QCBOR is a powerful, commercial-quality CBOR encoder/decoder that completely implements these RFCs except as noted:
- RFC8949 The CBOR Standard. Everything except sorting of encoded maps is implemented.
- RFC7049 The previous CBOR standard. Replaced by RFC 8949.
- RFC8742 CBOR Sequences
- RFC8943 CBOR Dates
New Version With Spiffy Decode
This new version of QCBOR adds a more powerful decoding API called Spiffy Decode.
- Much easier implementation of decoding of CBOR protocols
- Decoding implementations parallels encoding implementation
- Overall smaller code size for implementations decoding multiple and / or complex maps
- Backwards compatible with previous versions of QCBOR
See section below for more details.
Implemented in C with minimal dependency – The only dependencies are C99, <stdint.h>, <stddef.h>, <stdbool.h> and <string.h> making it highly portable. <math.h> and <fenv.h> are used too, but their use can disabled. No #ifdefs or compiler options need to be set for QCBOR to run correctly.
Focused on C / native data representation – Careful conversion of CBOR data types in to C data types, carefully handling over and underflow, strict typing and such so the caller doesn't have to worry so much about this and so code using QCBOR passes static analyzers easier. Simpler code because there is no support for encoding/decoding to/from JSON, pretty printing, diagnostic notation... Only encoding from native C representations and decoding to native C representations is supported.
Small simple memory model – Malloc is not needed. The encode context is 174 bytes, decode context is 312 bytes and the description of decoded data item is 56 bytes. Stack use is light and there is no recursion. The caller supplies the memory to hold the encoded CBOR and encode/decode contexts so caller has full control of memory usage making it good for embedded implementations that have to run in small fixed memory.
Supports most of RFC 8949 – With some size limits, all data types and formats in the specification are supported. Map sorting is main CBOR feature that is not supported. The same decoding API supports both definite and indefinite-length map and array decoding. Decoding indefinite length strings is supported but requires a string allocator be set up. Encoding of indefinite length strings is planned, but not yet supported.
Extensible and general – Provides a way to handle data types that are not directly supported.
Secure coding style – Uses a construct called UsefulBuf as a discipline for very safe coding and handling of binary data.
Small code size – In the smallest configuration the object code is less than 4KB on 64-bit x86 CPUs. The design is such that object code for QCBOR APIs not used is not referenced.
Clear documented public interface – The public interface is separated from the implementation. It can be put to use without reading the source.
Comprehensive test suite – Easy to verify on a new platform or OS with the test suite. The test suite dependencies are minimal and the same as the library's.
This version with spiffy decode in fall of 2020 is a big change from the previous versions but is thoroughly tested including regression for backwards compatibility with the previous version.
Should the previous version be necessary, it is available in the branch BeforeSpiffyDecode. Please file an issue in GitHub to report any problems.
QCBOR was originally developed by Qualcomm. It was open sourced through CAF with a permissive Linux license, September 2018 (thanks Qualcomm!).
This code in Laurence's GitHub has diverged from the CAF source with some simplifications, tidying up and feature additions.
There is a simple makefile for the UNIX style command line binary that compiles everything to run the tests.
These eleven files, the contents of the src and inc directories, make up the entire implementation.
For most use cases you should just be able to add them to your project. Hopefully the easy portability of this implementation makes this work straight away, whatever your development environment is.
The test directory includes the tests that are nearly as portable as the main implementation. If your development environment doesn't support UNIX style command line and make, you should be able to make a simple project and add the test files to it. Then just call RunTests() to invoke them all.
While this code will run fine without configuration, there are several C pre processor macros that can be #defined in order to:
- use a more efficient implementation
- to reduce code size
- to improve performance (a little)
- remove features to reduce code size
See the comment sections on "Configuration" in inc/UsefulBuf.h and the pre processor defines that start with QCBOR_DISABLE_XXX.
In Fall 2020 a large addition makes the decoder more powerful and easy to use. Backwards compatibility with the previous API is retained as the new decoding features layer on top of it.
The first noticable addition are functions to get particular data types. These are an alternative to and built on top of QCBORDecode_GetNext() that does the type checking and in some cases sophisticated type conversion. They track an error state internally so the caller doesn't need to. They also handle the CBOR tagged data types thoroughly and properly.
