OS-X & Ubuntu: Windows:

BREAKING: 0.5.3 changes behavour of builder create calls so arguments are always ordered by field id when id attributes are being used, for example MyGame_Example_Monster_create() in monster_test.fbs (#81).

The JSON parser may change the interface for parsing union vectors in a future release which requires code generation to match library versions.

FlatCC FlatBuffers in C for C

flatcc has no external dependencies except for build and compiler tools, and the C runtime library. With concurrent Ninja builds, a small client project can build flatcc with libraries, generate schema code, link the project and execute a test case in less than 2 seconds (4 incl. flatcc clone), rebuild in less than 0.2 seconds and produce binaries between 15K and 60K, read small buffers in 30ns, build FlatBuffers in about 600ns, and with a larger executable handle optional json parsing or printing in less than 2 us for a 10 field mixed type message.

• Online Forums
• Project Details
• Poll on Meson Build
• Reporting Bugs
• Status
  • Main features supported as of 0.5.3
  • Supported platforms (CI tested)
  • Platforms reported to work by users
  • Portability
• Time / Space / Usability Tradeoff
• Generated Files
  • Use of Macros in Generated Code
  • Extracting Documentation
• Using flatcc
• Quickstart
  • Reading a Buffer
  • Compiling for Read-Only
  • Building a Buffer
  • Verifying a Buffer
  • Potential Name Conflicts
  • Debugging a Buffer
• File and Type Identifiers
  • File Identifiers
  • Type Identifiers
• JSON Parsing and Printing
  • Base64 Encoding
  • Generic Parsing and Printing.
  • Performance Notes
• Global Scope and Included Schema
• Required Fields and Duplicate Fields
• Fast Buffers
• Types
• Unions
  • Union Scope Resolution
• Endianness
• Pitfalls in Error Handling
• Searching and Sorting
• Null Values
• Portability Layer
• Building
  • Unix Build (OS-X, Linux, related)
  • Windows Build (MSVC)
  • Docker
  • Cross-compilation
  • Custom Allocation
  • Shared Libraries
• Distribution
  • Unix Files
  • Windows Files
• Running Tests on Unix
• Running Tests on Windows
• Configuration
• Using the Compiler and Builder library
• FlatBuffers Binary Format
• Security Considerations
• Benchmarks

Online Forums

• Google Groups - FlatBuffers
• Gitter - FlatBuffers

Project Details

NOTE: see CHANGELOG. There are occassionally minor breaking changes as API inconsistencies are discovered. Unless clearly stated, breaking changes will not affect the compiled runtime library, only the header files. In case of trouble, make sure the flatcc tool is same version as the include/flatcc path.

The project includes:

• an executable flatcc FlatBuffers schema compiler for C and a corresponding library libflatcc.a. The compiler generates C header files or a binary flatbuffers schema.
• a typeless runtime library libflatccrt.a for building and verifying flatbuffers from C. Generated builder headers depend on this library. It may also be useful for other language interfaces. The library maintains a stack state to make it easy to build buffers from a parser or similar.
• a small flatcc/portable header only library for non-C11 compliant compilers, and small helpers for all compilers including endian handling and numeric printing and parsing.

See also:

• Reporting Bugs

• Google FlatBuffers

• Build Instructions

• Quickstart

• Builder Interface Reference

• Benchmarks

The flatcc compiler is implemented as a standalone tool instead of extending Googles flatc compiler in order to have a pure portable C library implementation of the schema compiler that is designed to fail graciously on abusive input in long running processes. It is also believed a C version may help provide schema parsing to other language interfaces that find interfacing with C easier than C++. The FlatBuffers team at Googles FPL lab has been very helpful in providing feedback and answering many questions to help ensure the best possible compatibility. Notice the name flatcc (FlatBuffers C Compiler) vs Googles flatc.

The JSON format is compatible with Googles flatc tool. The flatc tool converts JSON from the command line using a schema and a buffer as input. flatcc generates schema specific code to read and write JSON at runtime. While the flatcc approach is likely much faster and also easier to deploy, the flatc approach is likely more convenient when manually working with JSON such as editing game scenes. Both tools have their place.

NOTE: Big-endian platforms are only supported as of release 0.4.0.

Poll on Meson Build

It is being considered adding support for the Meson build system, but it would be good with some feedback on this via issue #56

Reporting Bugs

If possible, please provide a short reproducible schema and source file with a main program the returns 1 on error and 0 on success and a small build script. Preferably generate a hexdump and call the buffer verifier to ensure the input is valid and link with the debug library flatccrt_d.

See also Debugging a Buffer, and readfile.h useful for reading an existing buffer for verification.

Example:

samples/bugreport

eclectic.fbs :

namespace Eclectic;

enum Fruit : byte { Banana = -1, Orange = 42 }
table FooBar {
    meal      : Fruit = Banana;
    density   : long (deprecated);
    say       : string;
    height    : short;
}
file_identifier "NOOB";
root_type FooBar;

myissue.c :

/* Minimal test with all headers generated into a single file. */
#include "build/myissue_generated.h"
#include "flatcc/support/hexdump.h"

int main(int argc, char *argv[])
{
    int ret;
    void *buf;
    size_t size;
    flatcc_builder_t builder, *B;

    (void)argc;
    (void)argv;

    B = &builder;
    flatcc_builder_init(B);

    Eclectic_FooBar_start_as_root(B);
    Eclectic_FooBar_say_create_str(B, "hello");
    Eclectic_FooBar_meal_add(B, Eclectic_Fruit_Orange);
    Eclectic_FooBar_height_add(B, -8000);
    Eclectic_FooBar_end_as_root(B);
    buf = flatcc_builder_get_direct_buffer(B, &size);
#if defined(PROVOKE_ERROR) || 0
    /* Provoke error for testing. */
    ((char*)buf)[0] = 42;
#endif
    ret = Eclectic_FooBar_verify_as_root(buf, size);
    if (ret) {
        hexdump("Eclectic.FooBar buffer for myissue", buf, size, stdout);
        printf("could not verify Electic.FooBar table, got %s\n", flatcc_verify_error_string(ret));
    }
    flatcc_builder_clear(B);
    return ret;
}

build.sh :

#!/bin/sh
cd $(dirname $0)

FLATBUFFERS_DIR=../..
NAME=myissue
SCHEMA=eclectic.fbs
OUT=build

FLATCC_EXE=$FLATBUFFERS_DIR/bin/flatcc
FLATCC_INCLUDE=$FLATBUFFERS_DIR/include
FLATCC_LIB=$FLATBUFFERS_DIR/lib

mkdir -p $OUT
$FLATCC_EXE --outfile $OUT/${NAME}_generated.h -a $SCHEMA || exit 1
cc -I$FLATCC_INCLUDE -g -o $OUT/$NAME $NAME.c -L$FLATCC_LIB -lflatccrt_d || exit 1
echo "running $OUT/$NAME"
if $OUT/$NAME; then
    echo "success"
else
    echo "failed"
    exit 1
fi

Status

Release 0.5.3 (not released) BREAKING: 0.5.3 changes behavour of builder create calls so arguments are always ordered by field id when id attributes are being used, for example MyGame_Example_Monster_create() in monster_test.fbs (#81).

Release 0.5.2 introduces optional _get suffix to reader methods. By using flatcc -g only _get methods are valid. This removes potential name conficts for some field names. 0.5.2 also introduces the long awaited clone operation for tables and vectors. A C++ smoketest was added to reduce the number void pointer assignment errors that kept sneaking in. The runtime library now needs an extra file refmap.c.

Release 0.5.1 fixes a buffer overrun in the JSON printer and improves the portable libraries <stdalign.h> compatibility with C++ and the embedded newlib standard library. JSON printing and parsing has been made more consistent to help parse and print tables other than the schema root as seen in the test driver in test_json.c. The monster_test.fbs file has been reorganized to keep the Monster table more consistent with Googles flatc version and a minor schema namespace inconsistency has been resolved as a result. Explicit references to portable headers have been moved out of generated source. extern "C" C++ guards added around generated headers. 0.5.1 also cleaned up the low-level union interface so the terms { type, value } are used consistently over { type, member } and { types, members }.

Main features supported as of 0.5.3

• generated FlatBuffers reader and builder headers for C
• generated FlatBuffers verifier headers for C
• generated FlatBuffers JSON parser and printer for C
• ability to concatenate all output into one file, or to stdout
• robust dependency file generation for build systems
• binary schema (.bfbs) generation
• pre-generated reflection headers for handling .bfbs files
• cli schema compiler and library for compiling schema
• runtime library for builder, verifier and JSON support
• thorough test cases
• monster sample project
• fast build times
• support for big endian platforms (as of 0.4.0)
• support for big endian encoded flatbuffers on both le and be platforms. Enabled on be branch.
• size prefixed buffers - see also Builder Interface Reference
• flexible configuration of malloc alternatives and runtime aligned_alloc/free support in builder library.
• feature parity with C++ FlatBuffers schema features added in 2017 adding support for union vectors and mixed type unions of strings, structs, and tables, and type aliases for uint8, ..., float64.
• base64(url) encoded binary data in JSON.

There are no plans to make frequent updates once the project becomes stable, but input from the community will always be welcome and included in releases where relevant, especially with respect to testing on different target platforms.

