flatcc
0.6.1 Bugfix and maintenance release
Ubuntu，Macos和Windows：Windows：每周：
JSON解析器可能会在以后的版本中更改解析联合向量的接口，该版本需要代码生成匹配库版本。
flatcc除构建和编译器工具以及C运行时库外没有外部依赖关系。 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 a few seconds, produce binaries between 15K and 60K, read small buffers in 30ns, build FlatBuffers in about 600ns, and with a larger executable also handle optional json parsing or printing in less than 2 us for a 10 field mixed type message.
该项目构建FlatCC，这是一个编译器，该编译器为C flatbuffer模式文件生成Flatbuffers代码。本简介还使用传统的怪物示例创建了一个单独的测试项目，此处在C版本中。
现在，假设一个类似UNIX的系统，尽管这不是一般要求 - 另请参见构建。您将需要git，cmake，bash，a C编译器以及忍者构建系统或制造。
git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
# scripts/initbuild.sh ninja
scripts/initbuild.sh make
scripts/setup.sh -a ../mymonster
ls bin
ls lib
cd ../mymonster
ls src
scripts/build.sh
ls generated
scripts/initbuild.sh是可选的，并且选择了构建后端，该后端默认为忍者。
设置脚本使用CMAKE构建FlatCC，然后使用Monster示例创建一个测试项目目录，而构建脚本只是一个小型的Shell脚本。标题和库象征性地链接到测试项目中。收集FlatCC后，您不需要CMAKE就可以构建自己的项目。
要创建另一个名为Foobar的测试项目，请调用scripts/setup.sh -s -x ../foobar 。这将避免从头开始重建FlatCC项目。
注意：请参阅ChangElog。由于发现了API不一致，因此偶尔会有较小的破裂变化。除非明确说明，否则破坏更改不会影响编译的运行时库，而只会影响标头文件。如果遇到麻烦，请确保flatcc工具与include/flatcc路径相同。
该项目包括：
flatcc FlatBuffers架构编译器，用于C和相应的库libflatcc.a 。编译器生成C标头文件或二进制flatbuffers架构。libflatccrt.a用于构建和验证C的Flatbuffers。生成的构建器标头取决于此库。它也可能对其他语言界面有用。该图书馆维护堆栈状态，以使从解析器或类似产品构建缓冲区变得易于构建。flatcc/portable标头库，以及用于所有编译器的小型助手，包括Endian处理和数字打印和解析。参见：
报告错误
Google FlatBuffers
建立说明
Quickstart
构建器接口参考
基准
flatcc编译器被实现为独立工具，而不是扩展Google flatc编译器，以便具有纯粹的Portable C库的实现该模式编译器，该架构编译器的实现旨在长期运行过程中的滥用输入而失败。还认为，C版本可能有助于提供模式解析，以比C ++更容易找到与C更容易接口的语言接口。 Googles FPL实验室的Flatbuffers团队在提供反馈和回答许多问题以帮助确保最佳兼容性方面非常有帮助。请注意，名称flatcc （FlatBuffers C编译器）与Google flatc 。
JSON格式与Google flatc工具兼容。 flatc工具使用架构和缓冲区作为输入将命令行的JSON转换。 flatcc生成架构特定代码以在运行时读取和编写JSON。尽管flatcc方法可能更快且更易于部署，但在使用JSON（例如编辑游戏场景）手动工作时， flatc方法可能更方便。这两个工具都有自己的位置。
注意：截至版本0.4.0，大型平台仅受支持。
它被认为是对Meson Build System的增加的支持，但是对此Via第56条的反馈会很好
如果可能的话，请提供一个简短的可重复模式和源文件，并带有主程序，返回1时返回1，成功的脚本为0和一个小的构建脚本。最好生成六个人并调用缓冲区验证器，以确保输入有效，并与debug库flatccrt_d链接。
另请参阅调试缓冲区，然后readfile.h可用于读取现有的缓冲区以进行验证。
例子：
样本/bugreport
折衷的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 %sn" , 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 版本0.6.2（在开发中）主要是错误修复版本，有关详细信息，请参阅ChangElog。已经确定了一个长期存在的错误，在对象在调用_create_as_root之前创建的对象将无法正确对齐，而现在也将缓冲区端填充到缓冲区内看到的最大对象。请注意，对于clang debug builds，添加了-fsanitize = undefined，这可能需要依赖的源代码，也需要使用该标志来避免缺少链接器符号。该功能可以在cmakelists.txt中禁用。
