    join(maxBlockSize = Infinity) {

        // First pass, create a Map of address→length of contiguous blocks
        const sortedKeys = Array.from(this.keys()).sort((a,b)=>a-b);
        const blockSizes = new Map();
        let lastBlockAddr = -1;
        let lastBlockEndAddr = -1;

        for (let i=0,l=sortedKeys.length; i<l; i++) {
            const blockAddr = sortedKeys[i];
            const blockLength = this.get(sortedKeys[i]).length;

            if (lastBlockEndAddr === blockAddr && (lastBlockEndAddr - lastBlockAddr) < maxBlockSize) {
                // Grow when the previous end address equals the current,
                // and we don't go over the maximum block size.
                blockSizes.set(lastBlockAddr, blockSizes.get(lastBlockAddr) + blockLength);
                lastBlockEndAddr += blockLength;
            } else if (lastBlockEndAddr <= blockAddr) {
                // Else mark a new block.
                blockSizes.set(blockAddr, blockLength);
                lastBlockAddr = blockAddr;
                lastBlockEndAddr = blockAddr + blockLength;
            } else {
                throw new Error('Overlapping data around address 0x' + blockAddr.toString(16));
            }
        }

        // Second pass: allocate memory for the contiguous blocks and copy data around.
        const mergedBlocks = new MemoryMap();
        let mergingBlock;
        let mergingBlockAddr = -1;
        for (let i=0,l=sortedKeys.length; i<l; i++) {
            const blockAddr = sortedKeys[i];
            if (blockSizes.has(blockAddr)) {
                mergingBlock = new Uint8Array(blockSizes.get(blockAddr));
                mergedBlocks.set(blockAddr, mergingBlock);
                mergingBlockAddr = blockAddr;
            }
            mergingBlock.set(this.get(blockAddr), blockAddr - mergingBlockAddr);
        }

        return mergedBlocks;
    }

    /**
     * Given a {@link https://developer.mozilla.org/docs/Web/JavaScript/Reference/Global_Objects/Map|<tt>Map</tt>}
     * of {@linkcode MemoryMap}s, indexed by a alphanumeric ID,
     * returns a <tt>Map</tt> of address to tuples (<tt>Arrays</tt>s of length 2) of the form
     * <tt>(id, Uint8Array)</tt>s.
     *<br/>
     * The scenario for using this is having several {@linkcode MemoryMap}s, from several calls to
     * {@link module:nrf-intel-hex~hexToArrays|hexToArrays}, each having a different identifier.
     * This function locates where those memory block sets overlap, and returns a <tt>Map</tt>
     * containing addresses as keys, and arrays as values. Each array will contain 1 or more
     * <tt>(id, Uint8Array)</tt> tuples: the identifier of the memory block set that has
     * data in that region, and the data itself. When memory block sets overlap, there will
     * be more than one tuple.
     *<br/>
     * The <tt>Uint8Array</tt>s in the output are
     * {@link https://developer.mozilla.org/docs/Web/JavaScript/Reference/Global_Objects/TypedArray/subarray|subarrays}
     * of the input data; new memory is <strong>not</strong> allocated for them.
     *<br/>
     * The insertion order of keys in the output <tt>Map</tt> is guaranteed to be strictly
     * ascending. In other words, when iterating through the <tt>Map</tt>, the addresses
     * will be ordered in ascending order.
     *<br/>
     * When two blocks overlap, the corresponding array of tuples will have the tuples ordered
     * in the insertion order of the input <tt>Map</tt> of block sets.
     *<br/>
     *
     * @param {Map.MemoryMap} memoryMaps The input memory block sets
     *
     * @example
     * import MemoryMap from 'nrf-intel-hex';
     *
     * let memMap1 = MemoryMap.fromHex( hexdata1 );
     * let memMap2 = MemoryMap.fromHex( hexdata2 );
     * let memMap3 = MemoryMap.fromHex( hexdata3 );
     *
     * let maps = new Map([
     *  ['file A', blocks1],
     *  ['file B', blocks2],
     *  ['file C', blocks3]
     * ]);
     *
     * let overlappings = MemoryMap.overlapMemoryMaps(maps);
     *
     * for (let [address, tuples] of overlappings) {
     *     // if 'tuples' has length > 1, there is an overlap starting at 'address'
     *
     *     for (let [address, tuples] of overlappings) {
     *         let [id, bytes] = tuple;
     *         // 'id' in this example is either 'file A', 'file B' or 'file C'
     *     }
     * }
     * @return {Map.Array<mixed,Uint8Array>} The map of possibly overlapping memory blocks
     */
    static overlapMemoryMaps(memoryMaps) {
        // First pass: create a list of addresses where any block starts or ends.
        const cuts = new Set();
        for (const [, blocks] of memoryMaps) {
            for (const [address, block] of blocks) {
                cuts.add(address);
                cuts.add(address + block.length);
            }
        }