Supported platforms (CI tested)

The ci-more branch tests additional compilers:

• Ubuntu Trusty gcc 4.4, 4.6-4.9, 5, 6, 7 and clang 3.6, 3.8
• OS-X current clang / gcc
• Windows MSVC 2010, 2013, 2015, 2015 Win64, 2017, 2017 Win64
• C++11/C++14 user code on the above platforms.

C11/C++11 is the reference that is expected to always work.

MSVC 2017 is not always tested because the CI environment then won't support MSVC 2010.

Older/non-standard versions of C++ compilers cause problems because static_assert and alignas behave in strange ways where they are neither absent nor fully working as expected. There are often workarounds, but it is more reliable to use -std=c++11 or -std=c++14.

Some previously testet compiler versions may have been retired as the CI environment gets updated. See .travis.yml and appveyor.yml in the ci-more branch for the current configuration.

The monster sample does not work with MSVC 2010 because it intentionally uses C99 style code to better follow the C++ version.

Platforms reported to work by users

• ESP32 SoC SDK with FreeRTOS and newlib has been reported to compile cleanly with C++ 14 using flatcc generated JSON parsers, as of flatcc 0.5.1.
• FreeRTOS when using custom memory allocation methods.
• IBM XLC on AIX big endian Power PC has been tested for release 0.4.0 but is not part of regular release tests.

Portability

There is no reason why other or older compilers cannot be supported, but it may require some work in the build configuration and possibly updates to the portable library. The above is simply what has been tested and configured.

The portability layer has some features that are generally important for things like endian handling, and others to provide compatibility for optional and missing C11 features. Together this should support most C compilers around, but relies on community feedback for maturity.

The necessary size of the runtime include files can be reduced significantly by using -std=c11 and avoiding JSON (which needs a lot of numeric parsing support), and by removing include/flatcc/reflection which is present to support handling of binary schema files and can be generated from reflection/reflection.fbs, and removing include/flatcc/support which is only used for tests and samples. The exact set of required files may change from release to release, and it doesn't really matter with respect to the compiled code size.

Time / Space / Usability Tradeoff

The priority has been to design an easy to use C builder interface that is reasonably fast, suitable for both servers and embedded devices, but with usability over absolute performance - still the small buffer output rate is measured in millons per second and read access 10-100 millon buffers per second from a rough estimate. Reading FlatBuffers is more than an order of magnitude faster than building them.

For 100MB buffers with 1000 monsters, dynamically extended monster names, monster vector, and inventory vector, the bandwidth reaches about 2.2GB/s and 45ms/buffer on 2.2GHz Haswell Core i7 CPU. This includes reading back and validating all data. Reading only a few key fields increases bandwidth to 2.7GB/s and 37ms/op. For 10MB buffers bandwidth may be higher but eventually smaller buffers will be hit by call overhead and thus we get down to 300MB/s at about 150ns/op encoding small buffers. These numbers are just a rough guideline - they obviously depend on hardware, compiler, and data encoded. Measurements are excluding an ininitial warmup step.

The generated JSON parsers are roughly 4 times slower than building a FlatBuffer directly in C or C++, or about 2200ns vs 600ns for a 700 byte JSON message. JSON parsing is thus roughly two orders of magnitude faster than reading the equivalent Protocol Buffer, as reported on the Google FlatBuffers Benchmarks page. LZ4 compression would estimated double the overall processing time of JSON parsing. JSON printing is faster than parsing but not very significantly so. JSON compresses to roughly half the size of compressed FlatBuffers on large buffers, but compresses worse on small buffers (not to mention when not compressing at all).

It should be noted that FlatBuffer read performance exclude verification which JSON parsers and Protocol Buffers inherently include by their nature. Verification has not been benchmarked, but would presumably add less than 50% read overhead unless only a fraction of a large buffer is to be read.

See also Benchmarks.

The client C code can avoid almost any kind of allocation to build buffers as a builder stack provides an extensible arena before committing objects - for example appending strings or vectors piecemeal. The stack is mostly bypassed when a complete object can be constructed directly such as a vector from integer array on little endian platforms.

The reader interface should be pretty fast as is with less room for improvement performance wise. It is also much simpler than the builder.

Usability has also been prioritized over smallest possible generated source code and compile time. It shouldn't affect the compiled size by much.

The compiled binary output should be reasonably small for everything but the most restrictive microcontrollers. A 33K monster source test file (in addition to the generated headers and the builder library) results in a less than 50K optimized binary executable file including overhead for printf statements and other support logic, or a 30K object file excluding the builder library.

Read-only binaries are smaller but not necessarily much smaller than builders considering they do less work: The compatibility test reads a pre-generated binary monsterdata_test.golden monster file and verifies that all content is as expected. This results in a 13K optimized binary executable or a 6K object file. The source for this check is 5K excluding header files. Readers do not need to link with a library.

JSON parsers bloat the compiled C binary compared to pure Flatbuffer usage because they inline the parser decision tree. A JSON parser for monster.fbs may add 100K +/- optimization settings to the executable binary.

Generated Files

The generated code for building flatbuffers, and for parsing and printing flatbuffers, all need access to include/flatcc. The reader does no rely on any library but all other generated files rely on the libflatccrt.a runtime library. Note that libflatcc.a is only required if the flatcc compiler itself is required as a library.

The reader and builder rely on generated common reader and builder header files. These common file makes it possible to change the global namespace and redefine basic types (uoffset_t etc.). In the future this might move into library code and use macros for these abstractions and eventually have a set of predefined files for types beyond the standard 32-bit unsigned offset (uoffset_t). The runtime library is specific to one set of type definitions.

Refer to monster_test.c and the generated files for detailed guidance on use. The monster schema used in this project is a slight adaptation to the original to test some additional edge cases.

For building flatbuffers a separate builder header file is generated per schema. It requires a flatbuffers_common_builder.h file also generated by the compiler and a small runtime library libflatccrt.a. It is because of this requirement that the reader and builder generated code is kept separate. Typical uses can be seen in the monster_test.c file. The builder allows of repeated pushing of content to a vector or a string while a containing table is being updated which simplifies parsing of external formats. It is also possible to build nested buffers in-line - at first this may sound excessive but it is useful when wrapping a union of buffers in a network interface and it ensures proper alignment of all buffer levels.

For verifying flatbuffers, a myschema_verifier.h is generated. It depends on the runtime library and the reader header.

Json parsers and printers generate one file per schema file and included schema will have their own parsers and printers which including parsers and printers will depend upon, rather similar to how builders work.

Low level note: the builder generates all vtables at the end of the buffer instead of ad-hoc in front of each table but otherwise does the same deduplication of vtables. This makes it possible to cluster vtables in hot cache or to make sure all vtables are available when partially transmitting a buffer. This behavior can be disabled by a runtime flag.

Because some use cases may include very constrained embedded devices, the builder library can be customized with an allocator object and a buffer emitter object. The separate emitter ensures a buffer can be constructed without requiring a full buffer to be present in memory at once, if so desired.

The typeless builder library is documented in flatcc_builder.h and flatcc_emitter.h while the generated typed builder api for C is documented in Builder Interface Reference.

Use of Macros in Generated Code

Occasionally a concern is raised about the dense nature of the macros used in the generated code. These macros make it difficult to understand which functions are actually available. The Builder Interface Reference attempts to document the operations in general fashion. To get more detailed information, generated function prototypes can be extracted with the scripts/flatcc-doc.sh script.

Some are also concerned with macros being "unsafe". Macros are not unsafe when used with FlatCC because they generate static or static inline functions. These will trigger compile time errors if used incorrectly to the same extend that they would in direct C code.

The expansion compresses the generated output by more than a factor 10 ensuring that code under source control does not explode and making it possible to compare versions of generated code in a meaningful manner and see if it matches the intended schema. The macros are also important for dealing with platform abstractions via the portable headers.

Still, it is possible to see the generated output although not supported directly by the build system. As an example, include/flatcc/reflection contains pre-generated header files for the reflection schema. To see the expanded output using the clang compiler tool chain, run:

clang -E -DNDEBUG -I include \
		include/flatcc/reflection/reflection_reader.h | \
clang-format

Other similar commands are likely available on platforms not supporting clang.

Note that the compiler will optimize out nearly all of the generated code and only use the logic actually referenced by end-user code because the functions are static or static inline. The remaining parts generally inline efficiently into the application code resulting in a reasonably small binary code size.

More details can be found in #88

Extracting Documentation

The expansion of generated code can be used to get documentation for a specific object type.

The following script automates this process:

scripts/flatcc-doc.sh <schema-file> <name-prefix> [<outdir>]

writing function prototypes to <outdir>/<name-prefix>.doc.

Note that the script requires the clang compiler and the clang-format tool, but the script could likely be adapted for other tool chains as well.

The principle behind the script can be illustrated using the reflection schema as an example, where documentation for the Object table is extracted:

bin/flatcc reflection/reflection.fbs -a --json --stdout | \
    clang - -E -DNDEBUG -I include | \
    clang-format -style="WebKit" | \
	grep "^static.* reflection_Object_\w*(" | \
    cut -f 1 -d '{' | \
    grep -v deprecated | \
    grep -v ");" | \
    sed 's/__tmp//g' | \
    sed 's/)/);/g'

The WebKit style of clang-format ensures that parameters and the return type are all placed on the same line. Grep extracts the function headers and cut strips function bodies starting on the same line. Sed strips __tmp suffix from parameter names used to avoid macro name conflicts. Grep strips ); to remove redundant forward declarations and sed then adds ; to make each line a valid C prototype.

