import macros, types, stew/ranges/[memranges, ptr_arith], object_serialization, priv/defs export memranges type RlpWriter* = object pendingLists: seq[tuple[remainingItems, outBytes: int]] output: Bytes PrematureFinalizationError* = object of Exception IntLike* = concept x, y type T = type(x) # arithmetic ops x + y is T x * y is T x - y is T x div y is T x mod y is T # some int compatibility required for big endian encoding: x shr int is T x shl int is T x and 0xff is int x < 128 is bool Integer* = SomeInteger # or IntLike const wrapObjsInList* = true proc bytesNeeded(num: Integer): int = type IntType = type(num) var n = num while n != IntType(0): inc result n = n shr 8 proc writeBigEndian(outStream: var Bytes, number: Integer, lastByteIdx: int, numberOfBytes: int) = mixin `and`, `shr` var n = number for i in countdown(lastByteIdx, lastByteIdx - int(numberOfBytes) + 1): outStream[i] = byte(n and 0xff) n = n shr 8 proc writeBigEndian(outStream: var Bytes, number: Integer, numberOfBytes: int) {.inline.} = outStream.setLen(outStream.len + numberOfBytes) outStream.writeBigEndian(number, outStream.len - 1, numberOfBytes) proc writeCount(bytes: var Bytes, count: int, baseMarker: byte) = if count < THRESHOLD_LIST_LEN: bytes.add(baseMarker + byte(count)) else: let origLen = bytes.len lenPrefixBytes = count.bytesNeeded bytes.setLen(origLen + int(lenPrefixBytes) + 1) bytes[origLen] = baseMarker + (THRESHOLD_LIST_LEN - 1) + byte(lenPrefixBytes) bytes.writeBigEndian(count, bytes.len - 1, lenPrefixBytes) proc add(outStream: var Bytes, newChunk: BytesRange) = let prevLen = outStream.len outStream.setLen(prevLen + newChunk.len) # XXX: Use copyMem here for i in 0 ..< newChunk.len: outStream[prevLen + i] = newChunk[i] {.this: self.} {.experimental.} using self: var RlpWriter proc initRlpWriter*: RlpWriter = newSeq(result.pendingLists, 0) newSeq(result.output, 0) proc decRet(n: var int, delta: int): int = n -= delta return n proc maybeClosePendingLists(self) = while pendingLists.len > 0: let lastListIdx = pendingLists.len - 1 doAssert pendingLists[lastListIdx].remainingItems >= 1 if decRet(pendingLists[lastListIdx].remainingItems, 1) == 0: # A list have been just finished. It was started in `startList`. let listStartPos = pendingLists[lastListIdx].outBytes pendingLists.setLen lastListIdx # How many bytes were written since the start? let listLen = output.len - listStartPos # Compute the number of bytes required to write down the list length let totalPrefixBytes = if listLen < int(THRESHOLD_LIST_LEN): 1 else: int(listLen.bytesNeeded) + 1 # Shift the written data to make room for the prefix length output.setLen(output.len + totalPrefixBytes) let outputBaseAddr = output.baseAddr moveMem(outputBaseAddr.shift(listStartPos + totalPrefixBytes), outputBaseAddr.shift(listStartPos), listLen) # Write out the prefix length if listLen < THRESHOLD_LIST_LEN: output[listStartPos] = LIST_START_MARKER + byte(listLen) else: let listLenBytes = totalPrefixBytes - 1 output[listStartPos] = LEN_PREFIXED_LIST_MARKER + byte(listLenBytes) output.writeBigEndian(listLen, listStartPos + listLenBytes, listLenBytes) else: # The currently open list is not finished yet. Nothing to do. return proc appendRawList(self; bytes: BytesRange) = output.writeCount(bytes.len, LIST_START_MARKER) output.add(bytes) maybeClosePendingLists() proc appendRawBytes*(self; bytes: BytesRange) = output.add(bytes) maybeClosePendingLists() proc startList*(self; listSize: int) = if listSize == 0: appendRawList(BytesRange()) else: pendingLists.add((listSize, output.len)) template appendBlob(self; data, startMarker) = mixin baseAddr if data.len == 1 and byte(data[0]) < BLOB_START_MARKER: self.output.add byte(data[0]) else: self.output.writeCount(data.len, startMarker) let startPos = output.len self.output.setLen(startPos + data.len) copyMem(shift(baseAddr(self.output), startPos), baseAddr(data), data.len) maybeClosePendingLists() proc