Skip to content

Latest commit

 

History

History
592 lines (548 loc) · 31.3 KB

CHANGELOG.md

File metadata and controls

592 lines (548 loc) · 31.3 KB
Raw

Change Log

[0.6.2-pre]

  • CMake: avoid assuming location of build dir during configuration.
  • Use untyped integer constants in place of enums for public interface flags to allow for safe bit masking operations (PR #248).
  • Added experimental support for generating compile_commands.json via CMakeList.txt for use with clangd.
  • Remove fallthrough macro for improved portability (#247, #252).
  • Added parse_float/double_compare, parse_float/double_is_equal to portable library, and added parse_float/double_isnan to mirror isinf. This should help with GCC 32-bit double precision conversion issue.
  • Add Github Actions builds to replace stale Travis CI build. This also includes source code fixes for some build variants. Although Windows build is included it only covers recent 64-bit Windows. More work is need for older Windows variants. (#250).
  • Increase maximum allowed schema file size from 64 KiB to 1 MB (#256).
  • Fix seg fault in json parser while adding null characters to a too short input string for a fixed length char array struct field (#257).
  • Fix regression where empty namespace in schema does not reset root scope correctly in parser (#265).
  • Fix lexer checks that breaks with UTF-8, notably UTF-8 schema comments (#267).
  • Add sanitizer flag for clang debug and related warnings (input from several PRs incl. #237).
  • Fix missing runtime check for building too large tables (#235).
  • Fix alignment of large objects created outside root buffer (#127).
  • Pad top level buffer end to largest object in buffer
  • Add _with_size verifiers to enable verification of size prefixed buffers that are aligned above 4 bytes (sizeof(uoffset_t)) (#210, PR #213). Also fixes a (uoffset_t) cast that would truncate very large verifier bufsiz arguments above 4GB before the verifier checked for argument overflow (not expected to be a practical problem, but could be abused).
  • Fix __[portable_aligned_alloc, the fallback implementation of aligned_alloc so returns null proper and not a small offset on alloc failure. Most platforms do not depend on this function (#269). Also fix equivalent FLATCC_ALIGNED_ALLOC that does not necesserily use malloc.
  • Fix flatcc compiler error message when schema has a union as first table field with explicit id attribute 1. Explict id must leave space for the hidden type field, but id 1 is valid since id 0 is valid for the type field id. (#271).
  • Add C23 support for "pstdbool.h" header.
  • Fix strict aliasing UB. In the vast majority of cases *(T *)p style memory access is supported by compilers, but some both enable strict aliasing by default while being very aggressive with strict aliasing. This can lead to reading uninitialized memory and/or segfaults in the monster_test test case, but is less likely problematic in dynamically allocated memory buffers. For performance reasons this fix is not universally applied when deemed unnecessary. Added README section "Strict Aliasing". (#274).
  • Fix flatcc cli flags for --schema --outfile. (#216).
  • Replace cmetrohash and murmur hash with xxhash. (#238).

[0.6.1]