版本0.6.1主要包含错误修复和社区的大量贡献，以处理平台边缘案例。此外，Pendantic GCC警告是禁用的，而是依赖于Clang，因为GCC过于侵略性，因此断裂经常建立并不利于便携性。现有的C ++测试用例可确保C代码还可以与常见的C ++编译器一起使用，但是它可以破坏某些环境，因此现在有一个标志可以禁用该测试而无需禁用所有测试。已经添加了对FlatBuffer格式中可选标量值的支持。在各种平台上，还可以提高支持记忆分配的支持。 <table>_identifier已在生成的代码中对<table>_file_identifier偏爱，因为identifier很容易导致姓名冲突。现在，生成的代码中的file_extension常数没有前缀点（。）。
版本0.6.0引入了一个“主”属性，该属性与密钥属性一起使用，以选择用于查找和排序的默认键。如果不存在主，则具有最低ID的键将变为主要。现在可以在主密钥上递归对表和向量进行排序。打破：以前第一个列出的不是最低的ID是主要键。还引入了结构字段中的固定长度标量阵列（不支持结构和枚举元素）。结构支持固定长度数组字段，包括char数组。空结构从未完全工作，也不再支持，它们也不再受FlatC的支持。注意：Char数组当前不是Googles FlatC编译器的一部分-INT8阵列可以使用。打破：不再支持空结构 - 它们在Googles FlatC编译器中也无效。有关其他更改，请参见ChangElog。弃用： flatcc_accessors.h中的低级cast_to/from功能中删除，以删除read/write_from/to因为铸件接口断开某些罕见平台上的浮点转换。这不应影响正常使用，但在此版本中仍然有效。
发行0.5.3插入各种错误修复程序（请参见ChangElog）和一个破坏但可能的影响较低的影响：破坏：BREAKING：BREAKING：0.5.3构建器的行为创建呼叫，因此在使用ID属性时始终按字段ID订购参数，例如MyGame_Example_Monster_create() monster_test.fbs （＃81））。通过数字键字段对表进行排序时修复了未定义的行为。
版本0.5.2向阅读器方法引入可选的_get 。通过使用flatcc -g只有_get方法是有效的。这删除了某些字段名称的潜在名称。 0.5.2还引入了期待已久的克隆操作，用于表和向量。添加了一个C ++烟节，以减少数字void指针分配错误，但不断潜入。运行时库现在需要额外的文件refmap.c 。
版本0.5.1修复了JSON打印机中的缓冲区超支，并改进了便携式库<Stdalign.h>与C ++和嵌入式newlib标准库的兼容性。 JSON打印和解析已更加一致，可以帮助解析和打印表格以外的其他架构根，如Test_json.c中的测试驱动程序所示。 Monster_test.fbs文件已重新组织，以使Monster表与Googles FlatC版本更加一致，因此，较小的模式名称空间不一致得以解决。对便携式标头的明确引用已移出生成的来源。外部“ C” C ++后卫在生成的标头周围增加了。 0.5.1还清理了低级联合接口，因此{type，value}的术语在{type，embly}和{type，embers}上始终如一地使用。
be 。一旦项目变得稳定，就没有计划进行频繁更新，但是社区的输入将始终受到欢迎，并包含在相关的发行版中，尤其是在不同目标平台上测试。
此列表有些过时，添加了最新的编译器版本，当CI平台不再支持时，删除了一些旧版本，但很大程度上受支持的目标保持不变。 MSVC 2010将来可能会被弃用。
CI-More分支测试其他编译器：
C11/C ++ 11是期望始终工作的参考。
在GCC-8+的情况下，不支持GCC --pedantic编译器选项，因为它会迫使不可携带的代码更改，并且由于每个新的GCC发布都会破坏代码库。
MSVC 2017并不总是经过测试，因为CI环境将不支持MSVC 2010。
C ++编译器的较旧/非标准版本会引起问题，因为static_assert和alignas以奇怪的方式行事，在这种方式中，它们既不存在也不如预期的那样完全工作。通常会有解决方法，但是使用-std=c++11或-std=c++14更可靠。
该库中的库不支持GCC C ++ PRE 4.7，因为便携式库在stdalign.h和stdint.h中的C ++限制不起作用。这可以解决，但不是优先事项。
随着CI环境的更新，可能已经退休了一些先前的Testet编译器版本。有关当前配置，请参见ci-more分支中的.travis.yml和appveyor.yml 。
怪物样本与MSVC 2010不使用，因为它有意使用C99样式代码来更好地遵循C ++版本。
构建选项FLATCC_TEST可用于禁用所有可能使FlatCC编译在原本有问题的平台上的测试。可以专门用于C ++测试（包括生成C代码的简单的C ++文件），可以禁用大量选项FLATCC_CXX_TEST 。
没有理由不能支持其他或较旧的编译器，但是它可能需要在构建配置中进行一些工作，并且可能需要更新到便携式库。以上只是已测试和配置的内容。
便携性层具有一些对于诸如endian处理之类的功能，而其他功能则具有可选和缺失C11功能的兼容性。这应该支持大多数C编译器，但依靠社区反馈来成熟。
可以通过使用-STD = C11并避免使用JSON（需要大量的数字解析支持）来大大减少运行时的必要大小，并删除include/flatcc/reflection ，它可以支持二进制架构文件，并且可以从reflection/reflection.fbs include/flatcc/support sworts and flatcc/spects and and and spects and Same和Same and Same和Same and Same and Same。确切的必需文件可能会随着发布而变化，并且在编译的代码大小方面并不重要。
优先级是设计一个易于使用的C构建器界面，该界面相当快，适用于服务器和嵌入式设备，但是具有绝对性能的可用性 - 仍然以每秒毫秒为单位测量了小缓冲速度，并从粗略的估计中读取10-100 millon缓冲液。读取Flatbuffer的范围不仅仅是构建它们的数量级。
