Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 | 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 2x 64595x 64595x 29342x 29342x 29342x 29342x 64595x 17579x 17579x 17579x 17579x 17579x 17579x 17674x 64595x 64595x 2x 2x 2x 384x 384x 384x 2x 348x 1x 1x 347x 348x 347x 348x 347x 348x 347x 347x 347x 348x 347x 348x 681x 681x 681x 681x 681x 680x 680x 680x 680x 680x 680x 346x 346x 346x 346x 350x 350x 1026x 342x 1026x 4x 4x 2x 2x 2x 2x 1107022x 1107022x 2x 258399x 258399x 2x 271402x 271402x 271402x 271402x 271402x 271402x 579174x 579174x 579174x 271402x 271402x 2x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 2434x 271822x 271822x 271798x 271798x 271599x 271595x 271595x 271595x 271595x 271595x 271595x 271822x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271482x 271822x 271822x 271332x 271332x 2434x 2x 88176x 88176x 88176x 2x 2x 61578x 61578x 61578x 61578x 61578x 61578x 3078195x 3078195x 3078195x 61578x 61578x 2x 49863x 49863x 49863x 49863x 49863x 49863x 49863x 49863x 33238x 33238x 33238x 33238x 8291x 8291x 8291x 8291x 8291x 8287x 8290x 8287x 8287x 8287x 8287x 33238x 16623x 16623x 16623x 16623x 16623x 16623x 33235x 8306x 8306x 8306x 8306x 18x 24x 1x 1x 1x 49842x 49842x 2x 2x 2x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 16579x 2x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 1x 1x 1x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 49858x 1x 1x 1x 1x 1x 2x 228631x 29296x 29296x 29296x 228631x 51x 3x 3x 3x 228631x 228631x | /**
* BAM: the header, and one alignment record at a time.
*
* ## Why this is written here
*
* The same argument `hic/format.ts` makes about `hic-straw` and `tabix.ts`
* makes about `@gmod/tabix`, and it is not repeated at length: every read in
* this application goes through a source that counts its requests, caches its
* chunks and signs them where a bucket needs it, and a parser that fetches for
* itself sits outside all three. `@gmod/bam` is good and is a devDependency
* here for exactly one purpose — to be the oracle these decoders are tested
* against, the way `hic-straw` is for `.hic` — because a reader tested against
* itself is tested against nothing.
*
* What is left once the fetching is somebody else's problem is two structures:
* a header that is a length-prefixed string and a table, and a record that is a
* fixed 36-byte prefix followed by five variable-length blocks. Neither is
* large and neither is subtle.
*
* ## What is decoded, and what is stepped over
*
* A coverage pass over a deep window touches a few hundred thousand records and
* wants four numbers from each of them. Building an object per record to throw
* away eleven of its fields is most of the cost of the pass, so the reading
* comes in two shapes deliberately: {@link forEachRecord} walks the buffer and
* hands out offsets, allocating nothing, and {@link decodeRecord} builds the
* whole record for the one read a reader pointed at. The first is what a lane
* is drawn from; the second is what a dialog is filled from.
*
* `end` is the exception that is always computed. A record stores where it
* begins and how it aligns, never where it ends, so the reference span — the
* only thing a coverage pass actually needs — costs a walk of the CIGAR whether
* anybody asked for it or not.
*/
/** `BAM\1`, and the whole of the is-this-a-BAM question once BGZF is inflated. */
const BAM_MAGIC = [0x42, 0x41, 0x4d, 0x01]
/** The fixed part of an alignment record, before `read_name` begins. */
export const RECORD_PREFIX_BYTES = 32
/**
* The flag bits, named.
*
* Spelled out rather than left as literals at the call sites because six of
* them are read in three different modules and `(flag & 0x900) !== 0` is not a
* sentence anybody should have to decode twice.
*/
export const FLAG = {
PAIRED: 0x1,
PROPER_PAIR: 0x2,
UNMAPPED: 0x4,
MATE_UNMAPPED: 0x8,
REVERSE: 0x10,
MATE_REVERSE: 0x20,
FIRST_IN_PAIR: 0x40,
SECOND_IN_PAIR: 0x80,
SECONDARY: 0x100,
QC_FAIL: 0x200,
DUPLICATE: 0x400,
SUPPLEMENTARY: 0x800,
} as const
/**
* Step over one tag value, returning where the next tag begins, or `-1`.
*
* Here rather than beside either of its two readers — a read's `MD` for its
* mismatches, and a spliced read's `XS` for the strand its intron is on — so a
* tag type one of them learns to step over is a type the other steps over too.
