implement merge-logic - #263

This commit is contained in:
Kevin Jahns 2020-12-10 18:06:35 +01:00
parent 783c4d8209
commit 320da29b69

View File

@ -5,29 +5,13 @@ import * as encoding from 'lib0/encoding.js'
import {
createID,
readItemContent,
Item, GC, AbstractUpdateDecoder, AbstractUpdateEncoder, UpdateDecoderV1, UpdateDecoderV2, UpdateEncoderV1, UpdateEncoderV2 // eslint-disable-line
Item, GC, AbstractUpdateDecoder, UpdateDecoderV1, UpdateDecoderV2, UpdateEncoderV1, UpdateEncoderV2 // eslint-disable-line
} from '../internals.js'
/**
* @param {Array<Uint8Array>} updates
* @return {Uint8Array}
*/
export const mergeUpdates = updates => {
return updates[0]
}
/**
* @param {Uint8Array} update
* @param {Uint8Array} sv
*/
export const diffUpdate = (update, sv) => {
return update
}
/**
* @param {AbstractUpdateDecoder} decoder
*/
export function * lazyStructReaderGenerator (decoder) {
function * lazyStructReaderGenerator (decoder) {
const numOfStateUpdates = decoding.readVarUint(decoder.restDecoder)
for (let i = 0; i < numOfStateUpdates; i++) {
const numberOfStructs = decoding.readVarUint(decoder.restDecoder)
@ -96,6 +80,13 @@ export class LazyStructWriter {
*/
constructor (encoder) {
this.fresh = true
/**
* We keep the last written struct around in case we want to
* merge it with the next written struct.
* When a new struct is received: if mergeable merge; otherwise write curr and keep new struct around.
* @type {null | Item | GC}
*/
this.curr = null
this.currClient = 0
this.startClock = 0
this.written = 0
@ -112,57 +103,192 @@ export class LazyStructWriter {
*/
this.clientStructs = []
}
}
flushCurr () {
if (!this.fresh) {
this.clientStructs.push({ written: this.written, restEncoder: encoding.toUint8Array(this.encoder.restEncoder) })
this.encoder.restEncoder = encoding.createEncoder()
this.fresh = true
}
}
/**
* @param {Array<Uint8Array>} updates
* @return {Uint8Array}
*/
export const mergeUpdates = updates => mergeUpdatesV2(updates, UpdateDecoderV1, UpdateEncoderV2)
/**
* @param {Item | GC} struct
* @param {number} offset
*/
write (struct, offset) {
// flush curr if we start another client
if (!this.fresh && this.currClient !== struct.id.client) {
this.flushCurr()
this.currClient = struct.id.client
// write next client
this.encoder.writeClient(struct.id.client)
// write startClock
encoding.writeVarUint(this.encoder.restEncoder, struct.id.clock)
}
struct.write(this.encoder, offset)
this.written++
}
toUint8Array () {
this.flushCurr()
// this is a fresh encoder because we called flushCurr
const restEncoder = this.encoder.restEncoder
/**
* Now we put all the fragments together.
* This works similarly to `writeClientsStructs`
*/
// write # states that were updated - i.e. the clients
encoding.writeVarUint(restEncoder, this.clientStructs.length)
for (let i = 0; i < this.clientStructs.length; i++) {
const partStructs = this.clientStructs[i]
/**
* Works similarly to `writeStructs`
*/
// write # encoded structs
encoding.writeVarUint(restEncoder, partStructs.written)
// write the rest of the fragment
encoding.writeUint8Array(restEncoder, partStructs.restEncoder)
}
return this.encoder.toUint8Array()
/**
* This method is intended to slice any kind of struct and retrieve the right part.
* It does not handle side-effects, so it should only be used by the lazy-encoder.
*
* @param {Item | GC} left
* @param {number} diff
* @return {Item | GC}
*/
const sliceStruct = (left, diff) => {
if (left.constructor === GC) {
const { client, clock } = left.id
return new GC(createID(client, clock + diff), left.length - diff)
} else {
const leftItem = /** @type {Item} */ (left)
const { client, clock } = leftItem.id
return new Item(
createID(client, clock + diff),
null,
createID(client, clock + diff - 1),
null,
leftItem.rightOrigin,
leftItem.parent,
leftItem.parentSub,
leftItem.content.splice(diff)
)
}
}
/**
*
* This function works similarly to `readUpdateV2`.
*
* @param {Array<Uint8Array>} updates
* @param {typeof UpdateDecoderV1 | typeof UpdateDecoderV2} [YDecoder]
* @param {typeof UpdateEncoderV1 | typeof UpdateEncoderV2} [YEncoder]
* @return {Uint8Array}
*/
export const mergeUpdatesV2 = (updates, YDecoder = UpdateDecoderV2, YEncoder = UpdateEncoderV2) => {
const updateDecoders = updates.map(update => new UpdateDecoderV1(decoding.createDecoder(update)))
let lazyStructDecoders = updateDecoders.map(decoder => new LazyStructReader(decoder))
/**
* @todo we don't need offset because we always slice before
* @type {null | { struct: Item | GC, offset: number }}
*/
let currWrite = null
const updateEncoder = new YEncoder()
// write structs lazily