对于具有1000个怪物，动态扩展的怪物名称，怪物向量和库存向量的100MB缓冲区，带宽在2.2GHz Haswell Core core i7 cpu上达到约2.2GB/s和45ms/buffer。这包括读取并验证所有数据。仅读取几个关键字段将带宽提高到2.7GB/s和37ms/op。对于10MB缓冲区，带宽可能更高，但最终将被呼叫开销击中较小的缓冲区，因此我们以约150ns/op编码小型缓冲区的速度下降到300MB/s。这些数字只是一个粗略的指南 - 它们显然取决于硬件，编译器和编码数据。测量值不包括初始热身步骤。
生成的JSON解析器的速度比直接在C或C ++中构建FlatBuffer的速度约4倍，或者大约是2200NS vs 600NS，用于700字节JSON消息。因此，JSON解析比在Google Flatbuffers基准页面上报道的等效协议缓冲区大约要快两个数量级。 LZ4压缩将估计JSON解析的总体处理时间两倍。 JSON打印比解析更快，但不是很明显。 JSON压缩到大约一半的大型缓冲液在压缩的扁平按钮上的大小，但在小型缓冲液上压缩更糟（在根本不压缩时更不用说）。
应当指出的是，FlatBuffer阅读性能排除了JSON解析和协议缓冲区本质上包含的验证。验证尚未得到基准测试，但可能会增加少于50％的读取开销，除非仅读取大型缓冲液的一小部分。
另请参阅基准。
客户端C代码几乎可以避免任何形式的分配来构建缓冲区，因为建筑商堆栈提供了可扩展的竞技场，然后再进行对象 - 例如，附加字符串或向量零碎。当可以直接构造一个完整的对象时，例如在小Endian平台上的整数数组中的向量，大部分都会绕过堆栈。
读者界面应该非常快，而明智的改进性能的空间也较小。它也比构建器简单得多。
可用性还优先于最小的生成源代码和编译时间。它不应影响汇编的大小。
除了最限制的微控制器以外，所有的二进制输出应相当小。一个33K怪物源测试文件（除了生成的标头和构建器库）导致不到50k优化的二进制可执行文件，包括printf语句和其他支持逻辑的开销，或不包括构建器库的30k对象文件。
仅读取的二进制文件较小，但不一定比建筑商考虑的工作要少得多：兼容性测试读取预先生成的二进制monsterdata_test.golden Monster文件，并验证所有内容是否如预期的。这将产生13K优化的二进制可执行文件或6K对象文件。此检查的来源是5K不包括标头文件。读者不需要与库链接。
JSON解析器与纯flatbuffer使用量相比，将编译的二进制文件膨胀，因为它们会嵌入解析器决策树。 Monster.fbs的JSON解析器可以将100K +/-优化设置添加到可执行的二进制中。
生成的用于构建FlatBuffer的代码以及用于解析和打印FlatBuffers的代码，都需要访问include/flatcc 。读者不依赖任何库，但是所有其他生成的文件都依赖于libflatccrt.a运行时库。请注意，仅当需要flatcc编译器本身作为库时，才需要libflatcc.a 。
读者和建筑商依靠生成的普通读取器和构建器标题文件。这些常见的文件使更改全局名称空间并重新定义基本类型（ uoffset_t等）成为可能。将来，这可能会进入库代码并将宏用于这些抽象，并最终为超出标准32位未签名偏移量（ uoffset_t ）的类型提供了一组预定义的文件。运行时库是特定于一组类型定义的。
有关详细的使用指南，请参阅Monster_test.c和生成的文件。该项目中使用的怪物架构对原件进行了略微适应，以测试一些其他边缘案例。
对于构建FlatBuffer，每个模式都会生成单独的构建器标头文件。它需要一个编译器和小型运行时库libflatccrt.a生成的flatbuffers_common_builder.h文件。正是由于这种要求，读者和构建器生成的代码被分开保留。典型用途可以在monster_test.c文件中看到。构建器允许在更新包含表的同时重复将内容推向向量或字符串，从而简化了外部格式的解析。也可以在线构建嵌套缓冲液 - 起初这听起来可能会过大，但是在将缓冲区的结合在网络接口中并确保所有缓冲级别的对齐方式时很有用。
为了验证flatbuffer，生成myschema_verifier.h 。这取决于运行时库和读取器标题。
JSON解析器和打印机每个架构文件生成一个文件，其中包括模式将有其自己的解析器和打印机，其中包括解析器和打印机将依靠，与建筑商的工作方式相似。
低级别的注意：构建器在缓冲区的末端生成所有VTABLE，而不是在每个桌子前的临时临时生成，但否则可以对VTABLES进行相同的重复数据删除。这使得可以将VTABLES聚集在热缓存中，或确保在部分传输缓冲区时可用所有VTABLES。可以通过运行时标志禁用此行为。
由于某些用例可能包含非常受限的嵌入式设备，因此可以使用分配器对象和缓冲区发射极对象自定义构建器库。单独的发射极可确保可以构建缓冲区，而无需单独使用完整的缓冲区，如果需要的话。
在FlatCC_Builder.H和Flatcc_emitter.h中记录了无类型的构建器库，而生成的键入构建器API则记录在构建器接口参考中。
有时，人们对生成代码中使用的宏的密集性提出了一个问题。这些宏使很难理解哪些功能实际上可用。构建器接口参考试图以一般方式记录操作。为了获取更详细的信息，可以使用scripts/flatcc-doc.sh脚本提取生成的功能原型。
有些人还关注宏是“不安全”。与FlatCC一起使用时，宏不是不安全的，因为它们会生成静态或静态内联函数。如果错误地使用与直接C代码相同的扩展，这些将触发编译时间错误。
扩展将生成的输出压缩超过因子10，以确保源控制下的代码不会爆炸，并可以以有意义的方式比较生成的代码版本，并查看它是否与预期的架构匹配。宏对于通过便携式标头处理平台抽象也很重要。