*/
export function skipTag(bytes: Uint8Array, at: number, type: string, to: number): number {
const width: Record<string, number | undefined> = { c: 1, C: 1, s: 2, S: 2, i: 4, I: 4, f: 4, A: 1 }
if (type === 'Z' || type === 'H') {
let end = at
while (end < to && bytes[end] !== 0) end++
return end + 1
}
if (type === 'B') {
const sub = String.fromCharCode(bytes[at])
const count = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength).getInt32(at + 1, true)
const each = width[sub]
if (each === undefined || count < 0) return -1
return at + 5 + count * each
}
const each = width[type]
return each === undefined ? -1 : at + each
}
/**
* The reads no pile-up and no coverage should ever count.
*
* Secondary and supplementary alignments are *the same molecule* reported a
* second time, so counting them inflates depth at exactly the loci — repeats,
* fusion breakpoints — where somebody is reading the number most carefully.
* Unmapped records carry a position only as a courtesy to their mate. QC-fail
* is the aligner's own verdict.
*
* Duplicates are deliberately *not* in here: marking them is a pipeline's
* decision and reading a duplicate-marked file with them hidden by default
* would silently disagree with the same file read anywhere else. It is offered
* as a filter instead.
*/
export const DEFAULT_EXCLUDE =
FLAG.UNMAPPED | FLAG.SECONDARY | FLAG.SUPPLEMENTARY | FLAG.QC_FAIL
/** One entry of the reference table, in the file's own order. */
export interface BamRef {
name: string
length: number
}
/** What `@HD SO:` claimed, or `unknown` where there is no `@HD` line at all. */
export type SortOrder = 'coordinate' | 'queryname' | 'unsorted' | 'unknown'
export interface BamHeader {
/** The SAM text header, verbatim. Kept because `@RG` and `@PG` are read from it. */
text: string
refs: BamRef[]
sortOrder: SortOrder
/** Byte offset, within the inflated header region, where the first record begins. */
recordsFrom: number
}
/** `true` when these inflated bytes begin a BAM. */
export function isBam(bytes: Uint8Array): boolean {
if (bytes.length < BAM_MAGIC.length) return false
return BAM_MAGIC.every((byte, index) => bytes[index] === byte)
}
/**
* Parse the header out of the inflated start of a BAM.
*
* Returns `null` when the buffer stops mid-header rather than throwing, which
* is what lets the caller grow its read and try again. That is not a rare path:
* `l_text` is unbounded and a co-assembly of twenty thousand contigs writes a
* reference table of about a megabyte, spanning a dozen BGZF blocks — so
* "inflate the first block and parse the header" is wrong on precisely the
* files this viewer exists for.
*/
export function readHeader(bytes: Uint8Array): BamHeader | null {
if (!isBam(bytes)) {
throw new Error('This is not a BAM: the first four bytes are not BAM\\1.')
}
const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength)
if (bytes.length < 8) return null
const textBytes = view.getInt32(4, true)
if (textBytes < 0) throw new Error('This BAM header declares a negative length.')
let at = 8 + textBytes
if (bytes.length < at + 4) return null
const text = new TextDecoder().decode(bytes.subarray(8, at))
const refCount = view.getInt32(at, true)
at += 4
if (refCount < 0) throw new Error('This BAM header declares a negative reference count.')
const refs: BamRef[] = []
for (let i = 0; i < refCount; i++) {
if (bytes.length < at + 4) return null
const nameBytes = view.getInt32(at, true)
at += 4
if (nameBytes < 0) throw new Error('This BAM header declares a negative name length.')
if (bytes.length < at + nameBytes + 4) return null
// The stored name includes its own NUL terminator, which is not part of it.
const name = new TextDecoder().decode(bytes.subarray(at, at + nameBytes - 1))
at += nameBytes
const length = view.getInt32(at, true)
at += 4
refs.push({ name, length })
}
return { text, refs, sortOrder: sortOrderOf(text), recordsFrom: at }
}
/**
* What the `@HD` line's `SO:` says.
*
* Read rather than assumed because the difference is not cosmetic: every byte
* this module fetches is addressed by an index that only exists for a
* coordinate-sorted file, and a name-sorted BAM handed to a region query
* answers with whatever happened to be in those blocks. Refusing it by name is
* the only honest option, and `samtools sort` is the fix.
