Utilities for packing/unpacking fields of a bitstream
npm install @astronautlabs/bitstreamnpm install @astronautlabs/bitstream
typescript
import { BitstreamReader } from '@astronautlabs/bitstream';
let reader = new BitstreamReader();
reader.addBuffer(Buffer.from([0b11110000, 0b10001111]));
await reader.read(2); // == 0b11
await reader.read(3); // == 0b110
await reader.read(4); // == 0b0001
await reader.read(7); // == 0b0001111
`
The above will read the values as unsigned integers in big-endian (network byte order) format.
Asynchronous versus Synchronous
All read operations come in two flavors, asynchronous and synchronous. For instance to read an unsigned integer
asynchronously, use read(). For this and other asynchronous read operations, resolution of the resulting promise
is delayed until enough data is available to complete the operation. Note that there can be only one asynchronous
read operation in progress at a time for a given BitstreamReader object.
The synchronous method for reading unsigned integers is readSync(). When using synchronous methods, there must be
enough bytes available to the reader (via addBuffer()) to read the desired number of bits. If this is not the case,
an exception is thrown. You can check how many bits are available using the isAvailable() method:
`typescript
if (reader.isAvailable(10)) {
// 10 bits are available
let value = reader.readSync(10);
}
`
Alternatively, you can use .assure() to wait until the desired number of bits are available. Again, there can only be
one pending call to read() or assure() at a time. This allows you to "batch" synchronous reads in a single
"await" operation.
`typescript
await reader.assure(13);
let value1 = reader.readSync(3);
let value2 = reader.readSync(10);
`
Reading signed integers
Use the readSigned / readSignedSync methods to read a signed two's complement integer.
Reading floating-point integers [IEEE 754]
Use the readFloat / readFloatSync methods to read an IEEE 754 floating point value. The bit length passed must be
either 32 (32-bit single-precision) or 64 (64-bit double-precision).
Reading Strings
Use the readString / readStringSync methods to read string values.
`typescript
await reader.readString(10); // read a fixed length string with 10 characters.
`
By default readString() will cut off the string at the first character with value 0 (ie, the string is
considered null-terminated) and stop reading. You can disable this behavior so that the returned string always
contains all bytes that were in the bitstream:
`typescript
await reader.readString(10, { nullTerminated: false });
`
The default text encoding is UTF-8 (utf-8). You can read a string using any text encoding supported by the platform
you are on. For Node.js these are the encodings supported by Buffer. On the web, only utf-8 is available
(see documentation for TextEncoder/TextDecoder).
`typescript
await reader.readString(10, { encoding: 'utf16le' })
`
> Important: In cases like above where you are using encodings where a character spans
> multiple bytes (including UTF-8), the length given to readString() is always the _number of
> bytes_ not the _number of characters_. It is easy to make mistaken assumptions in this regard.
BitstreamWriter: Writing to bitstreams imperatively
`typescript
import { BitstreamWriter } from '@astronautlabs/bitstream';
let writer = new BitstreamWriter(writableStream, bufferLength);
writer.write(2, 0b10);
writer.write(10, 0b1010101010);
writer.write(length, value);
`
writableStream can be any object which has a write(chunk : Uint8Array) method (see exported Writable interface).
Examples of writables you can use include:
- Node.js' writable streams (Writable from the stream package)
- WritableStreamDefaultWriter from the WHATWG Streams specification in the browser
- Any custom object which implements the Writable interface
The bufferLength parameter determines how many bytes will be buffered before the buffer will be flushed out to the
passed writable stream. This parameter is optional, the default (and minimum value) is 1 (one byte per buffer).
Writing unsigned integers
`typescript
writer.write(2, 0b01);
writer.write(2, 0b1101);
writer.write(2, 0b1111101); // 0b01 will be written
`
> Note: Any bits in value above the length'th bit will be ignored, so all of the above are equivalent.
Writing signed integers
Use the writeSigned() method to write signed two's complement integers.
Writing floating point values
Use the writeFloat() method to write IEEE 754 floating point values. Only lengths of 32 (for 32-bit single-precision)
and 64 (for 64-bit double-precision) are supported.
Writing strings
Use the writeString() method to write string values. The default encoding is UTF-8 (utf-8).