In line with all the new get functions for non-aggregate types there are new functions for aggregate types. When a map is expected, QCBORDecode_EnterMap() can be called to descend into and searched by label. Duplicate detection of map items is performed. There is a similar facility for arrays and byte-string wrapped CBOR.
An outcome of all this is that now the decoding implementation of some data can look very similar to the encoding of some data and is generally easier to implement. Following is an example of first encoding a map with three items and then decoding it.
/* Encode */ QCBOREncode_Init(&EncodeCtx, Buffer); QCBOREncode_OpenMap(&EncodeCtx); QCBOREncode_AddTextToMap(&EncodeCtx, "Manufacturer", pE->Manufacturer); QCBOREncode_AddInt64ToMap(&EncodeCtx, "Displacement", pE->uDisplacement); QCBOREncode_AddInt64ToMap(&EncodeCtx, "Horsepower", pE->uHorsePower); QCBOREncode_CloseMap(&EncodeCtx); uErr = QCBOREncode_Finish(&EncodeCtx, &EncodedEngine); /* Decode */ QCBORDecode_Init(&DecodeCtx, EncodedEngine, QCBOR_DECODE_MODE_NORMAL); QCBORDecode_EnterMap(&DecodeCtx); QCBORDecode_GetTextStringInMapSZ(&DecodeCtx, "Manufacturer", &(pE->Manufacturer)); QCBORDecode_GetInt64InMapSZ(&DecodeCtx, "Displacement", &(pE->uDisplacement)); QCBORDecode_GetInt64InMapSZ(&DecodeCtx, "Horsepower", &(pE->uHorsePower)); QCBORDecode_ExitMap(&DecodeCtx); uErr = QCBORDecode_Finish(&DecodeCtx);
The spiffy decode version of QCBOR also handles CBOR tags in a simpler and more thorough way.
The spiffy decode functions will handle definite and indefinite length maps and arrays without the caller having to do anything. This includes mixed definite and indefinte maps and arrays. (Some work remains to support map searching with indefinite length strings.)
See the PR in GitHub for a more detailed list of changes.
Encoding of MIME tags now uses tag 257 instead of 36. Tag 257 accommodates binary and text-based MIME messages where tag 36 does not. Decoding supports either.
The number of nested tags on a data item is limited to four. Previously it was unlimited.
Some of the error codes have changed.
Floating Point Support & Configuration
By default, all QCBOR floating-point features are enabled:
- Encoding and decoding of basic float types, single and double-precision
- Encoding and decoding of half-precision with conversion to/from single and double-precision
- Preferred serialization of floating-point
- Floating point dates
- Methods that can convert big numbers, decimal fractions and other numbers to/from floating-point
If full floating-point is not needed, the following #defines can be used to reduce object code size and dependency.
See discussion in qcbor_encode.h for other details.
This removes dependency on:
- Floating-point hardware and floating-point instructions
- The math library (libm, -lm)
For most limited environments, this removes enough floating-point dependencies to be able to compile and run QCBOR.
Note that this does not remove use of the types double and float from QCBOR, but it limits QCBOR's use of them to converting the encoded byte stream to them and copying them. Converting and copying them usually don't require any hardware, libraries or includes. The C compiler takes care of it on its own.
QCBOR uses its own implementation of half-precision float-pointing that doesn't depend on math libraries. It uses masks and shifts instead. Thus, even with this define, half-precision encoding and decoding works.
When this is defined, the QCBOR functionality lost is minimal and only for decoding:
- Decoding floating-point format dates are not handled
- There is no conversion between floats and integers when decoding. For example, QCBORDecode_GetUInt64ConvertAll() will be unable to convert to and from float-point.
- Floats will be unconverted to double when decoding.
No interfaces are disabled or removed with this define. If input that
requires floating-point conversion or functions are called that
request floating-point conversion, an error code like
QCBOR_ERR_HW_FLOAT_DISABLED will be returned.
This saves only a small amount of object code. The primary purpose for defining this is to remove dependency on floating point hardware and libraries.
This eliminates support for half-precision and CBOR preferred serialization by disabling QCBOR's shift and mask based implementation of half-precision floating-point.
With this defined, single and double-precision floating-point numbers can still be encoded and decoded. Conversion of floating-point to and from integers, big numbers and such is also supported. Floating-point dates are still supported.