*/
function sortOrderOf(text: string): SortOrder {
for (const line of text.split('\n')) {
if (!line.startsWith('@HD')) continue
for (const field of line.split('\t')) {
if (!field.startsWith('SO:')) continue
const value = field.slice(3).trim()
if (value === 'coordinate' || value === 'queryname' || value === 'unsorted') return value
return 'unknown'
}
}
return 'unknown'
}
// ---------------------------------------------------------------------------
// CIGAR
// ---------------------------------------------------------------------------
/** `MIDNSHP=X`, indexed by the op code stored in the low four bits. */
export const CIGAR_OPS = 'MIDNSHP=X'
/** Ops that advance along the reference: M, D, N, =, X. */
const CONSUMES_REFERENCE = [true, false, true, true, false, false, false, true, true]
/** Ops that advance along the read: M, I, S, =, X. */
const CONSUMES_QUERY = [true, true, false, false, true, false, false, true, true]
/** Whether this op code moves along the reference. */
export function consumesReference(op: number): boolean {
return CONSUMES_REFERENCE[op] === true
}
/** Whether this op code moves along the read. */
export function consumesQuery(op: number): boolean {
return CONSUMES_QUERY[op] === true
}
/**
* How far along the reference a CIGAR reaches.
*
* Reads the packed ops straight out of the buffer rather than through an array,
* because this is the one thing computed for every record of every window and
* an `Uint32Array` view per record is an allocation per record.
*/
export function referenceLength(
view: DataView,
cigarAt: number,
opCount: number,
): number {
let span = 0
for (let i = 0; i < opCount; i++) {
const packed = view.getUint32(cigarAt + i * 4, true)
if (consumesReference(packed & 0xf)) span += packed >>> 4
}
return span
}
// ---------------------------------------------------------------------------
// Walking a buffer of records
// ---------------------------------------------------------------------------
/**
* Where one record is, and the four numbers every caller wants from it.
*
* Handed to a visitor and **reused between records** — this object is the same
* object every time, mutated in place. That is what makes a pass over three
* hundred thousand records allocate nothing, and it is why a visitor that wants
* to keep something must copy it out rather than keep the cursor.
*/
export interface RecordCursor {
/** Offset of this record's `refID` field — the start of its fixed prefix. */
at: number
/** Total bytes of this record, including its own four-byte length prefix. */
bytes: number
refId: number
pos: number
/** Reference end, exclusive. Zero-length for a record with no aligned op. */
end: number
mapq: number
flag: number
nameBytes: number
cigarOps: number
cigarAt: number
seqLength: number
seqAt: number
nextRefId: number
nextPos: number
tlen: number
}
/**
* Walk every whole record in `bytes`, from `from`.
*
* Stops cleanly at a truncated tail rather than throwing: a window is fetched
* as a byte span sized from the index and the last thing in it is routinely
* half a record belonging to the block after. The visitor returns `false` to
* stop early.
*
* `pos` is the file's own 0-based coordinate and is passed through untouched.
* A record with `pos === -1` is an unplaced read and is handed on with an `end`
* equal to its start, so a caller that does not filter it cannot accidentally
* paint it at the origin.
*
* A record whose name, CIGAR and sequence do not fit inside its own length is
* not a record, and stops the walk as a non-positive length does. Every caller
* sizes something by those fields — an `l_seq` sizes an `Int32Array` for base
* modifications and a loop that builds the read's string — so a corrupt one
* reaching a visitor was a two-billion-iteration loop over bytes that belong
* to the next record, or to nothing (§10).
*/
export function forEachRecord(
bytes: Uint8Array,
from: number,
visit: (cursor: RecordCursor) => boolean | void,
): void {
const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength)
const cursor: RecordCursor = {
at: 0, bytes: 0, refId: -1, pos: -1, end: -1, mapq: 0, flag: 0,
nameBytes: 0, cigarOps: 0, cigarAt: 0, seqLength: 0, seqAt: 0,
nextRefId: -1, nextPos: -1, tlen: 0,
}
let at = from
while (at + 4 <= bytes.length) {
const blockSize = view.getInt32(at, true)
// A non-positive block size is not a short read, it is a buffer that is not
// records — stop rather than loop forever on a zero.
if (blockSize <= 0) break
const recordAt = at + 4
if (recordAt + blockSize > bytes.length) break
if (blockSize < RECORD_PREFIX_BYTES) break
const nameBytes = view.getUint8(recordAt + 8)
const cigarOps = view.getUint16(recordAt + 12, true)
const cigarAt = recordAt + RECORD_PREFIX_BYTES + nameBytes
const seqLength = view.getInt32(recordAt + 16, true)
const pos = view.getInt32(recordAt + 4, true)
// Name, CIGAR, packed bases and qualities, in that order before the tags.