Any other encoding supported by the platform can be used (ie those supported by Buffer on Node.js). On the web, only utf-8 is supported (see TextEncoder / TextDecoder).
Writing byte arrays
Use the writeBuffer() method to write byte arrays. On Node.js you can also pass Buffer.
BitstreamElement: declarative structural serialization
Efficient structural (de)serialization can be achieved by building subclasses of the BitstreamElement class.
Deserialization (reading)
You can declaratively specify elements of bitstreams, then read and write them to bitstream readers/writers as needed. To do this, extend the BitstreamElement class:
`typescript
import { BitstreamElement, Field } from '@astronautlabs/bitstream';
class MyElement extends BitstreamElement {
@Field(2) field1 : number;
@Field(4) field2 : number;
@Field(3) field3 : number;
@Field(1) field4 : number;
@Field(1) field5 : boolean;
}
`
Then, read from a BitstreamReader using:
`typescript
let element = await MyElement.read(bitstreamReader);
`
Or, deserialize from a Buffer using:
`typescript
let element = await MyElement.deserialize(buffer);
`
$3
- null and undefined are written as 0 during serialization
- Numbers are treated as big-endian unsigned integers by default. Decimal portions of numbers are truncated.
- Use the { number: { format: 'signed' }} option to use signed two's complement integer. Decimals are truncated
- Use the { number: { format: 'float' }} option to use IEEE 754 floating point. Decimals are retained
- An error is thrown when trying to serialize NaN or infinite values (except when using the float format)
$3
If you specify type boolean for a field, the integer value 0 will be deserialized to false and all other values
will be deserialized as true. When booleans are serialized, 0 is used for false and 1 is used for true.
You can customize this behavior using the boolean field options (ie @Field(8, { boolean: { ... } })):
- true: The numeric value to use for true (default 1)
- false: The numeric value to use for false (default 0)
- undefined: The numeric value to use when writing undefined (default 0)
- mode: How to handle novel inputs when reading:
- "true-unless": The value is true unless the numeric value chosen for 'false' is observed (default mode). For
example 0 is false, 1 is true, 100 is true
- "false-unless": The value is false unless the numeric value chosen for 'true' is observed. For example
0 is false, 1 is true, 100 is false
- "undefined": The value is true if the numeric value for 'true' is observed, false if the numeric value for
'false' is observed and undefined otherwise. For example 0 is false, 1 is true, 100 is undefined.
If none of these options fit your use case, you can write a custom Serializer.
$3
Elements can also handle serializing fixed-length strings. For example, to represent a fixed-length 5-byte string,
set the length to 5 (note that this differs from other types where the length is specified in bits).
> Note: Null-termination is the default here, make sure to set nullTerminated to false if you do not desire
> that behavior.
`typescript
@Field(5) stringField : string;
`
If you wish to control the encoding options, use the string options group:
`typescript
@Field(5, { string: { encoding: 'ascii', nullTerminated: false } })
stringField : string;
`
For information about the available encodings, see "Reading Strings" above.
$3
You can represent a number of bytes as a byte array:
`typescript
@Field(10*8) read10BytesIntoBuffer : Uint8Array;
`
When running on Node.js you can also specify Buffer as the type instead:
`typescript
@Field(10*8) read10BytesIntoBuffer : Buffer;
`
$3
You can also nest element classes:
`typescript
class PartElement extends BitstreamElement {
@Field(3) a : number;
@Field(5) b : number;
}
class WholeElement extends BitstreamElement {
@Field() part1 : PartElement;
@Field() part2 : PartElement;
}
`
$3
Many kinds of arrays are supported.
#### Array with number values (unsigned/signed/float)
Arrays with number elements are natively supported for convenience.
For example, to read 10 8-bit unsigned integers:
`typescript
class BufferElement extends BitstreamElement {
@Field(10, { array: { type: Number, elementLength: 8 }})
array : number[];
}
`
> Important:
> - You must specify the array.type option when using elements.
> Typescript cannot convey the element type via reflection, only that the field is of type Array
> which is not enough information to deserialize the array.