  • Add flatcc_builder_alloc and flatcc_builder_free to handle situations where standard allocation has been redefined via macros so free is no longer safe to use. These are similar to the existing aligned_alloc/free functions.
  • Fix a potential, but never seen, low level race condition in the builder when writing a union field because the builder might reallocate between type field and value field. Affects flacc_common_builder.h but not builder.c.
  • Fix GCC 8.3 reversal on __alignas_is_defined for -std=c++11, affecting pstdalign.h (#130).
  • Make C++ test optional via build flag and disable for pre GCC-4.7 (#134).
  • Avoid conflicts with table or struct fields named identifier (#99, #135) by not generating <table-name>_identifier for the table (or structs) file identifier, instead <table-name>_file_identifier is generated and the old form is generated for backwards compatibility when there is no conflict.
  • DEPRECATED: <table-name>_identifier should now be <table-name>_file_identifer.
  • FLATCC_ASSERT and FLATCC_NO_ASSERT provided for custom runtime assert handling (#139).
  • Add guards in case bswap macroes are already defined and use POSIX compliant Linux detection (#129).
  • Fix precision loss of floating point default values in generated code (#140).
  • Fix anon union pedantic warning in internal builder struct (PR #143).
  • Remove warnings triggered by -Wsign-conversion.
  • BREAKING: some functions taking a string or buffer argument, and a separate size argument, have changed the size type from int to size_t in an effort to standardize and to remove sign conversion warnings.
  • Fixed size arrays renamed to fixed length arrays for consistency with Googles project.
  • Support aligned_alloc for MingW (#155).
  • Disable config flag FLATCC_ASCENDING_ENUM now that Googles flatc compiler also has support for reordered enums.
  • Fix cli options so common files can be generated without schema files (PR #156).
  • Make build.sh work for non-bash shells (PR #159).
  • Fix parsing json struct as roort (#157).
  • Add support for optional scalar values (#162).
  • Fix enum out of range (#176).
  • Add stdalign support for TCC compiler (#174).
  • Add RPC data to bfbs binary schema (#181).
  • Fix support for printing JSON enum vectors (#182).
  • Silence GCC 11 identation warning (#183).
  • Fix type of code field on json test (#184).
  • Add end_loc to parser state of root json parsers (#186).
  • BREAKING: Add per table <name>_file_extension and fixed wrong per table <name>_file_identifier assignment when multiple files are included due to macro expansion conflict. Extensions are now specified without extra dot prefix unless specified explicitly in the schema file. The default extension is now 'bin' instead of '.bin' (#187).
  • Fix buffer overrun when parser reports error on large symbols (#188).
  • BREAKING: Print --version to stdout, not stderr.
  • Fix schema parser returning on success on some failure modes (#193).
  • Fix larger integer literal types in JSON parser and printer (#194).
  • Add pattributes.h to portable library and replace GCC fallthrough comments with fallthough attribute to also silence clang warnings (#203).
  • Remove misguided include guards from portable/pdiagnostic_push/pop.h and fix related and expected warnings in other code. NOTE: End user code might be affected because warnigs were disabled more broadly than intended. Also note that warnings will still be disabled after pop if the compiler does not support push/pop diagnostics (#205).
  • Fix verifier crash on malicious string length input (#221).
  • Fix potential crash parsing unterminated JSON (#223).
  • Allow 0 (and other unknown values) as schema default value for enums with bit_flags attribute.
  • Disable -pedantic flag for GCC >= 8, it just keeps breaking perfectly valid code (#227).

[0.6.0]

  • BREAKING: if there are multiple table fields with a key attribute, the default key is now the field with the smaller id rather than the first listed in case these differ (which is not often the case).
  • The attribute primary_key has been added to choose a specific keyed field as the default key field for finding and sorting instead of the field with the smallest id.
  • Add mutable types: mutable_table_t, mutable_union_t, mutable_union_vec_t and casts: mutable_union_cast, mutable_union_vec_cast.
  • Disallow key attribute on deprecated fields.
  • Add 'sorted' attribute for scalar, string, table and struct vectors. Tables and structs must have a key field defined.
  • Add recursive table and union _sort operation in _reader.h for types that contain a sorted vector, either directly or indirectly. NOTE: shared vectors in a DAG will be sorted multiple times.
  • Allow attributes to be declared multiple times, including known attributes.
  • Allow table and struct field names to be reserved names such as 'namespace' or 'table'. This can be disabled in config/config.h. Separately do the same for enum member. This change was motivated by JSON fields that can have names such as "table" or "namespace".
  • Use FLATCC_INSTALL_LIB for install targets in addition to ordinary builds (#109).
  • Add missing case break in flatcc compiler - no impact on behavior (#110).
  • Make flatbuffers_not_found and flatbuffers_end constant values both because it is more correct, and to silence warnings on some systems.
  • Fix flatcc endless loop with our of order field id's (#112).
  • Add support for fixed length arrays as struct member fields, including fixed length char arrays. NOTE: bfbs schema uses byte arrays instead of char arrays since Googles flatc tool does not have char arrays.
  • Fix aligned_free when used with FLATCC_USE_GENERIC_ALIGNED_ALLOC (#118).
  • Fix potential buffer overrun when parsing JSON containing surrogate pairs with a resulting UTF-8 code point of length 4 (bug introduced in 0.5.3).
  • BREAKING: empty structs are no longer supported. They were fully supported and Googles flatc compiler also no longer support them.
  • Fix incorrect sprintf arg when printing NaN with the grisu3 algorithm (#119).
  • Fix gcc 9 regression on -fallthrough comment syntax (#120).
  • Silence false positive gcc 9 warning on strncpy truncation (-Wstringop-truncation) (#120).
  • Silence false positive gcc 9 warning on printf null string (-Wno-format-overflow) (#120).
  • Silence GCC -Wstrict-prototypes warnings that incorrectly complain about function definitions with empty instead of void arguments. Only for runtime source (#122).