尽管如此，尽管不直接由构建系统支持，但仍可以看到生成的输出。例如， include/flatcc/reflection包含反射架构的预生成的标头文件。要使用clang编译器工具链查看扩展的输出，请运行：
clang -E -DNDEBUG -I include
include/flatcc/reflection/reflection_reader.h |
clang-format
在不支持Clang的平台上可能还可以使用其他类似的命令。
请注意，编译器将优化几乎所有生成的代码，并且仅使用最终用户代码实际引用的逻辑，因为功能是静态的或静态的。其余部分通常会在应用程序代码中有效内联，从而导致相当小的二进制代码大小。
更多细节可以在＃88中找到
生成代码的扩展可用于获取特定对象类型的文档。
以下脚本自动化了此过程：
scripts/flatcc-doc.sh <schema-file> <name-prefix> [<outdir>]
写入函数原型<outdir>/<name-prefix>.doc 。
请注意，该脚本需要Clang编译器和Clang-Format工具，但是该脚本也可能适用于其他工具链。
可以使用反射模式作为一个例子来说明脚本背后的原理，其中提取了对象表的文档：
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'
Clang-Format的WebKit样式确保参数和返回类型都放在同一行上。 GREP提取功能标头，并切割带有从同一行开始的函数机构。 SED Strips __tmp后缀来自用于避免宏名称冲突的参数名称。格雷普条);要删除冗余的正向声明，然后添加；使每行成为有效的C原型。
不能保证以上内容始终可以随着输出而变化，但应该会有很长的路要走。
输出的一小部分，如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);
...
使用反射和怪物模式在以下脚本中提供了示例：
scripts/reflection-doc-example.sh
scripts/monster-doc-example.sh
Monster Doc示例本质上调用：
scripts/flatcc-doc.sh samples/monster/monster.fbs MyGame_Sample_Monster_
导致文件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);
...
也可以提取FlatBuffer本地类型，例如字符串操作：
scripts/flatcc-doc.sh samples/monster/monster.fbs flatbuffers_string_
导致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);
...
有关详细信息，请参阅flatcc -h 。
此处列出的在线版本：Flatcc -help.md，但请使用flatcc -h进行最新参考。
编译器可以生成一个单个标头文件，也可以为所有包含的架构和一个通用文件以及不支持阅读（默认）和写入（-W）FlatBuffers的标题。最简单的选项是对所有人使用（-a），并包括myschema_builder.h文件。
（-a）或（-v）还会生成一个验证者文件。
确保在c编译器中可见包含include夹下的flatcc在编译FlatBuffer构建器时包括路径。
flatcc （-i）Include路径将假设所有具有相同基本名称（案例insentive）的架构文件是相同的，并且仅包括第一个。所有生成的文件均使用输入Basename，并将降落在工作目录或（-o）设置的路径中。
可以使用（ - stdout）生成文件以生成stdout。 C标题将被订购和连接，但与单独的文件输出相同。每个包含的语句都受到保护，因此这不会导致丢失的包括文件。
生成的代码，尤其是所有与-stdout结合在一起的代码可能会显示大，但是实际上实际使用的零件才能在最终的可执行文件或对象文件中占用空间。现代编译器内联，仅包含静态链接的构建器库的必要部分。
如果仅需要一个，则可以使用-json flag或-json-printer或json-parser生成JSON打印机和解析器。有一些运行时库编译时标志可以优化打印符号枚举，但是在运行时也可以禁用这些枚举。
确保与libflatccrt （用于运行时的RT）而不是libflatcc （架构编译器）链接，否则构建器将不可用。另外，请确保将flatcc项目root的“包含”在包含路径中。
默认情况下，FlatCC将在阅读或验证缓冲区时期望缓冲区中的file_identifier 。
缓冲区可以在偏移4时具有意外的4字节标识符，或者可能不存在标识符。
并非所有语言界面都支持缓冲区中的文件标识符，如果这样做，它们可能不会在较旧的版本中使用。用户报告了Python和LUA接口的问题，但这很容易解决。
检查验证者的返回值：
int ret;
char *s;
ret = MyTable_verify_as_root(buf, size);
if (ret) {
s = flatcc_verify_error_string(ret);
printf("buffer failed: %sn", s);
}
要验证没有标识符的缓冲区，或者忽略了其他标识符，请使用null标识符使用验证程序的_with_identifier版本：
char *identifier = 0;
MyTable_verify_as_root_with_identifier(buf, size, identifier);
阅读缓冲区使用：
MyTable_as_root_with_identifier(buf, 0);
并在没有标识符的情况下构建缓冲区：
MyTable_start_as_root_with_identifier(builder, 0);
...