> - When using arrays of numbers you must specify the array.elementLength option for the library to know how many
> bits each array item represents within the bitstream. For other array item types such as BitstreamElement
> the length is already known and array.elementLength is ignored
You can read signed integers and floating point values instead of unsigned integers by specifying the number.format
option (see Number Fields for more information).
When handling arrays, the length parameter of @Field() is used to represent the item count for the array by default. Alternatively (and mainly for historical reasons) you can also set the options.array.count property.
#### Derived lengths
As with all fields, you can represent the number of items dynamically with a determinant fucntion, allowing the array
length to be dependent on any value already read before the field being read. For more information about this, see the
section on "Dynamic Lengths" below.
You can pair this with the writtenValue feature to ensure that the correct length is always written to the bitstream.
#### Array with preceding count
You can specify that a preceding count field should be read to determine what the length of the array is in the
countFieldLength option. If this option is not provided or undefined, no preceding count field is read.
#### Array with element values
You can also use arrays of elements:
`typescript
class ItemElement extends BitstreamElement {
@Field(3) a : number;
@Field(5) b : number;
}
class ListElement extends BitstreamElement {
@Field({ array: { type: PartElement, countFieldLength: 8 }})
parts : PartElement[];
}
`
The above will expect an 8-bit "count" field (indicates how many items are in the array), followed by that number of
PartElement syntax objects. Since ItemElement is 8 bits long, when the count is 3, there will be 3 additional
bytes (24 bits) following the count in the bitstream before the next element begins.
#### Dynamic arrays
Some bitstream formats do not send information about the number of items prior to the items themselves. In such cases
there is often a final "sentinel" value instead. This can be handled by using the hasMore discriminant:
For instance, consider the following set up where the array ends when a suffix field is false:
`typescript
class ItemElement extends BitstreamElement {
@Field(8) value : number;
@Field(8) hasMore : boolean;
}
class ListElement extends BitstreamElement {
@Field(0, { array: { type: ItemElement, hasMore: a => a[a.length - 1].hasMore }})
array : ItemElement[];
}
`
$3
You can make any field optional by using the presentWhen and/or excludedWhen options:
`typescript
class ItemElement extends BitstreamElement {
@Field(3) a : number;
@Field(5, { presentWhen: i => i.a === 10 }) b : number;
}
`
In the above case, the second field is only present when the first field is 10.
excludedWhen is the opposite of presentWhen and is provided for convenience and expressiveness.
$3
When using @Field() to present a child BitstreamElement relationship,
it can be useful to run some code to initialize the instance that is
created.
`ts
class Parent {
@Field(8)
context: number;
@Field({ initializer: (instance, parent) => instance.context = parent.context })
child: Child;
}
class Child {
context: number;
}
`
You can also use this from the static read() method:
`ts
Child.read(reader, { initializer: instance => instance.context = somePreviouslyParsedContextNumber });
`
This is a useful pattern for passing knowledge needed for parsing a child element down from unknown parent elements.
$3
Sometimes the length of a field depends on what has been read before the field being read. You can specify lengths as _determinant_ functions which determine lengths dynamically:
`typescript
class ItemElement extends BitstreamElement {
@Field(3) length : number;
@Field(i => i.length) value : number;
}
`
In the above, if length is read as 30, then the value field is read/written as 30 bits long.
Determinants can make use of any property of the instance which appear in the bitstream before the current element.
They are also passed the bitstream element which contains this one (if one exists).
Determinants are available on many other properties as well. Where booleans are expected _discriminants_ work the same
way. Examples of features that accept discriminants include hasMore, presentWhen and discriminants are the core
concept which enabling element variation (@Variant()).
$3
Many bitstream formats make use of the concept of specialization. BitstreamElement supports this using its variation
system. Any BitstreamElement class may have zero or more classes (which are marked with @Variant()) which are
automatically substituted for the class which is being read provided the discriminant passed to @Variant() holds true.
`typescript
class BaseElement extends BitstreamElement {
@Field(3) type : number;
}
@Variant(i => i.type === 0x1)
class Type1Element extends BaseElement {
@Field(5) field1 : number;
}
@Variant(i => i.type === 0x2)
class Type2Element extends BaseElement {
@Field(5) field1 : number;
}
`
When reading an instance of BaseElement the library automatically checks the discriminants specified on the defined
Variant classes to determine the appropriate subclass to substitute. In the case above, if type is read as 0x2 when
performing await BaseElement.read(reader), then the result will be an instance of Type2Element.