The above is not guaranteed to always work as output may change, but it should go a long way.

A small extract of the output, as of flatcc-v0.5.2

static inline size_t reflection_Object_vec_len(reflection_Object_vec_t vec);
static inline reflection_Object_table_t reflection_Object_vec_at(reflection_Object_vec_t vec, size_t i);
static inline reflection_Object_table_t reflection_Object_as_root_with_identifier(const void* buffer, const char* fid);
static inline reflection_Object_table_t reflection_Object_as_root_with_type_hash(const void* buffer, flatbuffers_thash_t thash);
static inline reflection_Object_table_t reflection_Object_as_root(const void* buffer);
static inline reflection_Object_table_t reflection_Object_as_typed_root(const void* buffer);
static inline flatbuffers_string_t reflection_Object_name_get(reflection_Object_table_t t);
static inline flatbuffers_string_t reflection_Object_name(reflection_Object_table_t t);
static inline int reflection_Object_name_is_present(reflection_Object_table_t t);
static inline size_t reflection_Object_vec_scan_by_name(reflection_Object_vec_t vec, const char* s);
static inline size_t reflection_Object_vec_scan_n_by_name(reflection_Object_vec_t vec, const char* s, int n);
...

Examples are provided in following script using the reflection and monster schema:

scripts/reflection-doc-example.sh
scripts/monster-doc-example.sh

The monster doc example essentially calls:

scripts/flatcc-doc.sh samples/monster/monster.fbs MyGame_Sample_Monster_

resulting in the file MyGame_Sample_Monster_.doc:

static inline size_t MyGame_Sample_Monster_vec_len(MyGame_Sample_Monster_vec_t vec);
static inline MyGame_Sample_Monster_table_t MyGame_Sample_Monster_vec_at(MyGame_Sample_Monster_vec_t vec, size_t i);
static inline MyGame_Sample_Monster_table_t MyGame_Sample_Monster_as_root_with_identifier(const void* buffer, const char* fid);
static inline MyGame_Sample_Monster_table_t MyGame_Sample_Monster_as_root_with_type_hash(const void* buffer, flatbuffers_thash_t thash);
static inline MyGame_Sample_Monster_table_t MyGame_Sample_Monster_as_root(const void* buffer);
static inline MyGame_Sample_Monster_table_t MyGame_Sample_Monster_as_typed_root(const void* buffer);
static inline MyGame_Sample_Vec3_struct_t MyGame_Sample_Monster_pos_get(MyGame_Sample_Monster_table_t t);
static inline MyGame_Sample_Vec3_struct_t MyGame_Sample_Monster_pos(MyGame_Sample_Monster_table_t t);
static inline int MyGame_Sample_Monster_pos_is_present(MyGame_Sample_Monster_table_t t);
static inline int16_t MyGame_Sample_Monster_mana_get(MyGame_Sample_Monster_table_t t);
static inline int16_t MyGame_Sample_Monster_mana(MyGame_Sample_Monster_table_t t);
static inline const int16_t* MyGame_Sample_Monster_mana_get_ptr(MyGame_Sample_Monster_table_t t);
static inline int MyGame_Sample_Monster_mana_is_present(MyGame_Sample_Monster_table_t t);
static inline size_t MyGame_Sample_Monster_vec_scan_by_mana(MyGame_Sample_Monster_vec_t vec, int16_t key);
static inline size_t MyGame_Sample_Monster_vec_scan_ex_by_mana(MyGame_Sample_Monster_vec_t vec, size_t begin, size_t end, int16_t key);
...

FlatBuffer native types can also be extracted, for example string operations:

scripts/flatcc-doc.sh samples/monster/monster.fbs flatbuffers_string_

resulting in flatbuffers_string_.doc:

static inline size_t flatbuffers_string_len(flatbuffers_string_t s);
static inline size_t flatbuffers_string_vec_len(flatbuffers_string_vec_t vec);
static inline flatbuffers_string_t flatbuffers_string_vec_at(flatbuffers_string_vec_t vec, size_t i);
static inline flatbuffers_string_t flatbuffers_string_cast_from_generic(const flatbuffers_generic_t p);
static inline flatbuffers_string_t flatbuffers_string_cast_from_union(const flatbuffers_union_t u);
static inline size_t flatbuffers_string_vec_find(flatbuffers_string_vec_t vec, const char* s);
static inline size_t flatbuffers_string_vec_find_n(flatbuffers_string_vec_t vec, const char* s, size_t n);
static inline size_t flatbuffers_string_vec_scan(flatbuffers_string_vec_t vec, const char* s);
static inline size_t flatbuffers_string_vec_scan_n(flatbuffers_string_vec_t vec, const char* s, size_t n);
static inline size_t flatbuffers_string_vec_scan_ex(flatbuffers_string_vec_t vec, size_t begin, size_t end, const char* s);
...

Using flatcc

Refer to flatcc -h for details.

An online version listed here: flatcc-help.md but please use flatcc -h for an up to date reference.

The compiler can either generate a single header file or headers for all included schema and a common file and with or without support for both reading (default) and writing (-w) flatbuffers. The simplest option is to use (-a) for all and include the myschema_builder.h file.

The (-a) or (-v) also generates a verifier file.

Make sure flatcc under the include folder is visible in the C compilers include path when compiling flatbuffer builders.

The flatcc (-I) include path will assume all schema files with same base name (case insentive) are identical and will only include the first. All generated files use the input basename and will land in working directory or the path set by (-o).

Files can be generated to stdout using (--stdout). C headers will be ordered and concatenated, but are otherwise identical to the separate file output. Each include statement is guarded so this will not lead to missing include files.

The generated code, especially with all combined with --stdout, may appear large, but only the parts actually used will take space up the the final executable or object file. Modern compilers inline and include only necessary parts of the statically linked builder library.

JSON printer and parser can be generated using the --json flag or --json-printer or json-parser if only one of them is required. There are some certain runtime library compile time flags that can optimize out printing symbolic enums, but these can also be disabled at runtime.

Quickstart

After building the flatcc tool, binaries are located in the bin and lib directories under the flatcc source tree.

You can either jump directly to the monster example that follows Googles FlatBuffers Tutorial, or you can read along the quickstart guide below. If you follow the monster tutorial, you may want to clone and build flatcc and copy the source to a separate project directory as follows:

git clone https://github.com/dvidelabs/flatcc.git
flatcc/scripts/setup.sh -a mymonster
cd mymonster
scripts/build.sh
build/mymonster

scripts/setup.sh will as a minimum link the library and tool into a custom directory, here mymonster. With (-a) it also adds a simple build script, copies the example, and updates .gitignore - see scripts/setup.sh -h. Setup can also build flatcc, but you still have to ensure the build environment is configured for your system.

To write your own schema files please follow the main FlatBuffers project documentation on writing schema files.

The Builder Interface Reference may be useful after studying the monster sample and quickstart below.

When looking for advanced examples such as sorting vectors and finding elements by a key, you should find these in the test/monster_test project.

The following quickstart guide is a broad simplification of the test/monster_test project - note that the schema is slightly different from the tutorial. Focus is on the C specific framework rather than general FlatBuffers concepts.

You can still use the setup tool to create an empty project and follow along, but there are no assumptions about that in the text below.

Reading a Buffer

Here we provide a quick example of read-only access to Monster flatbuffer - it is an adapted extract of the monster_test.c file.

First we compile the schema read-only with common (-c) support header and we add the recursion because monster_test.fbs includes other files.

flatcc -cr test/monster_test/monster_test.fbs

For simplicity we assume you build an example project in the project root folder, but in praxis you would want to change some paths, for example:

mkdir -p build/example
flatcc -cr -o build/example test/monster_test/monster_test.fbs
cd build/example

We get:

flatbuffers_common_reader.h
include_test1_reader.h
include_test2_reader.h
monster_test_reader.h

(There is also the simpler samples/monster/monster.fbs but then you won't get included schema files).

Namespaces can be long so we optionally use a macro to manage this.

#include "monster_test_reader.h"

#undef ns
#define ns(x) FLATBUFFERS_WRAP_NAMESPACE(MyGame_Example, x)

int verify_monster(void *buffer)
{
    ns(Monster_table_t) monster;
    /* This is a read-only reference to a flatbuffer encoded struct. */
    ns(Vec3_struct_t) vec;
    flatbuffers_string_t name;
    size_t offset;

    if (!(monster = ns(Monster_as_root(buffer)))) {
        printf("Monster not available\n");
        return -1;
    }
    if (ns(Monster_hp(monster)) != 80) {
        printf("Health points are not as expected\n");
        return -1;
    }
    if (!(vec = ns(Monster_pos(monster)))) {
        printf("Position is absent\n");
        return -1;
    }

    /* -3.2f is actually -3.20000005 and not -3.2 due to representation loss. */
    if (ns(Vec3_z(vec)) != -3.2f) {
        printf("Position failing on z coordinate\n");
        return -1;
    }

    /* Verify force_align relative to buffer start. */
    offset = (char *)vec - (char *)buffer;
    if (offset & 15) {
        printf("Force align of Vec3 struct not correct\n");
        return -1;
    }

    /*
     * If we retrieved the buffer using `flatcc_builder_finalize_aligned_buffer` or
     * `flatcc_builder_get_direct_buffer` the struct should also
     * be aligned without subtracting the buffer.
     */
    if (vec & 15) {
        printf("warning: buffer not aligned in memory\n");
    }

    /* ... */
    return 0;
}
/* main() {...} */

Compiling for Read-Only

Assuming our above file is monster_example.c the following are a few ways to compile the project for read-only - compilation with runtime library is shown later on.

cc -I include monster_example.c -o monster_example

cc -std=c11 -I include monster_example.c -o monster_example

cc -D FLATCC_PORTABLE -I include monster_example.c -o monster_example

The include path or source path is likely different. Some files in include/flatcc/portable are always used, but the -D FLATCC_PORTABLE flag includes additional files to support compilers lacking c11 features.