MyTable_end_as_root_with_identifier(builder, 0);
其他几个as_root调用具有as_root_with_identifier版本，包括JSON打印。
构建flatcc tool后，二进制文件位于flatcc源树下方的bin和lib目录中。
您可以直接跳到遵循Google flatbuffers教程的怪物示例，也可以沿着下面的Quickstart指南阅读。如果您遵循Monster教程，则可能需要克隆并构建FlatCC并将源复制到单独的项目目录，如下所示：
git clone https://github.com/dvidelabs/flatcc.git
flatcc/scripts/setup.sh -a mymonster
cd mymonster
scripts/build.sh
build/mymonster
scripts/setup.sh将最小链接库和工具链接到mymonster自定义目录。使用（-a）还添加了一个简单的scripts/setup.sh -h脚本，复制示例和更新.gitignore设置也可以构建FlatCC，但是您仍然必须确保为系统配置构建环境。
要编写您自己的模式文件，请按照编写模式文件的“ flatbuffers”项目文档。
在研究怪物样本和下面的快速启动后，构建器界面参考可能很有用。
在寻找高级示例时，例如对向量进行排序和通过密钥查找元素，您应该在test/monster_test项目中找到这些元素。
以下快速入门指南是对test/monster_test项目的广泛简化 - 请注意，该模式与教程略有不同。专注于C特定框架，而不是一般的FlatBuffers概念。
您仍然可以使用设置工具来创建一个空项目并遵循，但是在下面的文本中没有任何假设。
在这里，我们提供了一个仅阅读访问monster flatbuffer的快速示例 - 它是monster_test.c文件的改编提取物。
首先，我们使用Common（-C）支持标头编译架构，并添加递归，因为monster_test.fbs包含其他文件。
flatcc -cr --reader test/monster_test/monster_test.fbs
为简单起见，我们假设您在项目根文件夹中构建示例项目，但是在实践中，您需要更改一些路径，例如：
mkdir -p build/example
flatcc -cr --reader -o build/example test/monster_test/monster_test.fbs
cd build/example
我们得到：
flatbuffers_common_reader.h
include_test1_reader.h
include_test2_reader.h
monster_test_reader.h
（还有更简单的samples/monster/monster.fbs ，但是您将不会收到架构文件）。
名称空间可以很长，因此我们可以选择使用宏来管理它。
#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 availablen");
return -1;
}
if (ns(Monster_hp(monster)) != 80) {
printf("Health points are not as expectedn");
return -1;
}
if (!(vec = ns(Monster_pos(monster)))) {
printf("Position is absentn");
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 coordinaten");
return -1;
}
/* Verify force_align relative to buffer start. */
offset = (char *)vec - (char *)buffer;
if (offset & 15) {
printf("Force align of Vec3 struct not correctn");
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 memoryn");
}
/* ... */
return 0;
}
/* main() {...} */
假设我们上面的文件是monster_example.c c以下是编译项目仅阅读的几种方法 - 稍后显示使用运行时库的编译。
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
包括路径或源路径可能不同。始终使用include/flatcc/portable中的某些文件，但是-D FLATCC_PORTABLE标志包含其他文件，以支持缺少C11功能的编译器。
注意：在某些clang/GCC平台上，如果链接器无法找到posix_memalign ，则可能有必要使用-std = gnu99或-std = gnu11，另请参见paligned_alloc.h中的注释。
在这里，我们提供了一个非常有限的示例，说明如何构建缓冲区 - 只有几个字段已更新。请参阅Monster_test.c和DOC目录以获取更多信息。
首先，我们必须生成文件：
flatcc -a monster_test.fbs
这会产生：
flatbuffers_common_reader.h
flatbuffers_common_builder.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
注意：除非我们只打算读取缓冲区，否则我们实际上不会进行前面显示的可读生成 - 建筑商的生成也总是会生成读取acpes。
通过包括"monster_test_builder.h"所有其他文件都会自动包含。 C编译器需要-I include指令以访问flatcc/flatcc_builder.h ， flatcc/flatcc_verifier.h和其他文件，并根据细节的不同文件，假设项目根是当前目录。
不需要验证符，只是因为我们懒惰地选择了-a选项，因此不需要验证器。
必须先初始化构建器才能设置我们有效构建缓冲区所需的运行时环境 - 构建器依靠发射器对象来构建实际的缓冲区 - 在这里，我们隐含地使用默认值。一旦有了这一点，我们就可以将构建器视为手柄，并专注于生成的API，直到最终确定缓冲区（即访问结果）。对于非平凡用途，建议使用自定义的发射器，例如，在它们完成后立即在网络上发布页面，而不是使用flatcc_builder_finalize_buffer将所有页面合并到单个缓冲区中，或者简单的flatcc_builder_get_direct_buffer如果返回null（如果buffer均太大）。另请参见flatcc_builder.h和flatcc_emitter.h中的文档注释。另请参见builder.h中的flatc_builder_finalize_aligned_buffer ，当malloc对齐的缓冲区不足时，构建器接口参考。
#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;
}
编译示例项目：
cc -std=c11 -I include monster_example.c lib/libflatccrt.a -o monster_example
请注意，构建缓冲区需要运行时库，但不需要阅读它们。 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.