> Note: Variation which occurs at _the end_ of a bitstream element (as above) is referred to as "tail variation".
> Variation which occurs in _the middle_ of a bitstream element is referred to as "marked variation".
$3
Sometimes a bitstream format specifies that the specialized fields fall somewhere in the middle of the overall
structure, with the fields of the base class falling both before and after those found in the subclass. This is called
"marked variation".
BitstreamElement can accomodate this using @VariantMarker():
`typescript
class BaseElement extends BitstreamElement {
@Field(3) type : number;
@VariantMarker() $variant;
@Field(8) checksum : number;
}
@Variant(i => i.type === 0x1)
class Type1Element extends BaseElement {
@Field(5) field1 : number;
}
@Variant(i => i.type === 0x2)
class Type2Element extends BaseElement {
@Field(5) field1 : number;
}
`
In the above example, variation will occur after reading type but before reading checksum. After variation occurs
(resulting in a Type2Element instance), the checksum field will then be read.
$3
Sometimes it is desirable to override the value present in a field with a specific formula. This is useful when
representing the lengths of arrays, Buffers, or ensuring that a certain hardcoded value is always written regardless
of what is specified in the instance being written. Use the writtenValue option to override the value that is
specified on an instance:
`typescript
class Type2Element extends BaseElement {
@Field(version, { writtenValue: () => 123 })
version : number;
}
`
The above element will always write 123 in the field specified by version.
You can depend on properties of the containing object as well. Unlike determinants used during reading, these functions
have full access to the value being serialized, so they can look ahead as well as behind for determining the value to
write.
`typescript
class Type2Element extends BaseElement {
@Field(version, { writtenValue: i => i.items.length })
count : number;
@Field(i => i.count, { array: { type: Number, elementLength: 8 }})
items : number[];
}
`
This sort of arrangement is useful for handling more complex array length scenarios, as count is able to be parsed
and serialized independently from items, thus allowing items to reference count in it's determinant function.
Similarly, when writing the value to bitstream, the value found in items determines the value of count, ensuring
that the two are always in sync _on the wire_.
One downside of this approach is that callers interacting with the Type2Element instances must be aware that count
and items are not intrinsically linked (even if temporarily), and thus may not agree at all times.
A cleaner approach to handle this is to encapsulate this into the object and forbid writing to count outside of
parsing:
`typescript
class Type2Element extends BaseElement {
@Field(version, { writtenValue: i => i.items.length })
private $count : number;
get count() { return this.items?.length ?? this.$count; }
@Field(i => i.$count, { array: { type: Number, elementLength: 8 }})
private $items : number[];
get items() { return this.items; }
set items(value) {
this.$items = value ?? [];
this.$count = this.$items.length;
}
}
`
Here count and items.length will always agree at all times, with the bonus that assigning undefined to the array
is forbidden.
$3
It can be useful to measure the bitlength of a portion of a BitstreamElement. Such a measurement could be used when
defining the length of a bit field or when defining the value of a bitfield. Use the measure() method to accomplish
this.
`typescript
class Type2Element extends BaseElement {
@Field(8, { writtenValue: i => i.measure('version', 'checksum') })
length : number;
@Field(version, { writtenValue: () => 123 })
version : number;
@VariantMarker() $variant;
checksum : number;
}
`
In the above, the written value of length will be the number of bits occupied by the fields starting from version through checksum inclusive, including all fields specified by the variant of the element which is being written.
$3
Measurement is extremely useful, but because measure() only measures fields _inclusively_ it is tempting to introduce
fields with zero length which can be used as _markers_ for measurement purposes. You can absolutely do this with fields
marked @Field(0), but there is a dedicated decorator for this: @Marker(), which also marks the field as "ignored",
meaning it will never actually be read or written from the relevant object instance.
`typescript
class Type2Element extends BaseElement {
@Field(version, { writtenValue: () => 123 })
version : number;
@Field(8, { writtenValue: i => i.measure('$lengthStart', '$lengthEnd') })
length : number;
@Marker() $lengthStart;
@VariantMarker() $variant;
@Marker() $lengthEnd;
checksum : number;
}
`
In the above example, the written value of length will be the bit length of the fields provided by the variant subclass, not including any other fields. The fields $lengthStart and $lengthEnd will not contribute any data to the bitstream representation.