appendImpl(self; data: string) = appendBlob(self, data, BLOB_START_MARKER) proc appendBlob(self; data: openarray[byte]) = appendBlob(self, data, BLOB_START_MARKER) proc appendBlob(self; data: openarray[char]) = appendBlob(self, data, BLOB_START_MARKER) proc appendBytesRange(self; data: BytesRange) = appendBlob(self, data, BLOB_START_MARKER) proc appendImpl(self; data: MemRange) = appendBlob(self, data, BLOB_START_MARKER) proc appendInt(self; i: Integer) = # this is created as a separate proc as an extra precaution against # any overloading resolution problems when matching the IntLike concept. type IntType = type(i) if i == IntType(0): self.output.add BLOB_START_MARKER elif i < BLOB_START_MARKER.Integer: self.output.add byte(i) else: let bytesNeeded = i.bytesNeeded self.output.writeCount(bytesNeeded, BLOB_START_MARKER) self.output.writeBigEndian(i, bytesNeeded) self.maybeClosePendingLists() proc appendFloat(self; data: float64) = # This is not covered in the RLP spec, but Geth uses Go's # `math.Float64bits`, which is defined here: # https://github.com/gopherjs/gopherjs/blob/master/compiler/natives/src/math/math.go let uintWords = cast[ptr UncheckedArray[uint32]](unsafeAddr data) let uint64bits = (uint64(uintWords[1]) shl 32) or uint64(uintWords[0]) self.appendInt(uint64bits) template appendImpl(self; i: Integer) = appendInt(self, i) template appendImpl(self; e: enum) = appendImpl(self, int(e)) template appendImpl(self; b: bool) = appendImpl(self, int(b)) proc appendImpl[T](self; listOrBlob: openarray[T]) = mixin append # TODO: This append proc should be overloaded by `openarray[byte]` after # nim bug #7416 is fixed. when T is (byte or char): self.appendBlob(listOrBlob) else: self.startList listOrBlob.len for i in 0 ..< listOrBlob.len: self.append listOrBlob[i] proc appendRecordType*(self; obj: object|tuple, wrapInList = wrapObjsInList) = mixin enumerateRlpFields, append if wrapInList: self.startList(static obj.type.rlpFieldsCount) template op(field) = when hasCustomPragma(field, rlpCustomSerialization): append(self, obj, field) else: append(self, field) enumerateRlpFields(obj, op) proc appendImpl(self; data: object) {.inline.} = # TODO: This append proc should be overloaded by `BytesRange` after # nim bug #7416 is fixed. when data is BytesRange: self.appendBytesRange(data) else: self.appendRecordType(data) proc appendImpl(self; data: tuple) {.inline.} = self.appendRecordType(data) # We define a single `append` template with a pretty low specifity # score in order to facilitate easier overloading with user types: template append*[T](w: var RlpWriter; data: T) = when data is float64: # XXX: This works around an overloading bug. # Apparently, integer literals will be converted to `float64` # values with higher precedence than the generic match to Integer appendFloat(w, data) else: appendImpl(w, data) proc initRlpList*(listSize: int): RlpWriter = result = initRlpWriter() startList(result, listSize) # TODO: This should return a lent value proc finish*(self): Bytes = if pendingLists.len > 0: raise newException(PrematureFinalizationError, "Insufficient number of elements written to a started list") result = output proc encode*[T](v: T): Bytes = mixin append var writer = initRlpWriter() writer.append(v) return writer.finish proc encodeInt*(i: Integer): Bytes = var writer = initRlpWriter() writer.appendInt(i) return writer.finish macro encodeList*(args: varargs[untyped]): Bytes = var listLen = args.len writer = genSym(nskVar, "rlpWriter") body = newStmtList() append = bindSym("append", brForceOpen) for arg in args: body.add quote do: `append`(`writer`, `arg`) result = quote do: var `writer` = initRlpList(`listLen`) `body` finish(`writer`) when false: # XXX: Currently fails with a malformed AST error on the args.len expression template encodeList*(args: varargs[untyped]): BytesRange = mixin append var writer = initRlpList(args.len) for arg in args: writer.append(arg) writer.finish