        const orderedCuts = Array.from(cuts.values()).sort((a,b)=>a-b);
        const overlaps = new Map();

        // Second pass: iterate through the cuts, get slices of every intersecting blockset
        for (let i=0, l=orderedCuts.length-1; i<l; i++) {
            const cut = orderedCuts[i];
            const nextCut = orderedCuts[i+1];
            const tuples = [];

            for (const [setId, blocks] of memoryMaps) {
                // Find the block with the highest address that is equal or lower to
                // the current cut (if any)
                const blockAddr = Array.from(blocks.keys()).reduce((acc, val)=>{
                    if (val > cut) {
                        return acc;
                    }
                    return Math.max( acc, val );
                }, -1);

                if (blockAddr !== -1) {
                    const block = blocks.get(blockAddr);
                    const subBlockStart = cut - blockAddr;
                    const subBlockEnd = nextCut - blockAddr;

                    if (subBlockStart < block.length) {
                        tuples.push([ setId, block.subarray(subBlockStart, subBlockEnd) ]);
                    }
                }
            }

            if (tuples.length) {
                overlaps.set(cut, tuples);
            }
        }

        return overlaps;
    }


    /**
     * Given the output of the {@linkcode MemoryMap.overlapMemoryMaps|overlapMemoryMaps}
     * (a <tt>Map</tt> of address to an <tt>Array</tt> of <tt>(id, Uint8Array)</tt> tuples),
     * returns a {@linkcode MemoryMap}. This discards the IDs in the process.
     *<br/>
     * The output <tt>Map</tt> contains as many entries as the input one (using the same addresses
     * as keys), but the value for each entry will be the <tt>Uint8Array</tt> of the <b>last</b>
     * tuple for each address in the input data.
     *<br/>
     * The scenario is wanting to join together several parsed .hex files, not worrying about
     * their overlaps.
     *<br/>
     *
     * @param {Map.Array<mixed,Uint8Array>} overlaps The (possibly overlapping) input memory blocks
     * @return {MemoryMap} The flattened memory blocks
     */
    static flattenOverlaps(overlaps) {
        return new MemoryMap(
            Array.from(overlaps.entries()).map(([address, tuples]) => {
                return [address, tuples[tuples.length - 1][1] ];
            })
        );
    }


    /**
     * Returns a new instance of {@linkcode MemoryMap}, where:
     *
     * <ul>
     *  <li>Each key (the start address of each <tt>Uint8Array</tt>) is a multiple of
     *    <tt>pageSize</tt></li>
     *  <li>The size of each <tt>Uint8Array</tt> is exactly <tt>pageSize</tt></li>
     *  <li>Bytes from the input map to bytes in the output</li>
     *  <li>Bytes not in the input are replaced by a padding value</li>
     * </ul>
     *<br/>
     * The scenario is wanting to prepare pages of bytes for a write operation, where the write
     * operation affects a whole page/sector at once.
     *<br/>
     * The insertion order of keys in the output {@linkcode MemoryMap} is guaranteed
     * to be strictly ascending. In other words, when iterating through the
     * {@linkcode MemoryMap}, the addresses will be ordered in ascending order.
     *<br/>
     * The <tt>Uint8Array</tt>s in the output will be newly allocated.
     *<br/>
     *
     * @param {Number} [pageSize=1024] The size of the output pages, in bytes
     * @param {Number} [pad=0xFF] The byte value to use for padding
     * @return {MemoryMap}
     */
    paginate( pageSize=1024, pad=0xFF) {
        if (pageSize <= 0) {
            throw new Error('Page size must be greater than zero');
        }
        const outPages = new MemoryMap();
        let page;