// `Math.ceil` rather than `>> 1`: an `l_seq` near 2^31 wraps under a shift.
const fixed = RECORD_PREFIX_BYTES + nameBytes + cigarOps * 4 + Math.ceil(seqLength / 2) + seqLength
if (seqLength < 0 || fixed > blockSize) break
cursor.at = recordAt
cursor.bytes = blockSize + 4
cursor.refId = view.getInt32(recordAt, true)
cursor.pos = pos
cursor.mapq = view.getUint8(recordAt + 9)
cursor.flag = view.getUint16(recordAt + 14, true)
cursor.nameBytes = nameBytes
cursor.cigarOps = cigarOps
cursor.cigarAt = cigarAt
cursor.seqLength = seqLength
cursor.seqAt = cigarAt + cigarOps * 4
cursor.nextRefId = view.getInt32(recordAt + 20, true)
cursor.nextPos = view.getInt32(recordAt + 24, true)
cursor.tlen = view.getInt32(recordAt + 28, true)
cursor.end = pos < 0 ? pos : pos + referenceLength(view, cigarAt, cigarOps)
if (visit(cursor) === false) return
at += 4 + blockSize
}
}
/**
* The name of the read a cursor is on.
*
* Separate from the walk because it allocates a string, which is the one thing
* a coverage pass must not do per record.
*/
export function readNameAt(bytes: Uint8Array, cursor: RecordCursor): string {
const from = cursor.at + RECORD_PREFIX_BYTES
// `l_read_name` counts the NUL terminator; the name does not include it.
return new TextDecoder().decode(bytes.subarray(from, from + cursor.nameBytes - 1))
}
// ---------------------------------------------------------------------------
// SEQ, and the tag block
// ---------------------------------------------------------------------------
/**
* The 4-bit sequence alphabet, in the order the format numbers it.
*
* `=` at zero is "same as the reference", which no aligner in common use
* writes; the rest are IUPAC. Two bases per byte, high nibble first.
*/
export const SEQ_ALPHABET = '=ACMGRSVTWYHKDBN'
/** Unpack `l_seq` bases out of the packed nibbles at `seqAt`. */
export function readSequence(
bytes: Uint8Array,
seqAt: number,
seqLength: number,
): string {
let out = ''
for (let i = 0; i < seqLength; i++) {
const byte = bytes[seqAt + (i >> 1)]
out += SEQ_ALPHABET[i % 2 === 0 ? byte >> 4 : byte & 0xf]
}
return out
}
/** A tag's value, in the four shapes a caller can do anything with. */
export type TagValue = number | string | number[]
/**
* Read the tag block of one record into a plain object.
*
* Every type in the specification is handled, including `B` arrays, because a
* tag this reader skips is a tag that reads as absent — and `MD`, `NM`, `SA`
* and `RG` all being absent is indistinguishable from a file that does not
* carry them, which is the one confusion this feature cannot afford.
*
* An unknown type character stops the walk rather than throwing: the rest of
* the record has already been read, and losing the tags of one odd record is a
* better outcome than losing the window it is in.
*/
export function readTags(
bytes: Uint8Array,
from: number,
to: number,
): Record<string, TagValue> {
const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength)
const out: Record<string, TagValue> = {}
let at = from
while (at + 3 <= to) {
const tag = String.fromCharCode(bytes[at], bytes[at + 1])
const type = String.fromCharCode(bytes[at + 2])
at += 3
if (type === 'B') {
const subtype = String.fromCharCode(bytes[at])
const count = view.getInt32(at + 1, true)
at += 5
const width = NUMERIC_WIDTH[subtype]
// An array longer than the tag block is the block's end misread, and
// reading it would build a list of whatever follows: the next record's
// bytes, a few megabytes of them.