$3
There is also measureTo() and measureFrom() which measure the entire element up to and including a specific field, and from a specific field to the end of an element, respectively. You can also use measureField() to measure the size of a specific field.
$3
Note that in prior examples we specified field references as strings when calling measure(). You can also specify field references as functions which allow for better type safety:
`typescript
class Type2Element extends BaseElement {
@Field(version, { writtenValue: () => 123 })
version : number;
@Field(8, { writtenValue: (i : Type2Element) => i.measureFrom(i => i.$lengthStart) })
length : number;
@Marker() $lengthStart;
@VariantMarker() $variant;
}
`
In the case above the type of the determinant is carried through into the measureFrom() call, and Typescript will
alert you if you reference a field that doesn't exist on Type2Element.
$3
There is also @Reserved(length) which can be used in place of @Field(length). This decorator marks a field as
taking up space in the bitstream, but ignores the value in the object instance when reading/writing. Instead the value
is always high bits. That is, a field marked with @Reserved(8) will always be written as 0xFF (0b11111111). This
can be useful for standards which specify that reserved space should be filled with high bits. @ReservedLow() is also
provided which will do the opposite- filling the covered bits with all zeroes.
Other schemes can be accomodated with custom decorators. For instance:
`typescript
function RandomReserved(length : LengthDeterminant) {
return Field(length, { ignored: true, writtenValue: () => Math.floor(Math.random() 2*length) });
}
`
The above function can be used as a decorator to specify that the values read/written should be ignored (and not read/
written to the object), and when writing the value to a bitstream it should be populated with a random value. It could
be used like so:
`typescript
class MyElement extends BitstreamElement {
@RandomReserved() reserved : number;
}
`
Writing elements to bitstreams
Write to a BitstreamWriter with:
`typescript
element.write(bitstreamWriter);
`
Serialize to a Buffer using:
`typescript
let buffer = element.serialize();
`
Custom Serializers
BitstreamElement can be extended to support arbitrary field types by implementing new serializers. Simply
implement the Serializer interface and use the serializer option on the @Field() properties you wish to use it
with.
Advanced element serialization
If you need to dynamically omit or include some fields, or otherwise implement custom serialization,
you can override the deserializeFrom() and serializeTo() methods in your subclass. This allows you
to control exactly how fields are populated. You can call deserializeGroup(bitstreamReader, '*') to
immediately deserialize all fields, or deserializeGroup(bitstreamReader, 'groupName') to deserialize
fields which are in the group named groupName.
You may also require additional information to be available to your deserialization/serialization
routines. For instance:
`typescript
export class FirstElement extends BitstreamElement {
@Field(2) foo : number;
}
export class SecondElement extends BitstreamElement {
deserializeFrom(bitstreamReader : BitstreamReader, firstElement : BitstreamElement) {
}
}
let firstElement = FirstElement.deserializeSync(bitstreamReader);
let secondElement = new SecondElement();
secondElement.deserializeFrom(bitstreamReader, firstElement);
`
Here, we are passing a previously decoded element into the deserializeFrom() of the element being deserialized. You
could pass any arbitrary data in this fashion, giving you flexibility in how you handle advanced serialization.
$3
When using BitstreamElement#deserialize() to parse an element object from a byte array / buffer, you can use
allowExhaustion to suppress the exception when the available bits are exhausted and instead return the partially read
object. This can be very useful for diagnostics or for cases where there are optional tailing fields.
Architecture
Generators
The elements system uses generators internally in a coroutine-like fashion for allowing parsing to work asynchronously
while still taking little or no performance penalty when using the parser in a synchronous manner. When the bitstream
becomes exhausted during reading, the reader yields the number of bits it needs. The caller is then responsible
for waiting until that amount of bits is available, and then resuming the generator where it left off. The caller uses
BitstreamReader.assure() to accomplish this.
For this reason, many lower level read operations return Generator instead of the final value or the Promise. In
a fully synchronous setting (such as deserialize or an already assured read), the caller will use the generator to
get a single (and final) value back which will be the value that was read.