const lazyStructEncoder = new LazyStructWriter(updateEncoder)
// Note: We need to ensure that all lazyStructDecoders are fully consumed
// Note: Should merge document updates whenever possible - even from different updates
// Note: Should handle that some operations cannot be applied yet ()
while (true) {
// Write higher clients first ⇒ sort by clientID & clock and remove decoders without content
lazyStructDecoders = lazyStructDecoders.filter(dec => dec.curr !== null)
lazyStructDecoders.sort(
/** @type {function(any,any):number} */ (dec1, dec2) =>
dec1.curr.id.client === dec2.curr.id.client
? dec1.curr.id.clock - dec2.curr.id.clock
: dec1.curr.id.client - dec2.curr.id.client
)
if (lazyStructDecoders.length === 0) {
break
}
const currDecoder = lazyStructDecoders[0]
// write from currDecoder until the next operation is from another client or if filler-struct
// then we need to reorder the decoders and find the next operation to write
const firstClient = /** @type {Item | GC} */ (currDecoder.curr).id.client
if (currWrite !== null) {
let curr = /** @type {Item | GC} */ (currDecoder.curr)
if (firstClient !== currWrite.struct.id.client) {
writeStructToLazyStructWriter(lazyStructEncoder, currWrite.struct, currWrite.offset)
currWrite = { struct: curr, offset: 0 }
currDecoder.next()
} else if (currWrite.struct.id.clock + currWrite.struct.length < curr.id.clock) {
// @todo write currStruct & set currStruct = Skip(clock = currStruct.id.clock + currStruct.length, length = curr.id.clock - self.clock)
throw new Error('unhandled case') // @Todo !
} else if (currWrite.struct.id.clock + currWrite.struct.length >= curr.id.clock) {
const diff = currWrite.struct.id.clock + currWrite.struct.length - curr.id.clock
if (diff > 0) {
curr = sliceStruct(curr, diff)
}
if (!currWrite.struct.mergeWith(/** @type {any} */ (curr))) {
writeStructToLazyStructWriter(lazyStructEncoder, currWrite.struct, currWrite.offset)
currWrite = { struct: curr, offset: 0 }
currDecoder.next()
}
}
} else {
currWrite = { struct: /** @type {Item | GC} */ (currDecoder.curr), offset: 0 }
currDecoder.next()
}
for (
let next = currDecoder.curr;
next !== null && next.id.client === firstClient && next.id.clock === currWrite.struct.id.clock + currWrite.struct.length; // @Todo && next.constructor !== skippable
next = currDecoder.next()
) {
writeStructToLazyStructWriter(lazyStructEncoder, currWrite.struct, currWrite.offset)
currWrite = { struct: next, offset: 0 }
}
}
finishLazyStructWriting(lazyStructEncoder)
// Read DeleteSets and merge them.
// Write merged deleteset.
// -- updateEncoder.writeDs()
return updateEncoder.toUint8Array()
}
/**
* @param {Uint8Array} update
* @param {Uint8Array} sv
*/
export const diffUpdate = (update, sv) => {
return update
}
/**
* @param {LazyStructWriter} lazyWriter
*/
const flushLazyStructWriter = lazyWriter => {
if (!lazyWriter.fresh) {
lazyWriter.clientStructs.push({ written: lazyWriter.written, restEncoder: encoding.toUint8Array(lazyWriter.encoder.restEncoder) })
lazyWriter.encoder.restEncoder = encoding.createEncoder()
lazyWriter.fresh = true
}
}
/**
* @param {LazyStructWriter} lazyWriter
* @param {Item | GC} struct
* @param {number} offset
*/
const writeStructToLazyStructWriter = (lazyWriter, struct, offset) => {
// flush curr if we start another client
if (!lazyWriter.fresh && lazyWriter.currClient !== struct.id.client) {
flushLazyStructWriter(lazyWriter)
lazyWriter.currClient = struct.id.client
// write next client
lazyWriter.encoder.writeClient(struct.id.client)
// write startClock
encoding.writeVarUint(lazyWriter.encoder.restEncoder, struct.id.clock)
}
struct.write(lazyWriter.encoder, offset)
lazyWriter.written++
}
/**
* Call this function when we collected all parts and want to
* put all the parts together. After calling this method,
* you can continue using the UpdateEncoder.
*
* @param {LazyStructWriter} lazyWriter
*/
const finishLazyStructWriting = (lazyWriter) => {
flushLazyStructWriter(lazyWriter)
// this is a fresh encoder because we called flushCurr
const restEncoder = lazyWriter.encoder.restEncoder
/**
* Now we put all the fragments together.
* This works similarly to `writeClientsStructs`
*/
// write # states that were updated - i.e. the clients
encoding.writeVarUint(restEncoder, lazyWriter.clientStructs.length)
for (let i = 0; i < lazyWriter.clientStructs.length; i++) {
const partStructs = lazyWriter.clientStructs[i]
/**
* Works similarly to `writeStructs`
*/
// write # encoded structs
encoding.writeVarUint(restEncoder, partStructs.written)
// write the rest of the fragment
encoding.writeUint8Array(restEncoder, partStructs.restEncoder)
}
}