The primary reason to define this is to save object code. Roughly 900 bytes are saved, though about half of this can be saved just by not calling any functions that encode floating-point numbers.
This eliminates floating point support completely (along with related function headers). This is useful if the compiler options deny the usage of floating point operations completely, and the usage soft floating point ABI is not possible.
Compilers support a number of options that control
which float-point related code is generated. For example,
it is usually possible to give options to the compiler to avoid all
floating-point hardware and instructions, to use software
and replacement libraries instead. These are usually
bigger and slower, but these options may still be useful
in getting QCBOR to run in some environments in
-mfloat-abi=soft, disables use of
hardware instructions for the float and double
types in C for some architectures.
Comparison to TinyCBOR
TinyCBOR is a popular widely used implementation. Like QCBOR, it is a solid, well-maintained commercial quality implementation. This section is for folks trying to understand the difference in the approach between QCBOR and TinyCBOR.
TinyCBOR's API is a bit more minimalist and closer to the CBOR encoding mechanics than QCBOR's. QCBOR's API is at a somewhat higher level of abstraction.
QCBOR really does implement just about everything described in RFC 8949. The main part missing is sorting of maps when encoding. TinyCBOR implements a smaller part of the standard.
No detailed code size comparison has been made, but in a spot check that encodes and decodes a single integer shows QCBOR about 25% larger. QCBOR encoding is actually smaller, but QCBOR decoding is larger. This includes the code to call the library, which is about the same for both libraries, and the code linked from the libraries. QCBOR is a bit more powerful, so you get value for the extra code brought in, especially when decoding more complex protocols.
QCBOR tracks encoding and decoding errors internally so the caller doesn't have to check the return code of every call to an encode or decode function. In many cases the error check is only needed as the last step or an encode or decode. TinyCBOR requires an error check on each call.
QCBOR provides a substantial feature that allows searching for data items in a map by label. It works for integer and text string labels (and at some point byte-string labels). This includes detection of items with duplicate labels. This makes the code for decoding CBOR simpler, similar to the encoding code and easier to read. TinyCBOR supports search by string, but no integer, nor duplicate detection.
QCBOR provides explicit support many of the registered CBOR tags. For example, QCBOR supports big numbers and decimal fractions including their conversion to floats, uint64_t and such.
Generally, QCBOR supports safe conversion of most CBOR number formats into number formats supported in C. For example, a data item can be fetched and converted to a C uint64_t whether the input CBOR is an unsigned 64-bit integer, signed 64-bit integer, floating-point number, big number, decimal fraction or a big float. The conversion is performed with full proper error detection of overflow and underflow.
QCBOR has a special feature for decoding byte-string wrapped CBOR. It treats this similar to entering an array with one item. This is particularly use for CBOR protocols like COSE that make use of byte-string wrapping. The implementation of these protocols is simpler and uses less memory.
QCBOR's test suite is written in the same portable C that QCBOR is where TinyCBOR requires Qt for its test. QCBOR's test suite is designed to be able to run on small embedded devices the same as QCBOR.
These are approximate sizes on a 64-bit x86 CPU with the -Os optimization.
| | smallest | largest | |---------------|----------|---------| | encode only | 850 | 2100 | | decode only | 2000 | 13300 | | combined | 2850 | 15500 |
From the table above, one can see that the amount of code pulled in from the QCBOR library varies a lot, ranging from 1KB to 15KB. The main factor is in this is the number of QCBOR functions called and which ones they are. QCBOR is constructed with less internal interdependency so only code necessary for the called functions is brought in.
Encoding is simpler and smaller. An encode-only implementation may bring in only 1KB of code.
Encoding of floating-point brings in a little more code as does encoding of tagged types and encoding of bstr wrapping.
Basic decoding using QCBORDecode_GetNext() brings in 3KB.
Use of the supplied MemPool by calling QCBORDecode_SetMemPool() to setup to decode indefinite-length strings adds 0.5KB.
Basic use of spiffy decode to brings in about 3KB. Using more spiffy decode functions, such as those for tagged types bstr wrapping brings in more code.
Finally, use of all of the integer conversion functions will bring in about 5KB, though you can use the simpler ones like QCBORDecode_GetInt64() without bringing in very much code.