Building a Buffer

Here we provide a very limited example of how to build a buffer - only a few fields are updated. Pleaser refer to monster_test.c and the doc directory for more information.

First we must generate the files:

flatcc -a monster_test.fbs

This produces:

flatbuffers_common_reader.h
flatbuffers_common_builder.h
flatbuffers_common_verifier.h
include_test1_reader.h
include_test1_builder.h
include_test1_verifier.h
include_test2_reader.h
include_test2_builder.h
include_test2_verifier.h
monster_test_reader.h
monster_test_builder.h
monster_test_verifier.h

Note: we wouldn't actually do the readonly generation shown earlier unless we only intend to read buffers - the builder generation always generates read acces too.

By including "monster_test_builder.h" all other files are included automatically. The C compiler needs the -I include directive to access flatbuffers/flatcc_builder.h and related.

The verifiers are not required and just created because we lazily chose the -a option.

The builder must be initialized first to set up the runtime environment we need for building buffers efficiently - the builder depends on an emitter object to construct the actual buffer - here we implicitly use the default. Once we have that, we can just consider the builder a handle and focus on the FlatBuffers generated API until we finalize the buffer (i.e. access the result). For non-trivial uses it is recommended to provide a custom emitter and for example emit pages over the network as soon as they complete rather than merging all pages into a single buffer using flatcc_builder_finalize_buffer, or the simplistic flatcc_builder_get_direct_buffer which returns null if the buffer is too large. See also documentation comments in flatcc_builder.h and flatcc_emitter.h. See also flatc_builder_finalize_aligned_buffer in builder.h and the Builder Interface Reference when malloc aligned buffers are insufficent.

#include "monster_test_builder.h"

/* See [monster_test.c] for more advanced examples. */
void build_monster(flatcc_builder_t *B)
{
    ns(Vec3_t *vec);

    /* Here we use a table, but structs can also be roots. */
    ns(Monster_start_as_root(B));

    ns(Monster_hp_add(B, 80));
    /* The vec struct is zero-initalized. */
    vec = ns(Monster_pos_start(B));
    /* Native endian. */
    vec->x = 1, vec->y = 2, vec->z = -3.2f;
    /* _end call converts to protocol endian format - for LE it is a nop. */
    ns(Monster_pos_end(B));

    /* Name is required, or we get an assertion in debug builds. */
    ns(Monster_name_create_str(B, "MyMonster"));

    ns(Monster_end_as_root(B));
}

#include "flatcc/support/hexdump.h"

int main(int argc, char *argv[])
{
    flatcc_builder_t builder;
    void *buffer;
    size_t size;

    flatcc_builder_init(&builder);

    build_monster(&builder);
    /* We could also use `flatcc_builder_finalize_buffer` and free the buffer later. */
    buffer = flatcc_builder_get_direct_buffer(&builder, &size);
    assert(buffer);
    verify_monster(buffer);

    /* Visualize what we got ... */
    hexdump("monster example", buffer, size, stdout);

    /*
     * Here we can call `flatcc_builder_reset(&builder) if
     * we wish to build more buffers before deallocating
     * internal memory with `flatcc_builder_clear`.
     */

    flatcc_builder_clear(&builder);
    return 0;
}

Compile the example project:

cc -std=c11 -I include monster_example.c lib/libflatccrt.a -o monster_example

Note that the runtime library is required for building buffers, but not for reading them. If it is incovenient to distribute the runtime library for a given target, source files may be used instead. Each feature has its own source file, so not all runtime files are needed for building a buffer:

cc -std=c11 -I include monster_example.c \
    src/runtime/emitter.c src/runtime/builder.c \
    -o monster_example

Other features such as the verifier and the JSON printer and parser would each need a different file in src/runtime. Which file should be obvious from the filenames except that JSON parsing also requires the builder and emitter source files.

Verifying a Buffer

A buffer can be verified to ensure it does not contain any ranges that point outside the the given buffer size, that all data structures are aligned according to the flatbuffer principles, that strings are zero terminated, and that required fields are present.

In the builder example above, we can apply a verifier to the output:

#include "monster_test_builder.h"
#include "monster_test_verifier.h"
int ret;
...
... finalize
if ((ret = ns(Monster_verify_as_root(buffer, size, "MONS")))) {
    printf("Monster buffer is invalid: %s\n",
    flatcc_verify_error_string(ret));
}

The readfile.h utility may also be helpful in reading an existing buffer for verification.

Flatbuffers can optionally leave out the identifier, here "MONS". Use a null pointer as identifier argument to ignore any existing identifiers and allow for missing identifiers.

Nested flatbuffers are always verified with a null identifier, but it may be checked later when accessing the buffer.

The verifier does NOT verify that two datastructures are not overlapping. Sometimes this is indeed valid, such as a DAG (directed acyclic graph) where for example two string references refer to the same string in the buffer. In other cases an attacker may maliciously construct overlapping datastructures such that in-place updates may cause subsequent invalid buffers. Therefore an untrusted buffer should never be updated in-place without first rewriting it to a new buffer.

The CMake build system has build option to enable assertions in the verifier. This will break debug builds and not usually what is desired, but it can be very useful when debugging why a buffer is invalid. Traces can also be enabled so table offset and field id can be reported.

See also include/flatcc/flatcc_verifier.h.

When verifying buffers returned directly from the builder, it may be necessary to use the flatcc_builder_finalize_aligned_buffer to ensure proper alignment and use aligned_free to free the buffer (or as of v0.5.0 also flatcc_builder_aligned_free), see also the Builder Interface Reference. Buffers may also be copied into aligned memory via mmap or using the portable layers paligned_alloc.h feature which is available when including generated headers. test/flatc_compat/flatc_compat.c is an example of how this can be done. For the majority of use cases, standard allocation would be sufficient, but for example standard 32-bit Windows only allocates on an 8-byte boundary and can break the monster schema because it has 16-byte aligned fields.

Potential Name Conflicts

If unfortunate, it is possible to have a read accessor method conflict with other generated methods and typenames. Usually a small change in the schema will resolve this issue.

As of flatcc 0.5.2 read accors are generated with and without a _get suffix so it is also possible to use Monster_pos_get(monster) instead of Monster_pos(monster). When calling flatcc with option -g the read accesors will only be generated with _get suffix. This avoids potential name conflicts. An example of a conflict is a field name like pos_add when there is also a pos field because the builder interface generates the add suffix. Using the -g option avoids this problem, but it is preferable to choose another name such as added_pos when the schema can be modified.

The -g option only changes the content of the flatbuffers_common_reader.h file, so it is technically possible to use different versions of this file if they are not mixed.

If an external code generator depends on flatcc output, it should use the _get suffix because it will work with and without the -g option, but only as of version 0.5.2 or later. For human readable code it is probaly simpler to stick to the orignal naming convention without the _get suffix.

Even with the above, it is still possible to have a conflict with the union type field. If a union field is named foo, an additional field is automatically - this field is named foo_type and holds, unsurprisingly, the type of the union.

Namespaces can also cause conflicts. If a schema has the namespace Foo.Bar and table named MyTable with a field name hello, then a read accessor will be named: Foo_Bar_MyTable_hello_get. It is also possible to have a table named Bar_MyTable because _ are allowed in FlatBuffers schema names, but in this case we have name conflict in the generated the C code. FlatCC does not attempt to avoid such conflicts so such schema are considered invalid.

Debugging a Buffer

When reading a FlatBuffer does not provide the expected results, the first line of defense is to ensure that the code being tested is linked against flatccrt_d, the debug build of the runtime library. This will raise an assertion if calls to the builder are not properly balanced or if required fields are not being set.

To dig further into a buffer, call the buffer verifier and see if the buffer is actually valid with respect to the expected buffer type.

Strings and tables will be returned as null pointers when their corresponding field is not set in the buffer. User code should test for this but it might also be helpful to temporarily or permanently set the required attribute in the schema. The builder will then detect missing fields when cerating buffers and the verifier can will detect their absence in an existing buffer.

If the verifier rejects a buffer, the error can be printed (see Verifying a Buffer), but it will not say exactly where the problem was found. To go further, the verifier can be made to assert where the problem is encountered so the buffer content can be analyzed. This is enabled with:

-DFLATCC_DEBUG_VERIFY=1

Note that this will break test cases where a buffer is expected to fail verification.

To dump detailed contents of a valid buffer, or the valid contents up to the point of failure, use:

-DFLATCC_TRACE_VERIFY=1

Both of these options can be set as CMake options, or in the flatcc_rtconfig.h file.

When reporting bugs, output from the above might also prove helpful.