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_with_identifier(buffer, size,
"MONS")))) {
printf("Monster buffer is invalid: %sn",
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.
NOTE: as of August 2024 it has been discovered that C++ writer code has been aligning empty vectors to the size field only, even if elements require greater alignment like the double type which requires 8. This would cause the FlatCC verifier to (correctly) reject these vectors because it would result in an invalid C pointer type on some architectures. However, because this has been in effect for over 10 years, the consensus is to have verifiers tolerate this behaviour even if C++ will eventually fix this issue. The FlatCC verifier has been updated to accept such buffers by default with an optional compile time flag to enforce the strict behaviour as well ( FLATCC_ENFORCE_ALIGNED_EMPTY_VECTORS ). In principle the misaligned vectors can potentially lead to undefined behaviour in agressively optimized C compilers. As of now it appears to be safe to read such buffers on common platforms and it is preferable to avoid additional runtime reader overhead to deal with this. For more, see FlatCC #287, Google Flatbuffers #8374, FlatCC #289.
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.
Notably several users have experienced conflicts with a table or struct field named 'identifier' because <table-name>_identifier has been defined to be the file identifier to be used when creating a buffer with that table (or struct) as root. As of 0.6.1, the name is <table-name>_file_identifier to reduce the risk of conflicts. The old form is deprecated but still generated for tables without a field named 'identifier' for backwards compatibility. Mostly this macro is used for higher level functions such as mytable_create_as_root which need to know what identifier to use.
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.
As of April 2022, Googles flatc tool has implemented an --annotate feature. This provides an annotated hex dump given a binary buffer and a schema. The output can be used to troubleshoot and rule out or confirm suspected encoding bugs in the buffer at hand. The eclectic example in the FlatBuffers Binary Format document contains a hand written annotated example which inspired the --annotate feature, but it is not the exact same output format. Note also that flatc generated buffers tend to have vtables before the table it is referenced by, while flatcc normally packs all vtables at the end of the buffer for better padding and cache efficiency.
See also flatc --annotate.
Note: There is experimental support for text editor that supports clangd language server or similar. You can edit CMakeList.txt to generate build/Debug/compile_comands.json , at least when using clang as a compiler, and copy or symlink it from root. Or come with a better suggestion. There are .gitignore entries for compile_flags.txt and compile_commands.json in project root.
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.
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_file_identifier
#define MyGame_Example_Monster_file_identifier "MONS"
#endif
The user can now override the identifier for a given type, for example:
#define MyGame_Example_Vec3_file_identifier "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_file_extension
#define MyGame_Example_Monster_file_extension "mon"
#endif
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 "x81xf4x0ex33"
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 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 length 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, ie 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 absence 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]
}
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.
Fixed length arrays introduced in 0.6.0 allow for structs containing arrays of fixed length scalars, structs and chars. Arrays are parsed like vectors for of similar type but are zero padded if shorter than expected and fails if longer than expected. The flag reject_array_underflow will error if an array is shorter than expected instead of zero padding. The flag skip_array_overflow will allow overlong arrays and simply drop extra elements.
Char arrays are parsed like strings and zero padded if short than expected, but they are not zero terminated. A string like "hello" will exactly fit into a field of type [char:5] . Trailing zero characters are not printed, but embedded zero characters are. This allows for loss-less roundtrips without having to zero pad strings. Note that other arrays are always printed in full. If the flag skip_array_overflow is set, a string might be truncated in the middle of a multi-byte character. This is not checked nor enforced by the verifier.
Both the printer and the parser have the ability to accept runtime flags that modifies their behavior. Please refer to header file comments for documentation and test cases for examples. Notably it is possible to print unquoted symbols and to ignore unknown fields when parsing instead of generating an error.
Note that deprecated fields are considered unknown fields during parsing so in order to process JSON from an old schema version with deprecated fields present, unknown symbols must be skipped.
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.
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.
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.
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.
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.
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.
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, eg 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.
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 .
Fixed Length Arrays is a late feature to the FlatBuffers format introduced in flatc and flatcc mid 2019. Currently only scalars arrays are supported, and only as struct fields. To use fixed length arrays as a table field wrap it in a struct first. It would make sense to support struct elements and enum elements, but that has not been implemented. Char arrays are more controversial due to verification and zero termination and are also not supported. Arrays are aligned to the size of the first field and are equivalent to repeating elements within the struct.
The schema syntax is:
struct MyStruct {
my_array : [float:10];
}
See test_fixed_array in monster_test.c for an example of how to work with these arrays.
Flatcc opts to allow arbitrary length fixed length arrays but limit the entire struct to 2^16-1 bytes. Tables cannot hold larger structs, and the C language does not guarantee support for larger structs. Other implementations might have different limits on maximum array size. Arrays of 0 length are not permitted.
