https://github.com/tensorflow/tflite-micro makes use of flatbuffers with
a variety of DSP toolchains.
Without the change from this PR, we can get a double-promotion warning
with some of these DSP toolchains:
```
flatbuffers/include/flatbuffers/util.h:104:11: error: implicit conversion increases floating-point precision: 'std::numeric_limits<float>::_Ty' (aka 'float') to 'double' [-Werror,-Wdouble-promotion]
T eps = std::numeric_limits<float>::epsilon();
~~~ ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
iomanip isn't available for our DSP. Luckily, we compile with FLATBUFFERS_PREFER_PRINTF, so moved the #include directive there.
ctype.h has to be included explicilty for tolower() and toupper().
Parsing as bytes produces buffers that are unsafe to access unless passed thru a verifier,
whereas users could reasonably assume that any JSON parsed without errors is safe to access.
Users that still have legacy JSON files with such bytes in it will get a helpful error point them
to the option to turn on to have it work again.
This change checks if the current source directory is a git repository.
If this is not checked the git command picks the commit hash from
parent directory. e.g. when tarball is extracted in a packaging repository.
* Enable --gen-onefile in Python
Made it possible to generate all python code in one file. Modified
py_test.py so that it can switch between the multi-file code and
the one-file code.
Updated PythonTest.sh and py_test.py so that the multi-file code
and the one-file code can be tested based on the same test code.
* Sync with google/flatbuffers
* Add --gen-onefile to generate_code.py
* Make idl_parser deterministic
Some golden tests even exercise this [logic](df2df21ec1/tests/prototest/test.golden (L8)). Let's make the parser fully stable not depending on the implementation of std::sort
* Retry the ci
* Fix C/C++ Create<Type>Direct with sorted vectors
If a struct has a key the vector has to be sorted. To sort the vector
you can't use "const".
* Changes due to code review
* Improve code readability
* Add generate of JSON schema to string to lib
* option indent_step is supported
* Remove unused variables
* Fix break in test
* Fix style to be consistent with rest of the code
* [TS] Fix reserved words as arguments (#6955)
* initial hack to get new Lua generator into flatc
* Starting to output enum defs for Lua
* Continue to work on table generation for Lua
* Finished basic getter access for Lua
* Added ability to get object by index
* Finished struct builder
* aliased reflection to r
* finish table builder generation
* register requiring files
* better generated header info
* Tying up loose ends
* Updated reflection to handle struct padding
* Addd type sizes to reflection
* Fixed some vector indirect issues
* Lua tests passed
* Misc cleanup
* ci fixes 1
* ci fixes 2
* renaming
* up size of type sizes
* manually ran clang-format-11 -i src/idl_parser.cpp
* fixed some windows casting
* remove stupid auto import
* more static_casting
* remove std
* update other build environments
* remove scoped enums
* replaced std::to_string with NumToString
* more win fixes
* more win fixes
* replaced old lua with new
* removed auto import
* review responses
* more style fixes
* refactor bfbs_gen_len to use code +=
* added consts
* fix lambda capture for windows
* remove unused return type
When generating code with --grpc, --cpp and using filename-suffix, the generated grpc files where not including the correct header that had the filename-suffix. As a suffix, they used the default "_generated".
Free functions for these were used to get the suffix. FlatBufFile had such methods, but also needed to be into its base File and use these.
- grpc generated files include the correct message header.
- grpc generated files also have the suffix
- grpc generated cc file does not include initial message header
* removed test/generate_code.{sh|bat}
remove c++0x from generate_code.py
added check generate code script in python
add windows specific call
added flags to generate_code.py
Set c++-0x on BUILD_LEGACY
Skip generating monster_extra if requested
* added option to skip monster extra
* add conditional to skip 2010 check gen
Restore flatbuffers::FlatBufferBuilder::kFileIdentifierLength, which was a documented part of the public API, but the identifier was lost during the refactoring effected by comment commit 6c8c291559.
* Added Google benchmarks (and gtests)
* Separate benchmark CMakeLists.txt to its own file
* Move output directory to target just flatbenchmark
* Reduced from encoding 210ns -> 188ns
* store size_ as uoffset_t
* fixed windows c4267 warning
Detected instability when built `flatbuffers-2.0.0` on `gcc-12`:
[ 75%] Building CXX object CMakeFiles/flattests.dir/tests/test_builder.cpp.o
.../c++/12.0.0/bits/shared_ptr_base.h:397:45: error: 'size' may be used uninitialized [-Werror=maybe-uninitialized]
397 | explicit _Sp_ebo_helper(_Tp&& __tp) : _M_tp(std::move(__tp)) { }
| ^~~~~~~~~~~~~~~~~~~~~~
In file included from flatbuffers/tests/test_builder.cpp:1:
flatbuffers/tests/test_builder.h: In function 'void builder_move_assign_after_releaseraw_test(Builder) [with Builder = flatbuffers::FlatBufferBuilder]':
flatbuffers/tests/test_builder.h:63:10: note: 'size' was declared here
63 | size_t size, offset;
| ^~~~
...
In file included from flatbuffers/tests/test_builder.cpp:1:
flatbuffers/tests/test_builder.h: In function 'void builder_move_assign_after_releaseraw_test(Builder) [with Builder = GrpcLikeMessageBuilder]':
flatbuffers/tests/test_builder.h:63:10: note: 'size' was declared here
63 | size_t size, offset;
| ^~~~
cc1plus: all warnings being treated as errors
Here is the relevant bit of test:
template<class Builder>
void builder_move_assign_after_releaseraw_test(Builder b1) {
auto root_offset1 = populate1(b1);
b1.Finish(root_offset1);
size_t size, offset;
std::shared_ptr<uint8_t> raw(
b1.ReleaseRaw(size, offset), [size](uint8_t *ptr) {
flatbuffers::DefaultAllocator::dealloc(ptr, size);
});
Note how `b1.ReleaseRaw(size, offset)` is expected to populate `size`
and `[size](uint8_t *ptr) {` captures the result. But both are parameters
to the same function call and thus evaluation order is unspecified.
* chore: make flatc artifacts from CI executable
* chore: prepare dart 2.0.0 release
* refactor: update description in pubspec.yaml to make pub.dev happy
"The package description is too long.
Search engines display only the first part of the description. Try to keep the value of the description field in your package's pubspec.yaml file between 60 and 180 characters."
Ivan Dlugos