The JSON parser and printer can also be used to create and display buffers. The parser will use the builder API correctly or issue a syntax error or an error on required field missing. This can rule out some uncertainty about using the api correctly. The test_json.c file and test_json_parser.c have test functions that can be adapted for custom tests.

For advanced debugging the hexdump.h file can be used to dump the buffer contents. It is used in test_json.c and also in monster_test.c. See also FlatBuffers Binary Format.

File and Type Identifiers

There are two ways to identify the content of a FlatBuffer. The first is to use file identifiers which are defined in the schema. The second is to use type identifiers which are calculated hashes based on each tables name prefixed with its namespace, if any. In either case the identifier is stored at offset 4 in binary FlatBuffers, when present. Type identifiers are not to be confused with union types.

File Identifiers

The FlatBuffers schema language has the optional file_identifier declaration which accepts a 4 characer ASCII string. It is intended to be human readable. When absent, the buffer potentially becomes 4 bytes shorter (depending on padding).

The file_identifier is intended to match the root_type schema declaration, but this does not take into account that it is convenient to create FlatBuffers for other types as well. flatcc makes no special destinction for the root_type while Googles flatc JSON parser uses it to determine the JSON root object type.

As a consequence, the file identifier is ambigous. Included schema may have separate file_identifier declarations. To at least make sure each type is associated with its own schemas file_identifier, a symbol is defined for each type. If the schema has such identifier, it will be defined as the null identifier.

The generated code defines the identifiers for a given table:

#ifndef MyGame_Example_Monster_identifier
#define MyGame_Example_Monster_identifier flatbuffers_identifier
#endif

The flatbuffers_identifier is the schema specific file_identifier and is undefined and redefined for each generated _reader.h file.

The user can now override the identifier for a given type, for example:

#define MyGame_Example_Vec3_identifer "VEC3"
#include "monster_test_builder.h"

...
MyGame_Example_Vec3_create_as_root(B, ...);

The create_as_root method uses the identifier for the type in question, and so does other _as_root methods.

The file_extension is handled in a similar manner:

#ifndef MyGame_Example_Monster_extension
#define MyGame_Example_Monster_extension flatbuffers_extension
#endif

Type Identifiers

To better deal with the ambigouties of file identifiers, type identifiers have been introduced as an alternative 4 byte buffer identifier. The hash is standardized on FNV-1a for interoperability.

The type identifier use a type hash which maps a fully qualified type name into a 4 byte hash. The type hash is a 32-bit native value and the type identifier is a 4 character little endian encoded string of the same value.

In this example the type hash is derived from the string "MyGame.Example.Monster" and is the same for all FlatBuffer code generators that supports type hashes.

The value 0 is used to indicate that one does not care about the identifier in the buffer.

...
MyGame_Example_Monster_create_as_typed_root(B, ...);
buffer = flatcc_builder_get_direct_buffer(B);
MyGame_Example_Monster_verify_as_typed_root(buffer, size);
// read back
monster = MyGame_Example_Monster_as_typed_root(buffer);

switch (flatbuffers_get_type_hash(buffer)) {
case MyGame_Example_Monster_type_hash:
    ...

}
...
if (flatbuffers_get_type_hash(buffer) ==
    flatbuffers_type_hash_from_name("Some.Old.Buffer")) {
    printf("Buffer is the old version, not supported.\n");
}

More API calls are available to naturally extend the existing API. See monster_test.c for more.

The type identifiers are defined like:

#define MyGame_Example_Monster_type_hash ((flatbuffers_thash_t)0x330ef481)
#define MyGame_Example_Monster_type_identifier "\x81\xf4\x0e\x33"

The type_identifier can be used anywhere the original 4 character file identifier would be used, but a buffer must choose which system, if any, to use. This will not affect the file_extension.

NOTE: The generated _type_identifier strings should not normally be used when an identifier string is expected in the generated API because it may contain null bytes which will be zero padded after the first null before comparison. Use the API calls that take a type hash instead. The type_identifier can be used in low level flatcc_builder.h calls because it handles identifiers as a fixed byte array and handles type hashes and strings the same.

NOTE: it is possible to compile the flatcc runtime to encode buffers in big endian format rather than the standard little endian format regardless of the host platforms endianness. If this is done, the identifier field in the buffer is always byte swapped regardless of the identifier method chosen. The API calls make this transparent, so "MONS" will be stored as "SNOM" but should still be verified as "MONS" in API calls. This safeguards against mixing little- and big-endian buffers. Likewise, type hashes are always tested in native (host) endian format.

The flatcc/flatcc_identifier.h file contains an implementation of the FNV-1a hash used. The hash was chosen for simplicity, availability, and collision resistance. For better distribution, and for internal use only, a dispersion function is also provided, mostly to discourage use of alternative hashes in transmission since the type hash is normally good enough as is.

Note: there is a potential for collisions in the type hash values because the hash is only 4 bytes.

JSON Parsing and Printing

JSON support files are generated with flatcc --json.

This section is not a tutorial on JSON printing and parsing, it merely covers some non-obvious aspects. The best source to get started quickly is the test file:

test/json_test/json_test.c

For detailed usage, please refer to:

test/json_test/test_json_printer.c
test/json_test/test_json_parser.c
test/json_test/json_test.c
test/benchmark/benchflatccjson

See also JSON parsing section in the Googles FlatBuffers schema documentation.

By using the flatbuffer schema it is possible to generate schema specific JSON printers and parsers. This differs for better and worse from Googles flatc tool which takes a binary schema as input and processes JSON input and output. Here that parser and printer only rely on the flatcc runtime library, is faster (probably significantly so), but requires recompilition when new JSON formats are to be supported - this is not as bad as it sounds - it would for example not be difficult to create a Docker container to process a specific schema in a web server context.

The parser always takes a text buffer as input and produces output according to how the builder object is initialized. The printer has different init functions: one for printing to a file pointer, including stdout, one for printing to a fixed size external buffer, and one for printing to a dynamically growing buffer. The dynamic buffer may be reused between prints via the reset function. See flatcc_json_parser.h for details.

The parser will accept unquoted names (not strings) and trailing commas, i.e. non-strict JSON and also allows for hex \x03 in strings. Strict mode must be enabled by a compile time flag. In addition the parser schema specific symbolic enum values that can optionally be unquoted where a numeric value is expected:

color: Green
color: Color.Green
color: MyGame.Example.Color.Green
color: 2

The symbolic values do not have to be quoted (unless required by runtime or compile time configuration), but can be while numeric values cannot be quoted. If no namespace is provided, like color: Green, the symbol must match the receiving enum type. Any scalar value may receive a symbolic value either in a relative namespace like hp: Color.Green, or an absolute namespace like hp: MyGame.Example.Color.Green, but not hp: Green (since hp in the monster example schema) is not an enum type with a Green value). A namespace is relative to the namespace of the receiving object.

It is also possible to have multiple values, but these always have to be quoted in order to be compatible with Googles flatc tool for Flatbuffers 1.1:

color: "Green Red"

Unquoted multi-valued enums can be enabled at compile time but this is deprecated because it is incompatible with both Googles flatc JSON and also with other possible future extensions: color: Green Red

These value-valued expressions were originally intended for enums that have the bit flag attribute defined (which Color does have), but this is tricky to process, so therefore any symblic value can be listed in a sequence with or without namespace as appropriate. Because this further causes problems with signed symbols the exact definition is that all symbols are first coerced to the target type (or fail), then added to the target type if not the first this results in:

color: "Green Blue Red Blue"
color: 19

Because Green is 2, Red is 1, Blue is 8 and repeated.

NOTE: Duplicate values should be considered implemention dependent as it cannot be guaranteed that all flatbuffer JSON parsers will handle this the same. It may also be that this implementation will change in the future, for example to use bitwise or when all members and target are of bit flag type.

It is not valid to specify an empty set like:

color: ""

because it might be understood as 0 or the default value, and it does not unquote very well.

The printer will by default print valid json without any spaces and everything quoted. Use the non-strict formatting option (see headers and test examples) to produce pretty printing. It is possibly to disable symbolic enum values using the noenum option.

Only enums will print symbolic values are there is no history of any parsed symbolic values at all. Furthermore, symbolic values are only printed if the stored value maps cleanly to one value, or in the case of bit-flags, cleanly to multiple values. For exmaple if parsing color: Green Red it will print as "color":"Red Green" by default, while color: Green Blue Red Blue will print as color:19.

Both printer and parser are limited to roughly 100 table nesting levels and an additional 100 nested struct depths. This can be changed by configuration flags but must fit in the runtime stack since the operation is recursive descent. Exceedning the limits will result in an error.

Numeric values are coerced to the receiving type. Integer types will fail if the assignment does not fit the target while floating point values may loose precision silently. Integer types never accepts floating point values. Strings only accept strings.

Nested flatbuffers may either by arrays of byte sized integers, or a table or a struct of the target type. See test cases for details.

The parser will by default fail on unknown fields, but these can also be skipped silently with a runtime option.

Unions are difficult to parse. A union is two json fields: a table as usual, and an enum to indicate the type which has the same name with a _type suffix and accepts a numeric or symbolic type code:

{
  name: "Container Monster",
  test_type: Monster,
  test: { name: "Contained Monster" }
}

based on the schema is defined in monster_test.fbs.