if (width === undefined || count < 0 || at + count * width > to) return out
const values: number[] = []
for (let i = 0; i < count; i++) values.push(readNumeric(view, at + i * width, subtype))
at += count * width
out[tag] = values
continue
}
if (type === 'Z' || type === 'H') {
let end = at
while (end < to && bytes[end] !== 0) end++
out[tag] = new TextDecoder().decode(bytes.subarray(at, end))
at = end + 1
continue
}
if (type === 'A') {
out[tag] = String.fromCharCode(bytes[at])
at += 1
continue
}
const width = NUMERIC_WIDTH[type]
if (width === undefined) return out
out[tag] = readNumeric(view, at, type)
at += width
}
return out
}
const NUMERIC_WIDTH: Record<string, number | undefined> = {
c: 1, C: 1, s: 2, S: 2, i: 4, I: 4, f: 4,
}
function readNumeric(view: DataView, at: number, type: string): number {
switch (type) {
case 'c': return view.getInt8(at)
case 'C': return view.getUint8(at)
case 's': return view.getInt16(at, true)
case 'S': return view.getUint16(at, true)
case 'i': return view.getInt32(at, true)
case 'I': return view.getUint32(at, true)
case 'f': return view.getFloat32(at, true)
default: return 0
}
}
/** One alignment, decoded whole. Built for a dialog, never for a pass. */
export interface BamRecord {
refId: number
pos: number
end: number
mapq: number
flag: number
name: string
/** Packed ops, `(length << 4) | op`, with a long CIGAR already resolved. */
cigar: Uint32Array
seq: string
/** Phred scores, or `null` where the record stores the `0xff` "absent" run. */
qual: Uint8Array | null
nextRefId: number
nextPos: number
tlen: number
tags: Record<string, TagValue>
}
/**
* Decode the record a cursor is sitting on, whole.
*
* The `CG` tag is resolved here rather than left to the caller. A record with
* more than 65,535 CIGAR operations — a long-read alignment across a spliced
* transcript, routinely — cannot fit its op count in the `n_cigar_op` field, so
* the format stores a two-op placeholder (`<l_seq>S<ref_len>N`) and the real
* ops in a `B,I` tag. The reference *span* is right either way, which is why
* coverage never has to know; the picture of the read is not, which is why
* anything drawing one does.
*/
export function decodeRecord(bytes: Uint8Array, cursor: RecordCursor): BamRecord {
const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength)
const qualAt = cursor.seqAt + ((cursor.seqLength + 1) >> 1)
const tagsFrom = qualAt + cursor.seqLength
const tagsTo = cursor.at + cursor.bytes - 4
const tags = readTags(bytes, tagsFrom, tagsTo)
let cigar = new Uint32Array(cursor.cigarOps)
for (let i = 0; i < cursor.cigarOps; i++) cigar[i] = view.getUint32(cursor.cigarAt + i * 4, true)
const long = tags['CG']
if (Array.isArray(long) && isLongCigarPlaceholder(cigar, cursor.seqLength)) {
cigar = Uint32Array.from(long)
delete tags['CG']
}
const qual = bytes.subarray(qualAt, qualAt + cursor.seqLength)
return {
refId: cursor.refId,
pos: cursor.pos,
end: cursor.end,
mapq: cursor.mapq,
flag: cursor.flag,
name: readNameAt(bytes, cursor),
cigar,
seq: readSequence(bytes, cursor.seqAt, cursor.seqLength),
qual: qual.length > 0 && qual[0] === 0xff ? null : qual.slice(),
nextRefId: cursor.nextRefId,
nextPos: cursor.nextPos,
tlen: cursor.tlen,
tags,
}
}
/** `<l_seq>S<ref_len>N` — the two-op stand-in a `CG` tag replaces. */
function isLongCigarPlaceholder(cigar: Uint32Array, seqLength: number): boolean {
if (cigar.length !== 2) return false
const softClip = (cigar[0] & 0xf) === 4 && cigar[0] >>> 4 === seqLength
return softClip && (cigar[1] & 0xf) === 3
}
/**
* The real ops of a record whose CIGAR is too long to fit in the record.
*
* `null` for the overwhelming majority, which have an ordinary CIGAR and pay
* two comparisons for the question. Where it is not null, the ops in the record
* are the `<l_seq>S<ref_len>N` stand-in — and that matters to anything walking
* the CIGAR rather than merely measuring it, because the `N` is a *skip* and a
* pass that honours skips will emit no aligned block at all. The reference
* *span* is right either way, which is why the record's end never needs this.
*/
export function longCigarOf(bytes: Uint8Array, cursor: RecordCursor): Uint32Array | null {
if (cursor.cigarOps !== 2) return null
const view = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength)
const first = view.getUint32(cursor.cigarAt, true)
const second = view.getUint32(cursor.cigarAt + 4, true)
if ((first & 0xf) !== 4 || first >>> 4 !== cursor.seqLength) return null
if ((second & 0xf) !== 3) return null
const qualAt = cursor.seqAt + ((cursor.seqLength + 1) >> 1)
const tags = readTags(bytes, qualAt + cursor.seqLength, cursor.at + cursor.bytes - 4)
const long = tags['CG']
return Array.isArray(long) ? Uint32Array.from(long) : null
}
|