[0.5.3]

  • BREAKING: 0.5.3 changes behavour of builder create calls so arguments are always ordered by field id when id attributes are being used, for example MyGame_Example_Monster_create() in monster_test_fbs (#81).
  • Fix refmap_clear so that the refmap is always ready for reuse.
  • Remove C++ from CMake toolchain when not building tests (#94).
  • Add missing _get suffix in flatbuffers_common_reader.h needed when used with flatcc -g option (#96).
  • Remove stray break statement that prevented generation of search methods for scalar table fields with the key attribute set (#97).
  • Fix return value when creating a struct field fails which is unlikely to happen, and hence has low impact (#98).
  • Fix file identifier string to type hash cast as implicit cast was not safe on all platforms.
  • Fix reallocation bug when cloning vectors of non-scalar content (#102).
  • JSON parser: combine valid UTF-16 surrogate pairs such as "\uD834\uDD1E" (U+1D11E) into single 4 byte UTF-8 sequence as per spec. Unmatched half pairs are intentionally decoded into invalid UTF-8 as before.
  • Fix sorting tables by scalar keys. Sorting by integer key could lead to undefined behavior (#104).
  • Add FLATCC_INSTALL_LIB configuration to CMake to change the default /lib path, for example cmake -DFLATCC_INSTALL_LIB=lib64.
  • Fix return code in flatcc_parse_file library function.

[0.5.2]

  • Handle union vectors in binary schema generation (.bfbs).
  • Handle mixed union types in binary schema (.bfbs).
  • Fix .bfbs bug failing to export fields of string type correctly.
  • Fix how vectors are printed in samples/reflection project.
  • Add support for KeyValue attributes in binary schema (.bfbs).
  • Added __tmp suffix to macro variables in flatbuffers_common_reader.h to avoid potential name conflicts (#82).
  • Added _get suffix to all table and struct read accessors in addition to existing accesors (Monster_name() vs Monster_name_get() (#82).
  • Added -g option flatcc commandline to only generate read accessors with the _get suffix in order to avoid potential name conficts (#82).
  • Fix stdalign.h not available in MSVC C++ in any known version.
  • Added test case for building flatcc project with C++ compiler (#79, #80).
  • Fix flatbuffers_int8_vec_t type which was incorrrectly unsigned.
  • Added table, union, vector clone, and union vector operations. Table fields now also have a _pick method taking a source table of same type as argument which is roughly a combined get, clone and add operation for a single field. _pick will pick a field even if it is a default value and it will succedd as a no-operation if the source is absent. _clone discards deprecated fields. Unknown union types are also discarded along with unions of type NONE, even if present in source. Warning: _clone will expand DAGs.
  • Added _get_ptr reader method on scalar struct and table member fields which returns the equivalent of a single field struct _get method. This is used to clone scalar values without endian conversion. NOTE: scalars do not have assertions on required attribute, so be careful with null return values. For structs it is valid to apply _get_ptr to a null container struct such that navigation can be done without additional checks.
  • Added _push_clone synonym for scalar and struct vector _push_copy.
  • Add include/flatcc/flatcc_refmap_h and src/runtime/refmap.c to runtime library. The runtime library now depends of refmap.c but has very low overhead when not expeclity enabled for use with cloning.
  • Add flatbuffers_type_hash_from_string to avoid gcc-8 strncpy warning (#86).
  • Add long form flatcc options --common, --common-reader, --common-builder, --reader, --builder, --verifier.
  • Remove flatcc deprecated option for schema namespace.
  • Add scripts/flatcc-doc.sh to help document generated code.
  • Remove unnecessary memory.h includes (#92)