Because other json processors may sort fields, it is possible to receive the type field after the test field. The parser does not store temporary datastructures. It constructs a flatbuffer directly. This is not possible when the type is late. This is handled by parsing the field as a skipped field on a first pass, followed by a typed back-tracking second pass once the type is known (only the table is parsed twice, but for nested unions this can still expand). Needless to say this slows down parsing. It is an error to provide only the table field or the type field alone, except if the type is NONE or 0 in which case the table is not allowed to be present.

Union vectors are supported as of v0.5.0. A union vector is represented as two vectors, one with a vector of tables and one with a vector of types, similar to ordinary unions. It is more efficient to place the type vector first because it avoids backtracking. Because a union of type NONE cannot be represented by abasence of table field when dealing with vectors of unions, a table must have the value null if its type is NONE in the corresponding type vector. In other cases a table should be absent, and not null.

Here is an example of JSON containing Monster root table with a union vector field named manyany which is a vector of Any unions in the monster_test.fbs schema:

{
    "name": "Monster",
    "manyany_type": [ "Monster", "NONE" ],
    "manyany": [{"name": "Joe"}, null]
}

Base64 Encoding

As of v0.5.0 it is possible to encode and decode a vector of type [uint8] (aka [ubyte]) as a base64 encoded string or a base64url encoded string as documented in RFC 4648. Any other type, notably the string type, do not handle base64 encoding.

Limiting the support to [uint8] avoids introducing binary data into strings and also avoids dealing with sign and endian encoding of binary data of other types. Furthermore, array encoding of values larger than 8 bits are not necessarily less efficient than base64.

Base64 padding is always printed and is optional when parsed. Spaces, linebreaks, JSON string escape character '\', or any other character not in the base64(url) alphabet are rejected as a parse error.

The schema must add the attribute (base64) or (base64url) to the field holding the vector, for example:

table Monster {
    name: string;
    sprite: [uint8] (base64);
    token: [uint8] (base64url);
}

If more complex data needs to be encoded as base64 such as vectors of structs, this can be done via nested FlatBuffers which are also of type [uint8].

Note that for some use cases it might be desireable to read binary data as base64 into memory aligned to more than 8 bits. This is not currently possible, but it is recognized that a (force_align: n) attribute on [ubyte] vectors could be useful, but it can also be handled via nested flatbuffers which also align data.

Generic Parsing and Printing.

As of v0.5.1 test_json.c demonstrates how a single parser driver can be used to parse different table types without changes to the driver or to the schema.

For example, the following layout can be used to configure a generic parser or printer.

struct json_scope {
	const char *identifier;
	flatcc_json_parser_table_f *parser;
	flatcc_json_printer_table_f *printer;
	flatcc_table_verifier_f *verifier;
};

static const struct json_scope Monster = {
	/* The is the schema global file identifier. */
	ns(Monster_identifier),
	ns(Monster_parse_json_table),
	ns(Monster_print_json_table),
	ns(Monster_verify_table)
};

The Monster scope can now be used by a driver or replaced with a new scope as needed:

/* Abbreviated ... */
struct json_scope = Monster;
flatcc_json_parser_table_as_root(B, &parser_ctx, json, strlen(json), parse_flags,
        scope->identifier, scope->parser);
/* Printing and verifying works roughly the same. */

The generated table MyGame_Example_Monster_parse_json_as_root is a thin convenience wrapper roughly implementing the above.

The generated monster_test_parse_json is a higher level convenience wrapper named of the schema file itself, not any specific table. It parses the root_type configured in the schema. This is how the test_json.c test driver operated prior to v0.5.1 but it made it hard to test parsing and printing distinct table types.

Note that verification is not really needed for JSON parsing because a generated JSON parser is supposed to build buffers that always verify (except for binary encoded nested buffers), but it is useful for testing.

Performance Notes

Note that json parsing and printing is very fast reaching 500MB/s for printing and about 300 MB/s for parsing. Floating point parsing can signficantly skew these numbers. The integer and floating point parsing and printing are handled via support functions in the portable library. In addition the floating point include/flatcc/portable/grisu3_* library is used unless explicitly disable by a compile time flag. Disabling grisu3 will revert to sprintf and strtod. Grisu3 will fall back to strtod and grisu3 in some rare special cases. Due to the reliance on strtod and because strtod cannot efficiently handle non-zero-terminated buffers, it is recommended to zero terminate buffers. Alternatively, grisu3 can be compiled with a flag that allows errors in conversion. These errors are very small and still correct, but may break some checksums. Allowing for these errors can significantly improve parsing speed and moves the benchmark from below half a million parses to above half a million parses per second on 700 byte json string, on a 2.2 GHz core-i7.

While unquoted strings may sound more efficient due to the compact size, it is actually slower to process. Furthermore, large flatbuffer generated JSON files may compress by a factor 8 using gzip or a factor 4 using LZ4 so this is probably the better place to optimize. For small buffers it may be more efficient to compress flatbuffer binaries, but for large files, json may actually compress significantly better due to the absence of pointers in the format.

SSE 4.2 has been experimentally added, but it the gains are limited because it works best when parsing space, and the space parsing is already fast without SSE 4.2 and because one might just leave out the spaces if in a hurry. For parsing strings, trivial use of SSE 4.2 string scanning doesn't work well becasuse all the escape codes below ASCII 32 must be detected rather than just searching for \ and ". That is not to say there are not gains, they just don't seem worthwhile.

The parser is heavily optimized for 64-bit because it implements an 8-byte wide trie directly in code. It might work well for 32-bit compilers too, but this hasn't been tested. The large trie does put some strain on compile time. Optimizing beyond -O2 leads to too large binaries which offsets any speed gains.

Global Scope and Included Schema

Attributes included in the schema are viewed in a global namespace and each include file adds to this namespace so a schema file can use included attributes without namespace prefixes.

Each included schema will also add types to a global scope until it sees a namespace declaration. An included schema does not inherit the namespace of an including file or an earlier included file, so all schema files starts in the global scope. An included file can, however, see other types previously defined in the global scope. Because include statements always appear first in a schema, this can only be earlier included files, not types from a containing schema.

The generated output for any included schema is indendent of how it was included, but it might not compile without the earlier included files being present and included first. By including the toplevel myschema.h or myschema_builder.h all these dependencies are handled correctly.

Note: libflatcc.a can only parse a single schema when the schema is given as a memory buffer, but can handle the above when given a filename. It is possible to concatenate schema files, but a namespace; declaration must be inserted as a separator to revert to global namespace at the start of each included file. This can lead to subtle errors because if one parent schema includes two child schema a.fbs and b.fbs, then b.fbs should not be able to see anything in a.fbs even if they share namespaces. This would rarely be a problem in praxis, but it means that schema compilation from memory buffers cannot authoratively validate a schema. The reason the schema must be isolated is that otherwise code generation for a given schema could change with how it is being used leading to very strange errors in user code.

Required Fields and Duplicate Fields

If a field is required such as Monster.name, the table end call will assert in debug mode and create incorrect tables in non-debug builds. The assertion may not be easy to decipher as it happens in library code and it will not tell which field is missing.

When reading the name, debug mode will again assert and non-debug builds will return a default value.

Writing the same field twice will also trigger an assertion in debug builds.

Fast Buffers

Buffers can be used for high speed communication by using the ability to create buffers with structs as root. In addition the default emitter supports flatcc_emitter_direct_buffer for small buffers so no extra copy step is required to get a linear buffer in memory. Preliminary measurements suggests there is a limit to how fast this can go (about 6-7 mill. buffers/sec) because the builder object must be reset between buffers which involves zeroing allocated buffers. Small tables with a simple vector achieve roughly half that speed. For really high speed a dedicated builder for structs would be needed. See also monster_test.c.

Types

All types stored in a buffer has a type suffix such as Monster_table_t or Vec3_struct_t (and namespace prefix which we leave out here). These types are read-only pointers into endian encoded data. Enum types are just constants easily grasped from the generated code. Tables are dense so they are never accessed directly.

Enums support schema evolution meaning that more names can be added to the enumeration in a future schema version. As of v0.5.0 the function _is_known_value can be used ot check if an enum value is known to the current schema version.

Structs have a dual purpose because they are also valid types in native format, yet the native reprsention has a slightly different purpose. Thus the convention is that a const pointer to a struct encoded in a flatbuffer has the type Vec3_struct_t where as a writeable pointer to a native struct has the type Vec3_t * or struct Vec3 *.

All types have a _vec_t suffix which is a const pointer to the underlying type. For example Monster_table_t has the vector type Monster_vec_t. There is also a non-const variant with suffix _mutable_vec_t which is rarely used. However, it is possible to sort vectors in-place in a buffer, and for this to work, the vector must be cast to mutable first. A vector (or string) type points to the element with index 0 in the buffer, just after the length field, and it may be cast to a native type for direct access with attention to endian encoding. (Note that table_t types do point to the header field unlike vectors.) These types are all for the reader interface. Corresponding types with a _ref_t suffix such as _vec_ref_t are used during the construction of buffers.

Native scalar types are mapped from the FlatBuffers schema type names such as ubyte to uint8_t and so forth. These types also have vector types provided in the common namespace (default flatbuffers_) so a [ubyte] vector has type flatbuffers_uint8_vec_t which is defined as const uint8_t *.