        const sortedKeys = Array.from(this.keys()).sort((a,b)=>a-b);

        for (let i=0,l=sortedKeys.length; i<l; i++) {
            const blockAddr = sortedKeys[i];
            const block = this.get(blockAddr);
            const blockLength = block.length;
            const blockEnd = blockAddr + blockLength;

            for (let pageAddr = blockAddr - (blockAddr % pageSize); pageAddr < blockEnd; pageAddr += pageSize) {
                page = outPages.get(pageAddr);
                if (!page) {
                    page = new Uint8Array(pageSize);
                    page.fill(pad);
                    outPages.set(pageAddr, page);
                }

                const offset = pageAddr - blockAddr;
                let subBlock;
                if (offset <= 0) {
                    // First page which intersects the block
                    subBlock = block.subarray(0, Math.min(pageSize + offset, blockLength));
                    page.set(subBlock, -offset);
                } else {
                    // Any other page which intersects the block
                    subBlock = block.subarray(offset, offset + Math.min(pageSize, blockLength - offset));
                    page.set(subBlock, 0);
                }
            }
        }

        return outPages;
    }


    /**
     * Locates the <tt>Uint8Array</tt> which contains the given offset,
     * and returns the four bytes held at that offset, as a 32-bit unsigned integer.
     *
     *<br/>
     * Behaviour is similar to {@linkcode https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/DataView/getUint32|DataView.prototype.getUint32},
     * except that this operates over a {@linkcode MemoryMap} instead of
     * over an <tt>ArrayBuffer</tt>, and that this may return <tt>undefined</tt> if
     * the address is not <em>entirely</em> contained within one of the <tt>Uint8Array</tt>s.
     *<br/>
     *
     * @param {Number} offset The memory offset to read the data
     * @param {Boolean} [littleEndian=false] Whether to fetch the 4 bytes as a little- or big-endian integer
     * @return {Number|undefined} An unsigned 32-bit integer number
     */
    getUint32(offset, littleEndian) {
        const keys = Array.from(this.keys());

        for (let i=0,l=keys.length; i<l; i++) {
            const blockAddr = keys[i];
            const block = this.get(blockAddr);
            const blockLength = block.length;
            const blockEnd = blockAddr + blockLength;

            if (blockAddr <= offset && (offset+4) <= blockEnd) {
                return (new DataView(block.buffer, offset - blockAddr, 4)).getUint32(0, littleEndian);
            }
        }
        return;
    }


    /**
     * Returns a <tt>String</tt> of text representing a .hex file.
     * <br/>
     * The writer has an opinionated behaviour. Check the project's
     * {@link https://github.com/NordicSemiconductor/nrf-intel-hex#Features|README file} for details.
     *
     * @param {Number} [lineSize=16] Maximum number of bytes to be encoded in each data record.
     * Must have a value between 1 and 255, as per the specification.
     *
     * @return {String} String of text with the .hex representation of the input binary data
     *
     * @example
     * import MemoryMap from 'nrf-intel-hex';
     *
     * let memMap = new MemoryMap();
     * let bytes = new Uint8Array(....);
     * memMap.set(0x0FF80000, bytes); // The block with 'bytes' will start at offset 0x0FF80000
     *
     * let string = memMap.asHexString();
     */
    asHexString(lineSize = 16) {
        let lowAddress  = 0;    // 16 least significant bits of the current addr
        let highAddress = -1 << 16; // 16 most significant bits of the current addr
        const records = [];
        if (lineSize <=0) {
            throw new Error('Size of record must be greater than zero');
        } else if (lineSize > 255) {
            throw new Error('Size of record must be less than 256');
        }

