@gmod/tabix 1.5.0 → 1.5.3

package/src/tbi.ts

@@ -0,0 +1,250 @@
+import Long from 'long'
+import VirtualOffset, { fromBytes } from './virtualOffset'
+import Chunk from './chunk'
+import { unzip } from '@gmod/bgzf-filehandle'
+import { longToNumber, optimizeChunks, checkAbortSignal } from './util'
+import IndexFile, { Options } from './indexFile'
+
+const TBI_MAGIC = 21578324 // TBI\1
+const TAD_LIDX_SHIFT = 14
+
+/**
+ * calculate the list of bins that may overlap with region [beg,end) (zero-based half-open)
+ */
+function reg2bins(beg: number, end: number) {
+  beg += 1 // < convert to 1-based closed
+  end -= 1
+  return [
+    [0, 0],
+    [1 + (beg >> 26), 1 + (end >> 26)],
+    [9 + (beg >> 23), 9 + (end >> 23)],
+    [73 + (beg >> 20), 73 + (end >> 20)],
+    [585 + (beg >> 17), 585 + (end >> 17)],
+    [4681 + (beg >> 14), 4681 + (end >> 14)],
+  ]
+}
+
+export default class TabixIndex extends IndexFile {
+  async lineCount(refName: string, opts: Options = {}) {
+    const indexData = await this.parse(opts)
+    if (!indexData) {
+      return -1
+    }
+    const refId = indexData.refNameToId[refName]
+    const idx = indexData.indices[refId]
+    if (!idx) {
+      return -1
+    }
+    const { stats } = indexData.indices[refId]
+    if (stats) {
+      return stats.lineCount
+    }
+    return -1
+  }
+
+  // memoize
+  // fetch and parse the index
+  async _parse(opts: Options = {}) {
+    const bytes = await unzip((await this.filehandle.readFile(opts)) as Buffer)
+    checkAbortSignal(opts.signal)
+
+    // check TBI magic numbers
+    if (bytes.readUInt32LE(0) !== TBI_MAGIC /* "TBI\1" */) {
+      throw new Error('Not a TBI file')
+      // TODO: do we need to support big-endian TBI files?
+    }
+
+    // number of reference sequences in the index
+    const refCount = bytes.readInt32LE(4)
+    const formatFlags = bytes.readInt32LE(8)
+    const coordinateType =
+      formatFlags & 0x10000 ? 'zero-based-half-open' : '1-based-closed'
+    const formatOpts: { [key: number]: string } = {
+      0: 'generic',
+      1: 'SAM',
+      2: 'VCF',
+    }
+    const format = formatOpts[formatFlags & 0xf]
+    if (!format) {
+      throw new Error(`invalid Tabix preset format flags ${formatFlags}`)
+    }
+    const columnNumbers = {
+      ref: bytes.readInt32LE(12),
+      start: bytes.readInt32LE(16),
+      end: bytes.readInt32LE(20),
+    }
+    const metaValue = bytes.readInt32LE(24)
+    const depth = 5
+    const maxBinNumber = ((1 << ((depth + 1) * 3)) - 1) / 7
+    const maxRefLength = 2 ** (14 + depth * 3)
+    const metaChar = metaValue ? String.fromCharCode(metaValue) : null
+    const skipLines = bytes.readInt32LE(28)
+
+    // read sequence dictionary
+    const nameSectionLength = bytes.readInt32LE(32)
+    const { refNameToId, refIdToName } = this._parseNameBytes(
+      bytes.slice(36, 36 + nameSectionLength),
+    )
+
+    // read the indexes for each reference sequence
+    let currOffset = 36 + nameSectionLength
+    let firstDataLine: VirtualOffset | undefined
+    const indices = new Array(refCount).fill(0).map(() => {
+      // the binning index
+      const binCount = bytes.readInt32LE(currOffset)
+      currOffset += 4
+      const binIndex: { [key: number]: Chunk[] } = {}
+      let stats
+      for (let j = 0; j < binCount; j += 1) {
+        const bin = bytes.readUInt32LE(currOffset)
+        currOffset += 4
+        if (bin > maxBinNumber + 1) {
+          throw new Error(
+            'tabix index contains too many bins, please use a CSI index',
+          )
+        } else if (bin === maxBinNumber + 1) {
+          const chunkCount = bytes.readInt32LE(currOffset)
+          currOffset += 4
+          if (chunkCount === 2) {
+            stats = this.parsePseudoBin(bytes, currOffset)
+          }
+          currOffset += 16 * chunkCount
+        } else {
+          const chunkCount = bytes.readInt32LE(currOffset)