[0.5.1]

  • Fix parent namespace lookup in the schema parser when the namespace prefix is omitted.
  • Fix buffer overrun in JSON printer when exhausting flush buffer (#70).
  • More consistent name generation across verifier and json parsers allowing for namespace wrapped parse/verify/print table functions.
  • Fix unhelpful error on precision loss from float/double conversion in schema and JSON parser.
  • Align monster_test.fbs Monster table more closely with Googles flatc version after they diverged a bit. (Subtables may differ).
  • Some C++ compatiblity fixes so include/{support/flatcc} headers can be included into C++ without extern "C".
  • Fix missing null ptr check in fall-back aligned_free.
  • Enable posix_memalign as a default build option on recent GNU systems because -std=c11 blocks automatic detection. This avoids using a less efficient fallback implementation.
  • Add portable/include wrappers so build systems can add include paths to ensure that <stdint.h>, <stdbool.h> etc. is available. Flatcc does not currently rely on these.
  • Replace flatcc/portable/pdiagnostic_push/pop.h includes in generated code with flatcc/flatcc_pro/epilogue.h and add __cplusplus extern "C" guards in those. This removes explicit references to the portable headers in generated code and improves C++ compatibility (#72).
  • Change inconsistent const void * to const char * in JSON buffer argument to generated _as_root parsers (#73).
  • Simplify use of json printers by auto-flushing and terminating buffers when a root object has been printed (#74).
  • BREAKING: in extension of the changes in 0.5.0 for unions and union vectors, the low-level methods and structs now consistently use the terminology { type, value } for union types and values rather than { type, member } or { types, members }. The binary builder interface remains unchanged.
  • Silence (unjustified) uninitialized gcc warnings (#75).
  • Fix C++14 missing __alignas_is_defined.
  • Remove newlib stdalign conflict (#77).
  • Add flatcc_json_printer_total.
  • Add flatcc_builder_table_add_union_vector.

[0.5.0]