        // Placeholders
        const offsetRecord = new Uint8Array(6);
        const recordHeader = new Uint8Array(4);

        const sortedKeys = Array.from(this.keys()).sort((a,b)=>a-b);
        for (let i=0,l=sortedKeys.length; i<l; i++) {
            const blockAddr = sortedKeys[i];
            const block = this.get(blockAddr);

            // Sanity checks
            if (!(block instanceof Uint8Array)) {
                throw new Error('Block at offset ' + blockAddr + ' is not an Uint8Array');
            }
            if (blockAddr < 0) {
                throw new Error('Block at offset ' + blockAddr + ' has a negative thus invalid address');
            }
            const blockSize = block.length;
            if (!blockSize) { continue; }   // Skip zero-length blocks


            if (blockAddr > (highAddress + 0xFFFF)) {
                // Insert a new 0x04 record to jump to a new 64KiB block

                // Round up the least significant 16 bits - no bitmasks because they trigger
                // base-2 negative numbers, whereas subtracting the modulo maintains precision
                highAddress = blockAddr - blockAddr % 0x10000;
                lowAddress = 0;

                offsetRecord[0] = 2;    // Length
                offsetRecord[1] = 0;    // Load offset, high byte
                offsetRecord[2] = 0;    // Load offset, low byte
                offsetRecord[3] = 4;    // Record type
                offsetRecord[4] = highAddress >> 24;    // new address offset, high byte
                offsetRecord[5] = highAddress >> 16;    // new address offset, low byte

                records.push(
                    ':' +
                    Array.prototype.map.call(offsetRecord, hexpad).join('') +
                    hexpad(checksum(offsetRecord))
                );
            }

            if (blockAddr < (highAddress + lowAddress)) {
                throw new Error(
                    'Block starting at 0x' +
                    blockAddr.toString(16) +
                    ' overlaps with a previous block.');
            }

            lowAddress = blockAddr % 0x10000;
            let blockOffset = 0;
            const blockEnd = blockAddr + blockSize;
            if (blockEnd > 0xFFFFFFFF) {
                throw new Error('Data cannot be over 0xFFFFFFFF');
            }

            // Loop for every 64KiB memory segment that spans this block
            while (highAddress + lowAddress < blockEnd) {

                if (lowAddress > 0xFFFF) {
                    // Insert a new 0x04 record to jump to a new 64KiB block
                    highAddress += 1 << 16; // Increase by one
                    lowAddress = 0;

                    offsetRecord[0] = 2;    // Length
                    offsetRecord[1] = 0;    // Load offset, high byte
                    offsetRecord[2] = 0;    // Load offset, low byte
                    offsetRecord[3] = 4;    // Record type
                    offsetRecord[4] = highAddress >> 24;    // new address offset, high byte
                    offsetRecord[5] = highAddress >> 16;    // new address offset, low byte

                    records.push(
                        ':' +
                        Array.prototype.map.call(offsetRecord, hexpad).join('') +
                        hexpad(checksum(offsetRecord))
                    );
                }

                let recordSize = -1;
                // Loop for every record for that spans the current 64KiB memory segment
                while (lowAddress < 0x10000 && recordSize) {
                    recordSize = Math.min(
                        lineSize,                            // Normal case
                        blockEnd - highAddress - lowAddress, // End of block
                        0x10000 - lowAddress                 // End of low addresses
                    );

                    if (recordSize) {

                        recordHeader[0] = recordSize;   // Length
                        recordHeader[1] = lowAddress >> 8;    // Load offset, high byte
                        recordHeader[2] = lowAddress;    // Load offset, low byte
                        recordHeader[3] = 0;    // Record type