Using generators internally provides a massive performance boost over doing the same with Promises as the original
version of this library did. See below for an analysis of the relevant performance aspects.
Performance
When reading data from BitstreamReader, you have two options: use the synchronous methods which will throw if not
enough data is available, or the asynchronous methods which will wait for the data to arrive before completing the read
operation. If you know you have enough data to complete the operation, you can read synchronously to avoid the overhead
of creating and awaiting a Promise. If your application is less performance intensive you can instead receive a Promise
for when the data becomes available (which happens by a addBuffer() call). This allows you to create a
pseudo-blocking control flow similar to what is done in lower level languages like C/C++. However using promises in
this manner can cause a huge reduction in performance while reading data. You should only use the async API when
performance requirements are relaxed.
When reading data into a declarative BitstreamElement class however, ECMAscript generators are used to control
whether the library needs to wait for more data. When reading a BitstreamElement using the Promise-based API you are
only incurring the overhead of the Promise API once for the initial call, and once each time there is not enough data
available in the underlying BitstreamReader, which will only happen if the raw data is not arriving fast enough. In
that case, though the Promise will have the typical overhead, it will not impact throughput because the IO wait time
will be larger than the time necessary to handle the Promise overhead. We believe that this effectively eliminates the
overhead of using an async/await pattern with reasonably sized BitstreamElements, so we encourage you start there and
optimize only if you run into throughput / CPU / memory bottlenecks.
With generators at the core of BitstreamElement, implementing high throughput applications such as audio and video
processing are quite viable with this library. You are more likely to run into a bottleneck with Javascript or Node.js
itself than to be bottlenecked by using declarative BitstreamElement classes, of course your mileage may vary! If you
believe BitstreamElement is a performance bottleneck for you, please file an issue!
So is it faster to "hand roll" using BitstreamReader instead of using BitstreamElements?
Absolutely not. You should use BitstreamElement whereever possible, because it is using generators as the core
mechanism for handling bitstream exhaustion events. Using generators internally instead of promises is _dramatically_
faster. To see this, compare the performance of @astronautlabs/bitstream@1 with @astronautlabs/bitstream@2. Using
generators in the core was introduced in v2.0.0, prior to that a Promise was issued for every individual read call.
How many times can each version of the library read the following BitstreamElement structure within a specified period
of time?
`typescript
class SampleItem extends BitstreamElement {
@Field(1) b1 : number;
@Field(1) b2 : number;
@Field(1) b3 : number;
@Field(1) b4 : number;
@Field(1) b5 : number;
@Field(1) b6 : number;
@Field(1) b7 : number;
@Field(1) b8 : number;
}
class SampleContainer extends BitstreamElement {
@Field(0, { array: { type: SampleItem, countFieldLength: 32 }})
items : SampleItem[];
}
`
The results are night and day:
> Iteration count while parsing a 103-byte buffer in 500ms (_Intel Core i7 6700_)
> - @astronautlabs/bitstream@1.1.0: Read the buffer 104 times
> - @astronautlabs/bitstream@2.0.2: Read the buffer 1,163,576 times
While we're proud of the performance improvement, it really just shows the overhead of Promises and how that
architecture was the wrong choice for BitstreamElements in V1.
Similarly, we tried giving a 1GB buffer to each version with the above test -- V2 was able to parse the entire buffer
in less than a millisecond, whereas V1 effectively _did not complete_ -- it takes _minutes_ to parse such a Buffer with
V1 even once, and quite frankly we gave up waiting for it to complete.
Debugging Element Serialization
It can be tricky to know where your app gets stuck reading a Bitstream Element. If you set
globalThis.BITSTREAM_TRACE = true then Bitstream will begin outputting some trace messages to help you understand
what field your app got stuck on while reading. This can be very helpful when your protocol descriptions have too many
bits.
Contributing
We encourage you to file issues and send pull requests! Please be sure to follow the
Code of Conduct.
Tests
Use npm test to run the test suite after making changes. Code coverage will also be generated. Important: Istanbul (the code coverage tool we are using) instruments the code to perform its coverage analysis. This breaks line numbers in stack traces. Use npm test:nocov` to skip Code Coverage generation to obtain correct line numbers during testing.