  • New schema type aliases: int8, uint8, int16, uint16, int32, uint32, int64, uint64, float32, float64.
  • Low-level: access multiple user frames in builder via handles.
  • Support for _is_known_type and _is_known_value on union and enum types.
  • More casts for C++ compatiblity (#59).
  • Fix regressions in verifier fix in 0.4.3 that might report out of bounds in rare cases (#60).
  • Silence gcc 7.x warnings about implicit fallthrough (#61).
  • Fix rare special in JSON parser causing spurious unknown symbol.
  • Reading and writing union vectors. The C++ interface also supports these types, but other languages likely won't for a while.
  • New _union(t) method for accessing a union fields type and member table in a single call. The method also supports union vectors to retrieve the type vector and member vector as a single object.
  • BREAKING: In generated builder code for union references of the form <union-name>_union_ref_t the union members and the hidden _member field has been replaced with a single member field. Union constructors work as before: Any_union_ref_t uref = Any_as_weapon(weapon_ref) Otherwise use .type and .member fields directly. This change was necessary to support the builder API's new union vectors without hitting strict aliasing rules, for example as argument to flatcc_builder_union_vector_push. Expected impact: low or none. The physical struct layout remains unchanged.
  • BREAKING: flatbuffers_generic_table_[vec_]t has been renamed to flatbuffers_generic_[vec_]t.
  • BREAKING: The verifiers runtime library interface has changed argument order from align, size to size, align in order to be consistent with the builders interface so generated code must match library version. No impact on user code calling generated verifier functions.
  • BREAKING: generated json table parser now calls table_end and returns the reference in a new pref argument. Generated json struct parsers now renamed with an _inline suffix and the orignal name now parses a non-inline struct similar to the table parsers. No impact to user code that only calls the generated root parser.
  • Fix off-by-one indexing in flatbuffers_generic_vec_at. Impact low since it was hardly relevant before union vectors were introduced in this release.
  • Add document on security considerations (#63).
  • Add support for base64 and base64url attributes in JSON printing and parsing of [ubyte] table fields.
  • Added flatcc_builder_aligned_free and flatcc_builder_aligned_alloc to ensure aligned_free implementation matches allocation compiled into the runtime library. Note that alignment and size arguments are ordered opposite to most runtime library calls for consistency with the C11 aligned_alloc prototype.
  • Support for struct and string types in unions.
  • Add missing _create method on table union member fields.
  • Add _clone and _clone_as_[typed_]root[_with_size] methods on structs. _clone was already supported on structs inlined in table fields.
  • Fix harmless but space consuming overalignment of union types.
  • Add flatbuffers_union_type_t with flatbuffers_union_type_vec operations.
  • Fix scoping bug on union types in JSON parser: symbolic names of the form MyUnion.MyUnionMember were not accepted on a union type field but MyNamespace.MyUnion.MyMember and MyMember was supported. This has been fixed so all forms are valid. Plain enums did not have this issue.
  • Place type identifiers early in generated _reader.h file to avoid circular reference issue with nested buffers when nested buffer type is placed after referencing table in schema.
  • Fix verify bug on struct buffers - and in test case - not affecting ordinary buffers with table as root.

[0.4.3]

  • Fix issue with initbuild.sh for custom builds (#43)
  • Add casts to aid clean C++ builds (#47)
  • Add missing const specifier in generated buffer_start methods - removes C++ warnings (#48)
  • Update external/hash, removed buggy Sorted Robin Hood Hash that wasn't faster anyway - no impact on flatcc.
  • Fix JSON parsing bug where some names are prefixes of others (#50).
  • A Table of Contents in documentation :-)
  • Move repetitive generated JSON string parsing into library.
  • Add tests for JSON runtime compiled with different flags such as permitting unquoted keys.
  • Fix building nested buffers when the parent buffer has not yet emitted any data (#51).
  • Fix building nested buffers using the _nest() call (#52).
  • Add FLATCC_TRACE_VERIFY as build option.
  • Allow more costumization of allocation functions (#55).
  • Avoid dependency on PORTABLE_H include guard which is too common (#55).
  • (possibly breaking) Fix duplicate field check in flatcc_builder_table_add call.
  • Fix incorrect infinity result in grisu3 parser and double to float overflow handling in parse_float in portable library (affects JSON of abnormal numeric values).
  • Fix return value handling of parse_float, parse_double in JSON parser.
  • Fix verifier vector alignment check - affects vectors with element size 8+.
  • Fix missing static in generated enum and union functions in JSON printer (#57).

[0.4.2]

  • Fix SIGNIFICANT bug miscalculating the number of builder frames in use. Nesting 8 levels would cause memory corruption (#41).
  • Fix minor memory leak in flatcc compiler.
  • Reduce collisions in builders vtable hash.
  • Remove broken dependency on <mm_malloc.h> for some GCC versions in paligned_alloc.h (#40).
  • Allow C++ files to include pstdalign.h and paligned_alloc.h (#39).