                        const subBlock = block.subarray(blockOffset, blockOffset + recordSize);   // Data bytes for this record

                        records.push(
                            ':' +
                            Array.prototype.map.call(recordHeader, hexpad).join('') +
                            Array.prototype.map.call(subBlock, hexpad).join('') +
                            hexpad(checksumTwo(recordHeader, subBlock))
                        );

                        blockOffset += recordSize;
                        lowAddress += recordSize;
                    }
                }
            }
        }

        records.push(':00000001FF');    // EOF record

        return records.join('\n');
    }


    /**
     * Performs a deep copy of the current {@linkcode MemoryMap}, returning a new one
     * with exactly the same contents, but allocating new memory for each of its
     * <tt>Uint8Array</tt>s.
     *
     * @return {MemoryMap}
     */
    clone() {
        const cloned = new MemoryMap();

        for (let [addr, value] of this) {
            cloned.set(addr, new Uint8Array(value));
        }

        return cloned;
    }


    /**
     * Given one <tt>Uint8Array</tt>, looks through its contents and returns a new
     * {@linkcode MemoryMap}, stripping away those regions where there are only
     * padding bytes.
     * <br/>
     * The start of the input <tt>Uint8Array</tt> is assumed to be offset zero for the output.
     * <br/>
     * The use case here is dumping memory from a working device and try to see the
     * "interesting" memory regions it has. This assumes that there is a constant,
     * predefined padding byte value being used in the "non-interesting" regions.
     * In other words: this will work as long as the dump comes from a flash memory
     * which has been previously erased (thus <tt>0xFF</tt>s for padding), or from a
     * previously blanked HDD (thus <tt>0x00</tt>s for padding).
     * <br/>
     * This method uses <tt>subarray</tt> on the input data, and thus does not allocate memory
     * for the <tt>Uint8Array</tt>s.
     *
     * @param {Uint8Array} bytes The input data
     * @param {Number} [padByte=0xFF] The value of the byte assumed to be used as padding
     * @param {Number} [minPadLength=64] The minimum number of consecutive pad bytes to
     * be considered actual padding
     *
     * @return {MemoryMap}
     */
    static fromPaddedUint8Array(bytes, padByte=0xFF, minPadLength=64) {

        if (!(bytes instanceof Uint8Array)) {
            throw new Error('Bytes passed to fromPaddedUint8Array are not an Uint8Array');
        }

        // The algorithm used is naïve and checks every byte.
        // An obvious optimization would be to implement Boyer-Moore
        // (see https://en.wikipedia.org/wiki/Boyer%E2%80%93Moore_string_search_algorithm )
        // or otherwise start skipping up to minPadLength bytes when going through a non-pad
        // byte.
        // Anyway, we could expect a lot of cases where there is a majority of pad bytes,
        // and the algorithm should check most of them anyway, so the perf gain is questionable.

        const memMap = new MemoryMap();
        let consecutivePads = 0;
        let lastNonPad = -1;
        let firstNonPad = 0;
        let skippingBytes = false;
        const l = bytes.length;

        for (let addr = 0; addr < l; addr++) {
            const byte = bytes[addr];

            if (byte === padByte) {
                consecutivePads++;
                if (consecutivePads >= minPadLength) {
                    // Edge case: ignore writing a zero-length block when skipping
                    // bytes at the beginning of the input
                    if (lastNonPad !== -1) {
                        /// Add the previous block to the result memMap
                        memMap.set(firstNonPad, bytes.subarray(firstNonPad, lastNonPad+1));
                    }

                    skippingBytes = true;
                }
            } else {
                if (skippingBytes) {
                    skippingBytes = false;
                    firstNonPad = addr;
                }
                lastNonPad = addr;
                consecutivePads = 0;
            }
        }