+          currOffset += 4
+          const chunks = new Array(chunkCount)
+          for (let k = 0; k < chunkCount; k += 1) {
+            const u = fromBytes(bytes, currOffset)
+            const v = fromBytes(bytes, currOffset + 8)
+            currOffset += 16
+            firstDataLine = this._findFirstData(firstDataLine, u)
+            chunks[k] = new Chunk(u, v, bin)
+          }
+          binIndex[bin] = chunks
+        }
+      }
+
+      // the linear index
+      const linearCount = bytes.readInt32LE(currOffset)
+      currOffset += 4
+      const linearIndex = new Array(linearCount)
+      for (let k = 0; k < linearCount; k += 1) {
+        linearIndex[k] = fromBytes(bytes, currOffset)
+        currOffset += 8
+        firstDataLine = this._findFirstData(firstDataLine, linearIndex[k])
+      }
+      return { binIndex, linearIndex, stats }
+    })
+
+    return {
+      indices,
+      metaChar,
+      maxBinNumber,
+      maxRefLength,
+      skipLines,
+      firstDataLine,
+      columnNumbers,
+      coordinateType,
+      format,
+      refIdToName,
+      refNameToId,
+      maxBlockSize: 1 << 16,
+    }
+  }
+
+  parsePseudoBin(bytes: Buffer, offset: number) {
+    const lineCount = longToNumber(
+      Long.fromBytesLE(
+        bytes.slice(offset + 16, offset + 24) as unknown as number[],
+        true,
+      ),
+    )
+    return { lineCount }
+  }
+
+  _parseNameBytes(namesBytes: Buffer) {
+    let currRefId = 0
+    let currNameStart = 0
+    const refIdToName: string[] = []
+    const refNameToId: { [key: string]: number } = {}
+    for (let i = 0; i < namesBytes.length; i += 1) {
+      if (!namesBytes[i]) {
+        if (currNameStart < i) {
+          let refName = namesBytes.toString('utf8', currNameStart, i)
+          refName = this.renameRefSeq(refName)
+          refIdToName[currRefId] = refName
+          refNameToId[refName] = currRefId
+        }
+        currNameStart = i + 1
+        currRefId += 1
+      }
+    }
+    return { refNameToId, refIdToName }
+  }
+
+  async blocksForRange(
+    refName: string,
+    min: number,
+    max: number,
+    opts: Options = {},
+  ) {
+    if (min < 0) {
+      min = 0
+    }
+
+    const indexData = await this.parse(opts)
+    if (!indexData) {
+      return []
+    }
+    const refId = indexData.refNameToId[refName]
+    const ba = indexData.indices[refId]
+    if (!ba) {
+      return []
+    }
+
+    const minOffset = ba.linearIndex.length
+      ? ba.linearIndex[
+          min >> TAD_LIDX_SHIFT >= ba.linearIndex.length
+            ? ba.linearIndex.length - 1
+            : min >> TAD_LIDX_SHIFT
+        ]
+      : new VirtualOffset(0, 0)
+    if (!minOffset) {
+      console.warn('querying outside of possible tabix range')
+    }
+
+    // const { linearIndex, binIndex } = indexes
+
+    const overlappingBins = reg2bins(min, max) // List of bin #s that overlap min, max
+    const chunks: Chunk[] = []
+
+    // Find chunks in overlapping bins.  Leaf bins (< 4681) are not pruned
+    for (const [start, end] of overlappingBins) {
+      for (let bin = start; bin <= end; bin++) {
+        if (ba.binIndex[bin]) {
+          const binChunks = ba.binIndex[bin]
+          for (let c = 0; c < binChunks.length; ++c) {
+            chunks.push(new Chunk(binChunks[c].minv, binChunks[c].maxv, bin))
+          }
+        }
+      }
+    }
+
+    // Use the linear index to find minimum file position of chunks that could
+    // contain alignments in the region
+    const nintv = ba.linearIndex.length
+    let lowest = null
+    const minLin = Math.min(min >> 14, nintv - 1)
+    const maxLin = Math.min(max >> 14, nintv - 1)
+    for (let i = minLin; i <= maxLin; ++i) {
+      const vp = ba.linearIndex[i]
+      if (vp) {
+        if (!lowest || vp.compareTo(lowest) < 0) {
+          lowest = vp
+        }
+      }
+    }
+
+    return optimizeChunks(chunks, lowest)
+  }
+}