The FlatBuffers boolean type is strictly 8 bits wide so we cannot use or emulate <stdbool.h> where sizeof(bool) is implementation dependent. Therefore flatbuffers_bool_t is defined as uint8_t and used to represent FlatBuffers boolean values and the constants of same type: flatbuffers_true = 1 and flatbuffers_false = 0. Even so, pstdbool.h is available in the include/flatcc/portable directory if bool, true, and false are desired in user code and <stdbool.h> is unavailable.

flatbuffers_string_t is const char * but imply the returned pointer has a length prefix just before the pointer. flatbuffers_string_vec_t is a vector of strings. The flatbufers_string_t type guarantees that a length field is present using flatbuffers_string_len(s) and that the string is zero terminated. It also suggests that it is in utf-8 format according to the FlatBuffers specification, but not checks are done and the flatbuffers_create_string(B, s, n) call explicitly allows for storing embedded null characters and other binary data.

All vector types have operations defined as the typename with _vec_t replaced by _vec_at and _vec_len. For example flatbuffers_uint8_vec_at(inv, 1) or Monster_vec_len(inv). The length or _vec_len will be 0 if the vector is missing whereas _vec_at will assert in debug or behave undefined in release builds following out of bounds access. This also applies to related string operations.

The FlatBuffers schema uses the following scalar types: ubyte, byte, ushort, short, uint, int, ulong, and long to represent unsigned and signed integer types of length 8, 16, 32, and 64 respectively. The schema syntax has been updated to also support the type aliases uint8, int8, uint16, int16, uint32, int32, uint64, int64 to represent the same basic types. Likewise, the schema uses the types float and double to represent IEEE-754 binary32 and binary64 floating point formats where the updated syntax also supports the type aliases float32 and float64.

The C interface uses the standard C types such as uint8 and double to represent scalar types and this is unaffected by the schema type name used, so the schema vector type [float64] is represented as flatbuffers_double_vec_t the same as [double] would be.

Note that the C standard does not guarantee that the C types float and double are represented by the IEEE-754 binary32 single precision format and the binary64 double precision format respectively, although they usually are. If this is not the case FlatCC cannot work correctly with FlatBuffers floating point values. (If someone really has this problem, it would be possible to fix).

Unions are represented with a two table fields, one with a table field and one with a type field. See separate section on Unions. As of flatcc v0.5.0 union vectors are also supported.

Unions

A union represents one of several possible tables. A table with a union field such as Monster.equipped in the samples schema will have two accessors: MyGame_Sample_Monster_equipped(t) of type flatbuffers_generic_t and MyGame_Sample_Monster_equipped_type(t) of type MyGame_Sample_Equipment_union_type_t. A generic type is is just a const void pointer that can be assigned to the expected table type, struct type, or string type. The enumeration has a type code for member of the union and also MyGame_Sample_Equipment_NONE which has the value 0.

The union interface were changed in 0.5.0 and 0.5.1 to use a consistent { type, value } naming convention for both unions and union vectors in all interfaces and to support unions and union vectors of multiple types.

A union can be accessed by its field name, like Monster MyGame_Sample_Monster_equipped(t) and its type is given by MyGame_Sample_Monster_type(t), or a flatbuffers_union_t struct can be returned with MyGame_Sample_monster_union(t) with the fields { type, value }. A union vector is accessed in the same way but { type, value } represents a type vector and a vector of the given type, e.g. a vector Monster tables or a vector of strings.

There is a test in monster_test.c covering union vectors and a separate test focusing on mixed type unions that also has union vectors.

Union Scope Resolution

Googles monster_test.fbs schema has the union (details left out):

namespace MyGame.Example2;
table Monster{}

namespace MyGame.Example;
table Monster{}

union Any { Monster, MyGame.Example2.Monster }

where the two Monster tables are defined in separate namespaces.

flatcc rejects this schema due to a name conflict because it uses the basename of a union type, here Monster to generate the union member names which are also used in JSON parsing. This can be resolved by adding an explicit name such as Monster2 to resolve the conflict:

union Any { Monster, Monster2: MyGame.Example2.Monster }

This syntax is accepted by both flatc and flatcc.

Both versions will implement the same union with the same type codes in the binary format but generated code will differ in how the types are referred to.

In JSON the monster type values are now identified by MyGame.Example.Any.Monster, or just Monster, when assigning the first monster type to an Any union field, and MyGame.Example.Any.Monster2, or just Monster2 when assigning the second monster type. C uses the usual enum namespace prefixed symbols like MyGame_Example_Any_Monster2.

Endianness

The include/flatcc/portable/pendian_detect.h file detects endianness for popular compilers and provides a runtime fallback detection for others. In most cases even the runtime detection will be optimized out at compile time in release builds.

The FLATBUFFERS_LITTLEENDIAN flag is respected for compatibility with Googles flatc compiler, but it is recommended to avoid its use and work with the mostly standard flags defined and/or used in pendian_detect.h, or to provide for additional compiler support.

As of flatcc 0.4.0 there is support for flatbuffers running natively on big endian hosts. This has been tested on IBM AIX. However, always run tests against the system of interest - the release process does not cover automated tests on any BE platform.

As of flatcc 0.4.0 there is also support for compiling the flatbuffers runtime library with flatbuffers encoded in big endian format regardless of the host platforms endianness. Longer term this should probably be placed in a separate library with separate name prefixes or suffixes, but it is usable as is. Redefine FLATBUFFERS_PROTOCOL_IS_LE/BE accordingly in include/flatcc/flatcc_types.h. This is already done in the be branch. This branch is not maintained but the master branch can be merged into it as needed.

Note that standard flatbuffers are always encoded in little endian but in situations where all buffer producers and consumers are big endian, the non standard big endian encoding may be faster, depending on intrinsic byteswap support. As a curiosity, the load_test actually runs faster with big endian buffers on a little endian MacOS platform for reasons only the optimizer will know, but read performance of small buffers drop to 40% while writing buffers generally drops to 80-90% performance. For platforms without compiler intrinsics for byteswapping, this can be much worse.

Flatbuffers encoded in big endian will have the optional file identifier byteswapped. The interface should make this transparent, but details are still being worked out. For example, a buffer should always verify the monster buffer has the identifier "MONS", but internally the buffer will store the identifier as "SNOM" on big endian encoded buffers.

Because buffers can be encode in two ways, flatcc uses the term native endianness and protocol endianess. _pe is a suffix used in various low level API calls to convert between native and protocol endianness without caring about whether host or buffer is little or big endian.

If it is necessary to write application code that behaves differently if the native encoding differs from protocol encoding, use flatbuffers_is_pe_native(). This is a function, not a define, but for all practical purposes it will have same efficience while also supporting runtime endian detection where necessary.

The flatbuffer environment only supports reading either big or little endian for the time being. To test which is supported, use the define FLATBUFFERS_PROTOCOL_IS_LE or FLATBUFFERS_PROTOCOL_IS_BE. They are defines as 1 and 0 respectively.

Pitfalls in Error Handling

The builder API often returns a reference or a pointer where null is considered an error or at least a missing object default. However, some operations do not have a meaningful object or value to return. These follow the convention of 0 for success and non-zero for failure. Also, if anything fails, it is not safe to proceed with building a buffer. However, to avoid overheads, there is no hand holding here. On the upside, failures only happen with incorrect use or allocation failure and since the allocator can be customized, it is possible to provide a central error state there or to guarantee no failure will happen depending on use case, assuming the API is otherwise used correctly. By not checking error codes, this logic also optimizes out for better performance.

Searching and Sorting

The builder API does not support sorting due to the complexity of customizable emitters, but the reader API does support sorting so a buffer can be sorted at a later stage. This requires casting a vector to mutable and calling the sort method available for fields with keys.

The sort uses heap sort and can sort a vector in-place without using external memory or recursion. Due to the lack of external memory, the sort is not stable. The corresponding find operation returns the lowest index of any matching key, or flatbuffers_not_found.

When configured in config.h (the default), the flatcc compiler allows multiple keyed fields unlike Googles flatc compiler. This works transparently by providing <table_name>_vec_sort_by_<field_name> and <table_name>_vec_find_by_<field_name> methods for all keyed fields. The first field maps to <table_name>_vec_sort and <table_name>_vec_find. Obviously the chosen find method must match the chosen sort method. The find operation is O(logN).

As of v0.4.1 <table_name>_vec_scan_by_<field_name> and the default <table_name>_vec_scan are also provided, similar to find, but as a linear search that does not require the vector to be sorted. This is especially useful for searching by a secondary key (multiple keys is a non-standard flatcc feature). _scan_ex searches a sub-range [a, b) where b is an exclusive index. b = flatbuffers_end == flatbuffers_not_found == (size_t)-1 may be used when searching from a position to the end, and b can also conveniently be the result of a previous search.

rscan searches in the opposite direction starting from the last element. rscan_ex accepts the same range arguments as scan_ex. If a >= b or a >= len the range is considered empty and flatbuffers_not_found is returned. [r]scan[_ex]_n[_by_name] is for length terminated string keys. See monster_test.c for examples.

Note that find requires key attribute in the schema. scan is also available on keyed fields. By default flatcc will also enable scan by any other field but this can be disabled by a compile time flag.

Basic types such as uint8_vec also have search operations.

See also Builder Interface Reference and monster_test.c.

Null Values

The FlatBuffers format does not fully distinguish between default values and missing or null values but it is possible to force values to be written to the buffer. This is discussed further in the Builder Interface Reference. For SQL data roundtrips this may be more important that having compact data.

The _is_present suffix on table access methods can be used to detect if value is present in a vtable, for example Monster_hp_present. Unions return true of the type field is present, even if it holds the value None.