        // At EOF, add the last block if not skipping bytes already (and input not empty)
        if (!skippingBytes && lastNonPad !== -1) {
            memMap.set(firstNonPad, bytes.subarray(firstNonPad, l));
        }

        return memMap;
    }


    /**
     * Returns a new instance of {@linkcode MemoryMap}, containing only data between
     * the addresses <tt>address</tt> and <tt>address + length</tt>.
     * Behaviour is similar to {@linkcode https://developer.mozilla.org/docs/Web/JavaScript/Reference/Global_Objects/Array/slice|Array.prototype.slice},
     * in that the return value is a portion of the current {@linkcode MemoryMap}.
     *
     * <br/>
     * The returned {@linkcode MemoryMap} might be empty.
     *
     * <br/>
     * Internally, this uses <tt>subarray</tt>, so new memory is not allocated.
     *
     * @param {Number} address The start address of the slice
     * @param {Number} length The length of memory map to slice out
     * @return {MemoryMap}
     */
    slice(address, length = Infinity){
        if (length < 0) {
            throw new Error('Length of the slice cannot be negative');
        }

        const sliced = new MemoryMap();

        for (let [blockAddr, block] of this) {
            const blockLength = block.length;

            if ((blockAddr + blockLength) >= address && blockAddr < (address + length)) {
                const sliceStart = Math.max(address, blockAddr);
                const sliceEnd = Math.min(address + length, blockAddr + blockLength);
                const sliceLength = sliceEnd - sliceStart;
                const relativeSliceStart = sliceStart - blockAddr;

                if (sliceLength > 0) {
                    sliced.set(sliceStart, block.subarray(relativeSliceStart, relativeSliceStart + sliceLength));
                }
            }
        }
        return sliced;
    }

    /**
     * Returns a new instance of {@linkcode https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/DataView/getUint32|Uint8Array}, containing only data between
     * the addresses <tt>address</tt> and <tt>address + length</tt>. Any byte without a value
     * in the input {@linkcode MemoryMap} will have a value of <tt>padByte</tt>.
     *
     * <br/>
     * This method allocates new memory.
     *
     * @param {Number} address The start address of the slice
     * @param {Number} length The length of memory map to slice out
     * @param {Number} [padByte=0xFF] The value of the byte assumed to be used as padding
     * @return {MemoryMap}
     */
    slicePad(address, length, padByte=0xFF){
        if (length < 0) {
            throw new Error('Length of the slice cannot be negative');
        }
        
        const out = (new Uint8Array(length)).fill(padByte);

        for (let [blockAddr, block] of this) {
            const blockLength = block.length;

            if ((blockAddr + blockLength) >= address && blockAddr < (address + length)) {
                const sliceStart = Math.max(address, blockAddr);
                const sliceEnd = Math.min(address + length, blockAddr + blockLength);
                const sliceLength = sliceEnd - sliceStart;
                const relativeSliceStart = sliceStart - blockAddr;

                if (sliceLength > 0) {
                    out.set(block.subarray(relativeSliceStart, relativeSliceStart + sliceLength), sliceStart - address);
                }
            }
        }
        return out;
    }

    /**
     * Checks whether the current memory map contains the one given as a parameter.
     *
     * <br/>
     * "Contains" means that all the offsets that have a byte value in the given
     * memory map have a value in the current memory map, and that the byte values
     * are the same.
     *
     * <br/>
     * An empty memory map is always contained in any other memory map.
     *
     * <br/>
     * Returns boolean <tt>true</tt> if the memory map is contained, <tt>false</tt>
     * otherwise.
     *
     * @param {MemoryMap} memMap The memory map to check
     * @return {Boolean}
     */
    contains(memMap) {
        for (let [blockAddr, block] of memMap) {

            const blockLength = block.length;

            const slice = this.slice(blockAddr, blockLength).join().get(blockAddr);

            if ((!slice) || slice.length !== blockLength ) {
                return false;
            }

            for (const i in block) {
                if (block[i] !== slice[i]) {
                    return false;
                }
            }
        }
        return true;
    }
}


export default MemoryMap;