Optional scalar table fields were introduced to FlatBuffers mid 2020 in order to better handle null values also for scalar data types, as is common in SQL databases. Before describing optional values, first understand how ordinary scalar values work in FlatBuffers:
Imagine a FlatBuffer table with a mana field from the monster sample schema. Ordinarily a scalar table field has implicit default value of 0 like mana : uint8; , or an explicit default value specified in the schema like mana : uint8 = 100; 。 When a value is absent from a table field, the default value is returned, and when a value is added during buffer construction, it will not actually be stored if the value matches the default value, unless the force_add option is used to write a value even if it matches the default value. Likewise the is_present method can be used to test if a field was actually stored in the buffer when reading it.
When a table has many fields, most of which just hold default settings, signficant space can be saved using default values, but it also means that an absent value does not indicate null. Field absence is essentially just a data compression technique, not a semantic change to the data. However, it is possible to use force_add and is_present to interpret values as null when not present, except that this is not a standardized technique. Optional fields represents a standardized way to achieve this.
Scalar fields can be marked as optional by assigning null as a default value. For example, some objects might not have a meaningful mana value, so it could be represented as lifeforce : uint8 = null . Now the lifeforce field has become an optional field. In the FlatCC implementation this means that the field is written, it will always be written also if the value is 0 or any other representable value. It also means that the force_add method is not available for the field because force_add is essentially always in effect for the field. On the read side, optional scalar fields behave exactly is ordinary scalar fields that have not specified a default value, that is, if the field is absent, 0 will be returned and is_present will return false. Instead optional scalar fields get a new accessor method with the suffix _option() which returns a struct with two fiels: { is_null, value } where _option().is_null == !is_present() and _option().value is the same value is the _get() method, which will be 0 if is_null is true. The option struct is named after the type similar to unions, for example flatbuffers_uint8_option_t or MyGame_Example_Color_option_t , and the option accessor method also works similar to unions. Note that _get() will also return 0 for optional enum values that are null (ie absent), even if the enum value does not have an enumerated element with the value 0. Normally enums without a 0 element is not allowed in the schema unless a default value is specified, but in this case it is null, and _get() needs some value to return in this case.
By keeping the original accessors, read logic can be made simpler and faster when it is not important whether a value is null or 0 and at the same time the option value can be returned and stored.
Note that struct fields cannot be optional. Also note that, non-scalar table fields are not declared optional because these types can already represent null via a null pointer or a NONE union type.
JSON parsing and printing change behavior for scalar fields by treating absent fields differently according the optional semantics. For example parsing a missing field will not store a default value even if the parser is executed with a flag to force default values to be stored and the printer will not print absent optional fields even if otherwise flagged to print default values. Currenlty the JSON printers and parsers do not print or parse JSON null and can only represent null be absence of a field.
For an example of reading and writing, as well as printing and parsing JSON, optional scalar fields, please refer to optional_scalars_test.fbs and optional_scalars_test.c.
The 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 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.
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.
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 v0.6.0 the default key used for find and and sort without the by_name suffix is the field with the smaller id instead of the first listed in the schema which is often but not always the same thing.
v0.6.0 also introduces the primary_key attribute that can be used instead of the key attribute on at most one field. The two attributes are mutually exclusive. This can be used if a key field with a higher id should be the default key. There is no difference when only one field has a key or primary_key attribute, so in that case choose key for compatiblity. Googles flatc compiler does not recognize the primary_key attribute.
As of v0.6.0 a 'sorted' attribute has been introduced together with the sort operations <table_name>_sort and <union_name>_sort . If a table or a union, directly or indirectly, contains a vector with the 'sorted' attribute, then the sort operation is made available. The sort will recursively visit all children with vectors marked sorted. The sort operatoin will use the default (primary) key. A table or union must first be cast to mutable, for example ns(Monster_sort((ns(Monster_mutable_table_t))monster) . The actual vector sort operations are the same as before, they are just called automatically. The sorted attribute can only be set on vectors that are not unions. The vector can be of scalar, string, struct, or table type. sorted is only valid for a struct or table vector if the struct or table has a field with a key or primary_key attribute. NOTE: A FlatBuffer can reference the same object multiple times. The sort operation will be repeated if this is the case. Sometimes that is OK, but if it is a concern, remove the sorted attribute and sort the vector manually. Note that sharing can also happen via a shared containing object. The sort operations are generated in _reader.h files and only for objects directly or indirectly affected by the sorted attribute. Unions have a new mutable case operator for use with sorting unions: ns(Any_sort(ns(Any_mutable_cast)(my_any_union)) . Usually unions will be sorted via a containing table which performs this cast automatically. See also test_recursive_sort in monster_test.c.
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.
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 .
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.
Note: if a test fails, see Strict Aliasing for a possible resolution.
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.
To enforce a 32-bit build on a 64-bit machine the following configuration can be used:
scripts/initbuild.sh make-32bit
which uses make and provides the -m32 flag to the compiler. 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
Install CMake, MSVC, and git (tested with MSVC 14 2015).
In PowerShell:
git clone https://github.com/dvidelabs/flatcc.git
cd flatcc
mkdir buildMSVC
cd buildMSVC
cmake -G "Visual Studio 14 2015" ....
Optionally also build from the command line (in buildMSVC):
cmake --build . --target --config Debug
cmake --build . --target --config Release
In Visual Studio:
open flatccbuildMSVCFlatCC.sln
build solution
choose Release build configuration menu
rebuild solution
Note that flatccCMakeList.txt sets the -DFLATCC_PORTABLE flag and that includeflatccportablepwarnings.h disable certain warnings for warning level -W3.