[0.4.1]

  • Test for posix_memalign on GCC platforms and fix fallback aligned_alloc.
  • Fix JSON parser handling of empty objects and tables.
  • Fix JSON parser - some fields would not be accepted as valid (#17).
  • Fix rare uncompilable doc comment in schema (#21).
  • Avoid crash on certain table parser error cases (#30).
  • Add support for scan similar to find in reader API, but for O(N) unsorted search, or search by a secondary key, and in sub-ranges.
  • Optionally, and by default, allow scan by any field (#29), not just keys.
  • More compact code generation for reader (hiding scan methods).
  • Use __flatbuffers_utype_t for union type in reader instead of uint8_t.
  • Add unaligned write to punaligned for completeness.
  • Promote use of flatcc_builder_finalize_aligned_buffer in doc and samples over flatcc_builder_finalize_buffer.
  • Add scope counter to pstatic_assert.h to avoid line number conflicts.
  • Fix compiler error/warning for negative enums in generated JSON parser (#35).
  • Fix potential compiler error/warnings for large enum/defaults in generated reader/builder (#35).
  • Fix tab character in C++ style comments (#34)
  • Fix incorrect api usage in binary schema builder (#32)
  • Support hex constants in fbs schema (flatc also supports these now) (#33).

[0.4.0]

  • Fix Windows detection in flatcc/support/elapsed.h used by benchmark.

  • Fix #8 surplus integer literal suffix in portable byteswap fallback.

  • Fix pstatic_assert.h missing fallback case.

  • Fix #9 return values from allocation can be zero without being an error.

  • Fix #11 by avoiding dependency on -lm (libmath) by providing a cleaner over/underflow function in include/flatcc/portable/pparsefp.h.

  • Fix #12 infinite loop during flatbuffer build operations caused by rare vtable dedupe hash table collision chains.

  • Added include/flatcc/support/cdump.h tool for encoding buffers in C.

  • JSON code generators no longer use non-portable PRIszu print modifiers. Fixes issue on IBM XLC AIX.

  • Deprecated support for PRIsz? print modifiers in include/flatcc/portable/pinttypes.h, they would require much more work to be portable.

  • Fix and clean up __STDC__ version checks in portable library.

  • Improve IBM XLC support in pstdalign.h.

  • Always include pstdalign.h in flatcc_flatbuffers.h because some C11 compilers fail to provide stdalign.h.

  • Buffer verifier used to mostly, but not always, verify buffer alignment relative to buffer start. With size prefixed buffers it is necessary to verify relative to the allocated buffer, which is also safer and more consistent, but adds requirements to aligned allocation.

  • monster_test and flatc_compat test now uses aligned alloc.

  • Add aligned_alloc and aligned_free to pstdalign.h.

  • flatcc_builder_finalize_aligned_buffer now requires aligned_free to be fully portable and no longer use unaligned malloc as fallback, but still works with free on most platforms (not Windows).

  • BREAKING: Size prefixed buffers added requires a minor change to the low-level flatcc builder library with a flag argument to create and start buffer calls. This should not affect user code.

Changes related to big endian support which do not affect little endian platforms with little endian wire format.

  • Support for big endian platforms, tested on IBM AIX Power PC.
  • Support for big endian encoded flatbuffers on both little and big endian host platforms via FLATBUFFERS_PROTOCOL_IS_LE/BE in include/flatcc/flatcc_types.h. Use flatbuffers_is_native_pe() to see if the host native endian format matches the buffer protocol. NOTE: file identifier at buffer offset 4 is always byteswapped.

In more detail:

  • Fix vtable conversion to protocol endian format. This keeps cached vtables entirely in native format and reduces hash collisions and only converts when emitting the vtable to a buffer location.
  • Fix structs created with parameter list resulting in double endian conversion back to native.
  • Fix string swap used in sort due to endian sensitive diff math.
  • Disable direct vector access test case when running on non-native endian platform.
  • Update JSON printer test to handle FLATBUFFERS_PROTOCOL_IS_BE.
  • Fix emit test case. Incorrect assumption on acceptable null pointer breaks with null pointer conversion. Also add binary check when FLATBUFFERS_PROTOCOL_IS_BE.
  • Add binary test case to json_test when FLATBUFFERS_PROTOCOL_IS_BE.
  • Fix endian sensitive voffset access in json printer.
  • Update flatc_compat to reverse acceptance of 'golden' little endian reference buffer when FLATBUFFERS_PROTOCOL_IS_BE.

[0.3.5a]

  • Fix regression introduced in 0.3.5 that caused double memory free on input file buffer. See issue #7.