In addition to using fewer QCBOR functions, the following are some ways to make the code smaller.
The gcc compiler output is usually smaller than llvm because stack guards are off by default (be sure you actually have gcc and not llvm installed to be invoked by the gcc command). You can also turn off stack gaurds with llvm. It is safe to turn off stack gaurds with this code because Usefulbuf provides similar defenses and this code was carefully written to be defensive.
Disable features with defines like:
QCBOR_DISABLE_EXP_AND_MANTISSA (saves about 400 bytes)
QCBOR_DISABLE_ENCODE_USAGE_GUARDS (saves about 150), and
QCBOR_DISABLE_PREFERRED_FLOAT (saves about 900 bytes), and
QCBOR_DISABLE_INDEFINITE_LENGTH_STRINGS (saves about 400 bytes).
QCBOR_DISABLE_INDEFINITE_LENGTH_ARRAYS (saves about 200 bytes). QCBOR_DISABLE_UNCOMMON_TAGS (saves about 100 bytes).
If QCBOR is installed as a shared library, then of course only one copy of the code is in memory no matter how many applications use it.
Size of spiffy decode
When creating a decode implementation, there is a choice of whether or not to use spiffy decode features or to just use QCBORDecode_GetNext().
The implementation using spiffy decode will be simpler resulting in the calling code being smaller, but the amount of code brought in from the QCBOR library will be larger. Basic use of spiffy decode brings in about 2KB of object code. If object code size is not a concern, then it is probably better to use spiffy decode because it is less work, there is less complexity and less testing to worry about.
If code size is a concern, then use of QCBORDecode_GetNext() will probably result in smaller overall code size for simpler CBOR protocols. However, if the CBOR protocol is complex then use of spiffy decode may reduce overall code size. An example of a complex protocol is one that involves decoding a lot of maps or maps that have many data items in them. The overall code may be smaller because the general purpose spiffy decode map processor is the one used for all the maps.
Other Software Using QCBOR
ctoken is an implementation of EAT and CWT.
Changes from CAF Version
- Float support is restored
- Minimal length float encoding is added
- indefinite length arrays/maps are supported
- indefinite length strings are supported
- Tag decoding is changed; unlimited number of tags supported, any tag value supported, tag utility function for easier tag checking
- Addition functions in UsefulBuf
- QCBOREncode_Init takes a UsefulBuf instead of a pointer and size
- QCBOREncode_Finish takes a UsefulBufC and EncodedCBOR is remove
- bstr wrapping of arrays/maps is replaced with OpenBstrwrap
- AddRaw renamed to AddEncoded and can now only add whole arrays or maps, not partial maps and arrays (simplification; was a dangerous feature)
- Finish cannot be called repeatedly on a partial decode (some tests used this, but it is not really a good thing to use in the first place)
- UsefulOutBuf_OutUBuf changed to work differently
- UsefulOutBuf_Init works differently
- The "_3" functions are replaced with a small number of simpler functions
- There is a new AddTag functon instead of the "_3" functions, making the interface simpler and saving some code
- QCBOREncode_AddRawSimple_2 is removed (the macros that referenced still exist and work the same)
- Ganesh Kanike for porting to QSEE
- Mark Bapst for sponsorship and release as open source by Qualcomm
- Sachin Sharma for release through CAF
- Tamas Ban for porting to TF-M and 32-bit ARM
- Michael Eckel for Makefile improvements
- Jan Jongboom for indefinite length encoding
- Peter Uiterwijk for error strings and other
- Michael Richarson for CI set up and fixing some compiler warnings
- Máté Tóth-Pál for float-point disabling and other
- Dave Thaler for portability to Windows
Copyright and License
QCBOR is available under what is essentially the 3-Clause BSD License.
Files created inside Qualcomm and open-sourced through CAF (The Code Aurora Forum) have a slightly modified 3-Clause BSD License. The modification additionally disclaims NON-INFRINGEMENT.
Files created after release to CAF use the standard 3-Clause BSD License with no modification. These files have the SPDX license identifier, "SPDX-License-Identifier: BSD-3-Clause" in them.
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Copyright for this README
Copyright (c) 2018-2021, Laurence Lundblade. All rights reserved. Copyright (c) 2021, Arm Limited. All rights reserved.