package/src/util.ts

@@ -0,0 +1,103 @@
+import Chunk from './chunk'
+import VirtualOffset from './virtualOffset'
+
+export function longToNumber(long: Long) {
+  if (
+    long.greaterThan(Number.MAX_SAFE_INTEGER) ||
+    long.lessThan(Number.MIN_SAFE_INTEGER)
+  ) {
+    throw new Error('integer overflow')
+  }
+  return long.toNumber()
+}
+
+class AbortError extends Error {
+  public code: string | undefined
+}
+/**
+ * Properly check if the given AbortSignal is aborted.
+ * Per the standard, if the signal reads as aborted,
+ * this function throws either a DOMException AbortError, or a regular error
+ * with a `code` attribute set to `ERR_ABORTED`.
+ *
+ * For convenience, passing `undefined` is a no-op
+ *
+ * @param {AbortSignal} [signal] an AbortSignal, or anything with an `aborted` attribute
+ * @returns nothing
+ */
+export function checkAbortSignal(signal?: AbortSignal) {
+  if (!signal) {
+    return
+  }
+
+  if (signal.aborted) {
+    // console.log('bam aborted!')
+    if (typeof DOMException !== 'undefined') {
+      // eslint-disable-next-line  no-undef
+      throw new DOMException('aborted', 'AbortError')
+    } else {
+      const e = new AbortError('aborted')
+      e.code = 'ERR_ABORTED'
+      throw e
+    }
+  }
+}
+
+/**
+ * Skips to the next tick, then runs `checkAbortSignal`.
+ * Await this to inside an otherwise synchronous loop to
+ * provide a place to break when an abort signal is received.
+ * @param {AbortSignal} signal
+ */
+export async function abortBreakPoint(signal?: AbortSignal) {
+  await Promise.resolve()
+  checkAbortSignal(signal)
+}
+
+export function canMergeBlocks(chunk1: Chunk, chunk2: Chunk) {
+  return (
+    chunk2.minv.blockPosition - chunk1.maxv.blockPosition < 65000 &&
+    chunk2.maxv.blockPosition - chunk1.minv.blockPosition < 5000000
+  )
+}
+
+export function optimizeChunks(chunks: Chunk[], lowest: VirtualOffset) {
+  const mergedChunks: Chunk[] = []
+  let lastChunk: Chunk | null = null
+
+  if (chunks.length === 0) {
+    return chunks
+  }
+
+  chunks.sort(function (c0, c1) {
+    const dif = c0.minv.blockPosition - c1.minv.blockPosition
+    if (dif !== 0) {
+      return dif
+    } else {
+      return c0.minv.dataPosition - c1.minv.dataPosition
+    }
+  })
+
+  chunks.forEach(chunk => {
+    if (!lowest || chunk.maxv.compareTo(lowest) > 0) {
+      if (lastChunk === null) {
+        mergedChunks.push(chunk)
+        lastChunk = chunk
+      } else {
+        if (canMergeBlocks(lastChunk, chunk)) {
+          if (chunk.maxv.compareTo(lastChunk.maxv) > 0) {
+            lastChunk.maxv = chunk.maxv
+          }
+        } else {
+          mergedChunks.push(chunk)
+          lastChunk = chunk
+        }
+      }
+    }
+    // else {
+    //   console.log(`skipping chunk ${chunk}`)
+    // }
+  })
+
+  return mergedChunks
+}

package/src/virtualOffset.ts

@@ -0,0 +1,47 @@
+export default class VirtualOffset {
+  public blockPosition: number
+  public dataPosition: number
+  constructor(blockPosition: number, dataPosition: number) {
+    this.blockPosition = blockPosition // < offset of the compressed data block
+    this.dataPosition = dataPosition // < offset into the uncompressed data
+  }
+
+  toString() {
+    return `${this.blockPosition}:${this.dataPosition}`
+  }
+
+  compareTo(b: VirtualOffset) {
+    return (
+      this.blockPosition - b.blockPosition || this.dataPosition - b.dataPosition
+    )
+  }
+
+  static min(...args: VirtualOffset[]) {
+    let min
+    let i = 0
+    for (; !min; i += 1) {
+      min = args[i]
+    }
+    for (; i < args.length; i += 1) {
+      if (min.compareTo(args[i]) > 0) {
+        min = args[i]
+      }
+    }
+    return min
+  }
+}
+export function fromBytes(bytes: Buffer, offset = 0, bigendian = false) {
+  if (bigendian) {
+    throw new Error('big-endian virtual file offsets not implemented')
+  }
+
+  return new VirtualOffset(
+    bytes[offset + 7] * 0x10000000000 +
+      bytes[offset + 6] * 0x100000000 +
+      bytes[offset + 5] * 0x1000000 +
+      bytes[offset + 4] * 0x10000 +
+      bytes[offset + 3] * 0x100 +
+      bytes[offset + 2],
+    (bytes[offset + 1] << 8) | bytes[offset],
+  )
+}

package/dist/declare.d.js

@@ -1,2 +0,0 @@
-"use strict";
-//# sourceMappingURL=data:application/json;charset=utf-8;base64,eyJ2ZXJzaW9uIjozLCJzb3VyY2VzIjpbXSwibmFtZXMiOltdLCJtYXBwaW5ncyI6IiIsInNvdXJjZXNDb250ZW50IjpbXX0=