Docker图像：
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.
2024-03-08: WARNING: -O2 -mcpu=cortex-m7 targets using the arm-none-eabi 13.2.Rel1 toolchain can result in uninitialized stack access when not compiled with -fno-strict-aliasing -mcpu=cortex-m0 and -mcpu=cortex-m1 appears to be unaffected. See also issue #274. 2024-10-03: Fix available on flatcc master branch when you read this. See also CHANGELOG comments for release 0.6.2.
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.
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.
On systems where the default POSIX assert call is unavailable, or when a different assert behaviour is desirable, it is possible to override the default behaviour in runtime part of flatcc library via logic defined in flatcc_assert.h.
By default Posix assert is beeing used. It can be changed by preprocessor definition:
-DFLATCC_ASSERT=own_assert
but it will not override assertions used in the portable library, notably the Grisu3 fast numerical conversion library used with JSON parsing.
Runtime assertions can be disabled using:
-DFLATCC_NO_ASSERT
This will also disable Grisu3 assertions. See flatcc_assert.h for details.
The <assert.h> file will in all cases remain a dependency for C11 style static assertions. Static assertions are needed to ensure the generated structs have the correct physical layout on all compilers. The portable library has a generic static assert implementation for older compilers.
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.
The Flatcc build files should take care of strict aliasing issues on common platforms, but it is not a solved problem, so here is some background information.
In most cases this is a non-issue with the current flatcc code base, but that does not help in the cases where it is an issue.
Compilers have become increasingly aggressive with applying, and defaulting to, strict aliasing rules.
FlatCC does not guarantee that strict aliasing rules are followed, but the code base is updated as issues are detected. If a test fails or segfaults the first thing to check is -fno-strict-aliasing , or the platform equivalent, or to disable pointer casts, as discussed below.
Strict aliasing means that a cast like p2 = *(T *)p1 is not valid because the compiler thinks that p2 does not depend on data pointed to by p1. In most cases compilers are sensible enough to handle this, but not always. It can, and will, lead to reading from uninitialized memory or segfaults. There are two ways around this, one is to use unions to convert from integer to float, which is valid in C, but not in C++, and the other is to use memcpy for small constant sizes, which is guaranteed safe, but can be slow if not optimized, and it is not always optimized. (Not strictly memcpy but access via cast to char * or other "narrow" type).
FlatCC manages this in flatcc_accessors.h which forwards to platform dependent code in pmemaccess.h. Note that is applies to the runtime code base only. For compile time the only issue should be hash tables and these should also be safe.
FlatCC either uses optimized memcpy or non-compliant pointer casts depending on the platform. Essentially, buffer memory is first copied, or pointer cast, into an unsigned integer of a given size. This integer is then endian converted into another unsigned integer. Then that integer is converted into a final integer type or floating point type using union casts. This generally optimizes out to very few assembly instructions, but when it does not, code size and execution time can grow significantly.
It has been observed that targets both default to strict aliasing with -O2 optimization, and at the same to uses a function call for memcpy(dest, src, sizeof(uint32_t)) , but where __builtin_memcpy does optimize well, hence requiring detection of a fast memcpy operation.
This is a game between being reasonably performant and compliant.
-DPORTABLE_MEM_PTR_ACCESS=0 will force the runtime code to not use pointer casts but it can potentially generate suboptimal code and can be set 1 if the compiler and build configuration is known to not have issues with strict aliasing. It is set to 1 for most x86/64 targets since this has been working for a long time in FlatCC builds and tests, while memcpy might not work efficient.
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.
By default lib files a built into the lib subdirectory of the project. This can be changed, for example like -DFLATCC_INSTALL_LIB=lib64 .
To distribute the compiled binaries the following files are required:
编译器：
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)
运行时：
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
The build products from MSVC are placed in the bin and lib subdirectories:
flatccbinDebugflatcc.exe
flatcclibDebugflatcc_d.lib
flatcclibDebugflatccrt_d.lib
flatccbinReleaseflatcc.exe
flatcclibReleaseflatcc.lib
flatcclibReleaseflatccrt.lib
Runtime includeflatcc directory is distributed like other platforms.
跑步
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
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 buildMSVC
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.
配置
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 ...
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.
Mostly for implementers: FlatBuffers Binary Format
See Security Considerations.
FlatCC coding style is largely similar to the WebKit Style, with the following notable exceptions:
<stdint.h> types are made available.if (err) return -1; 。0 ./* A comment. */true and false keywords are not used (pre C99).snake_case is used over camelCase .#pragma once because it is non-standard and not always reliable in filesystems with ambigious paths.config.h inclusion might be handled differently in that flatbuffers.h includes the config file.unsigned is not used without int for historical reasons. Generally a type like uint32_t is preferred.TODO: instead of FIXME: in comments for historical reasons.All the main source code in compiler and runtime aim to be C11 compatible and uses many C11 constructs. This is made possible through the included portable library such that older compilers can also function. Therefore any platform specific adaptations will be provided by updating the portable library rather than introducing compile time flags in the main source code.
See Benchmarks