The add methods have corresponding force_add methods for scalar and enum values to force storing the value even if it is default and thus making it detectable by is_present.

Portability Layer

The portable library is placed under include/flatcc/portable and is required by flatcc, but isn't strictly part of the flatcc project. It is intended as an independent light-weight header-only library to deal with compiler and platform variations. It is placed under the flatcc include path to simplify flatcc runtime distribution and to avoid name and versioning conflicts if used by other projects.

The license of portable is different from flatcc. It is mostly MIT or Apache depending on the original source of the various parts.

A larger set of portable files is included if FLATCC_PORTABLE is defined by the user when building.

cc -D FLATCC_PORTABLE -I include monster_test.c -o monster_test

Otherwise a targeted subset is included by flatcc_flatbuffers.h in order to deal with non-standard behavior of some C11 compilers.

pwarnings.h is also always included so compiler specific warnings can be disabled where necessary.

The portable library includes the essential parts of the grisu3 library found in external/grisu3, but excludes the test cases. The JSON printer and parser relies on fast portable numeric print and parse operations based mostly on grisu3.

If a specific platform has been tested, it would be good with feedback and possibly patches to the portability layer so these can be made available to other users.

Building

Unix Build (OS-X, Linux, related)

To initialize and run the build (see required build tools below):

scripts/build.sh

The bin and lib folders will be created with debug and release build products.

The build depends on CMake. By default the Ninja build tool is also required, but alternatively make can be used.

Optionally switch to a different build tool by choosing one of:

scripts/initbuild.sh make
scripts/initbuild.sh make-concurrent
scripts/initbuild.sh ninja

where ninja is the default and make-concurrent is make with the -j flag. A custom build configuration X can be added by adding a scripts/build.cfg.X file.

scripts/initbuild.sh cleans the build if a specific build configuration is given as argument. Without arguments it only ensures that CMake is initialized and is therefore fast to run on subsequent calls. This is used by all test scripts.

To install build tools on OS-X, and build:

brew update
brew install cmake ninja
git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
scripts/build.sh

To install build tools on Ubuntu, and build:

sudo apt-get update
sudo apt-get install cmake ninja-build
git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
scripts/build.sh

To install build tools on Centos, and build:

sudo yum group install "Development Tools"
sudo yum install cmake
git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
scripts/initbuild.sh make # there is no ninja build tool
scripts/build.sh

OS-X also has a HomeBrew package:

brew update
brew install flatcc

or for the bleeding edge:

brew update
brew install flatcc --HEAD

Windows Build (MSVC)

Install CMake, MSVC, and git (tested with MSVC 14 2015).

In PowerShell:

git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
mkdir build\MSVC
cd build\MSVC
cmake -G "Visual Studio 14 2015" ..\..

Optionally also build from the command line (in build\MSVC):

cmake --build . --target --config Debug
cmake --build . --target --config Release

In Visual Studio:

open flatcc\build\MSVC\FlatCC.sln
build solution
choose Release build configuration menu
rebuild solution

Note that flatcc\CMakeList.txt sets the -DFLATCC_PORTABLE flag and that include\flatcc\portable\pwarnings.h disable certain warnings for warning level -W3.

Docker

Docker image:

• https://github.com/neomantra/docker-flatbuffers

Cross-compilation

Users have been reporting some degree of success using cross compiles from Linux x86 host to embedded ARM Linux devices.

For this to work, FLATCC_TEST option should be disabled in part because cross-compilation cannot run the cross-compiled flatcc tool, and in part because there appears to be some issues with CMake custom build steps needed when building test and sample projects.

The option FLATCC_RTONLY will disable tests and only build the runtime library.

The following is not well tested, but may be a starting point:

mkdir -p build/xbuild
cd build/xbuild
cmake ../.. -DBUILD_SHARED_LIBS=on -DFLATCC_RTONLY=on \
  -DCMAKE_BUILD_TYPE=Release

Overall, it may be simpler to create a separate Makefile and just compile the few src/runtime/*.c into a library and distribute the headers as for other platforms, unless flatcc is also required for the target. Or to simply include the runtime source and header files in the user project.

Note that no tests will be built nor run with FLATCC_RTONLY enabled. It is highly recommended to at least run the tests/monster_test project on a new platform.

Custom Allocation

Some target systems will not work with Posix malloc, realloc, free and C11 aligned_alloc. Or they might, but more allocation control is desired. The best approach is to use flatcc_builder_custom_init to provide a custom allocator and emitter object, but for simpler case or while piloting a new platform flatcc_alloc.h can be used to override runtime allocation functions. Carefully read the comments in this file if doing so. There is a test case implementing a new emitter, and a custom allocator can be copied from the one embedded in the builder library source.

Shared Libraries

By default libraries are built statically.

Occasionally there are requests #42 for also building shared libraries. It is not clear how to build both static and shared libraries at the same time without choosing some unconvential naming scheme that might affect install targets unexpectedly.

CMake supports building shared libraries out of the box using the standard library name using the following option:

CMAKE ... -DBUILD_SHARED_LIBS=ON ...

See also CMake Gold: Static + shared.

Distribution

Install targes may be built with:

mkdir -p build/install
cd build/install
cmake ../.. -DBUILD_SHARED_LIBS=on -DFLATCC_RTONLY=on \
  -DCMAKE_BUILD_TYPE=Release -DFLATCC_INSTALL=on
make install

However, this is not well tested and should be seen as a starting point. The normal scripts/build.sh places files in bin and lib of the source tree.

Unix Files

To distribute the compiled binaries the following files are required:

Compiler:

bin/flatcc                 (command line interface to schema compiler)
lib/libflatcc.a            (optional, for linking with schema compiler)
include/flatcc/flatcc.h    (optional, header and doc for libflatcc.a)

Runtime:

include/flatcc/**          (runtime header files)
include/flatcc/reflection  (optional)
include/flatcc/support     (optional, only used for test and samples)
lib/libflatccrt.a          (runtime library)

In addition the runtime library source files may be used instead of libflatccrt.a. This may be handy when packaging the runtime library along with schema specific generated files for a foreign target that is not binary compatible with the host system:

src/runtime/*.c

Windows Files

The build products from MSVC are placed in the bin and lib subdirectories:

flatcc\bin\Debug\flatcc.exe
flatcc\lib\Debug\flatcc_d.lib
flatcc\lib\Debug\flatccrt_d.lib
flatcc\bin\Release\flatcc.exe
flatcc\lib\Release\flatcc.lib
flatcc\lib\Release\flatccrt.lib

Runtime include\flatcc directory is distributed like other platforms.

Running Tests on Unix

Run

scripts/test.sh [--no-clean]

NOTE: The test script will clean everything in the build directy before initializing CMake with the chosen or default build configuration, then build Debug and Release builds, and run tests for both.

The script must end with TEST PASSED, or it didn't pass.

To make sure everything works, also run the benchmarks:

scripts/benchmark.sh

Running Tests on Windows

In Visual Studio the test can be run as follows: first build the main project, the right click the RUN_TESTS target and chose build. See the output window for test results.

It is also possible to run tests from the command line after the project has been built:

cd build\MSVC
ctest

Note that the monster example is disabled for MSVC 2010.

Be aware that tests copy and generate certain files which are not automatically cleaned by Visual Studio. Close the solution and wipe the MSVC directory, and start over to get a guaranteed clean build.

Please also observe that the file .gitattributes is used to prevent certain files from getting CRLF line endings. Using another source control systems might break tests, notably test/flatc_compat/monsterdata_test.golden.

Note: Benchmarks have not been ported to Windows.

Configuration

The configuration

config/config.h

drives the permitted syntax and semantics of the schema compiler and code generator. These generally default to be compatible with Googles flatc compiler. It also sets things like permitted nesting depth of structs and tables.

The runtime library has a separate configuration file

include/flatcc/flatcc_rtconfig.h

This file can modify certain aspects of JSON parsing and printing such as disabling the Grisu3 library or requiring that all names in JSON are quoted.

For most users, it should not be relevant to modify these configuration settings. If changes are required, they can be given in the build system - it is not necessary to edit the config files, for example to disable trailing comma in the JSON parser:

cc -DFLATCC_JSON_PARSE_ALLOW_TRAILING_COMMA=0 ...

Using the Compiler and Builder library

The compiler library libflatcc.a can compile schemas provided in a memory buffer or as a filename. When given as a buffer, the schema cannot contain include statements - these will cause a compile error.

When given a filename the behavior is similar to the commandline flatcc interface, but with more options - see flatcc.h and config/config.h.

libflatcc.a supports functions named flatcc_.... reflection... may also be available which are simple the C generated interface for the binary schema. The builder library is also included. These last two interfaces are only present because the library supports binary schema generation.

The standalone runtime library libflatccrt.a is a collection of the src/runtime/*.c files. This supports the generated C headers for various features. It is also possible to distribute and compile with the source files directly. For debugging, it is useful to use the libflatccrt_d.a version because it catches a lot of incorrect API use in assertions.

The runtime library may also be used by other languages. See comments in flatcc_builder.h. JSON parsing is on example of an alternative use of the builder library so it may help to inspect the generated JSON parser source and runtime source.

FlatBuffers Binary Format

Mostly for implementers: FlatBuffers Binary Format

Security Considerations

See Security Considerations.

Benchmarks

See Benchmarks