[0.3.5]

  • Allow flatcc cli options anywhere in the argument list.
  • Add --outfile option similar to --stdout, but to a file.
  • Add --depfile and --deptarget options for build dependencies.
  • Allow some test cases to accept arguments to avoid hardcoded paths.
  • Deprecate --schema-namespace=no option to disable namespace prefixes in binary schema as Google flatbuffers now also includes namespaces according to google/flatbuffers#4025

[0.3.4]

  • Add FLATCC_RTONLY and FLATCC_INSTALL build options.
  • Fix issue4: when building a buffer and the first thing created is an empty table, the builder wrongly assumed allocation failure. Affects runtime library.
  • scripts/setup.sh now also links to debug libraries useful for bug reporting.
  • Add ULL suffix to large printed constants in generated code which would otherwise required --std=c99 to silence warnings.

[0.3.3]

  • BREAKING: verify_as_root no longer takes an identifier argument, use verify_as_root_with_identifier. myschema_verifier.h now includes myschema_reader.h to access identifier. identifer argument, and variants for type identifiers;.
  • Added scripts/setup.sh to quickly get started on small user projects.
  • Support namespace ; for reverting to global namespace in schema.
  • Enable block comments now that they are supported in flatc.
  • Parse and validate new rpc_service schema syntax, but with no support for code generation.
  • Add type hash support (create/verify_as_typed_root etc.) to optionally store and verify file identifiers based on hashed fully qualified type names.
  • Fix potential issue with detection of valid file identifiers in buffer.
  • Moved include/support into include/flatcc/support, renamed include/support/readfile.h function read_file to readfile.
  • Make FLATCC_TEST build option skip building samples and test files, as opposed to just skip running the tests.
  • vec_at, vec_find, etc. now use index type size_t instead of flatbuffers_uoffset_t.
  • Removed size_t conversion warnings on Win64.

[0.3.2]

  • Move compiler warning handling from generated headers to portable library.
  • Clean up warnings and errors for older gcc, clang and MSVC compilers.
  • CI builds.
  • Fix and improve portable version of static_assert.
  • Add integer parsing to portable library.

[0.3.1]

  • Add support for MSVC on Windows.
  • Allow FlatBuffer enums to be used in switch statements without warnings.
  • Remove warnings for 32-bit builds.
  • Fix runtime detection of endianness and add support for more platforms with compile time detection of endianness.
  • Fix scope bug where global namespace symbols from included schema would be invisible in parent schema.
  • Add missing static for generated union verifiers.
  • Fix bug in json printer unicode escape and hash table bug in compiler.
  • Fix json parser under allocation bug.
  • Fix emit_test early dealloc bug.

[0.3.0]

  • Rename examples folder to samples folder.
  • Add samples/monster example.
  • BREAKING: added missing _vec infix on some operations related to building vectors. For example Weapon_push -> Weapon_vec_push.
  • BREAKING: vector and string start functions no longer takes a count/len argument as it proved tedious and not very useful. The return value is now 0 on success rather than a buffer pointer. Use _extend call after start when the length argument is non-zero.

[0.2.1]

  • Disallow unquoted symbolic list in JSON parser by default for Google flatc compatibility.
  • Remove PRIVATE flags from CMake build files to support older CMake versions.
  • Simplify switching between ninja and make build tools.
  • Fix incorrectly named unaligned read macros - impacts non-x86 targets.
  • Mirror grisu3 headers in portable library to avoid dependency on external/grisu3.
  • Use correct grisu3 header for parsing, improving json parsing times.
  • Move include/portable to include/flatcc/portable to simplify runtime distribution and to prevent potential name and versioning conflicts.
  • Fix is_union in bfbs2json.c example.

[0.2.0]

  • BREAKING: flatcc verify functions now return a distinct error code. This breaks existing code. Before non-zero was success, now flatcc_verify_ok == 0. The error code can be converted to a string using flatcc_verify_error_string(ret).
  • BREAKING, minor: Remove user state from builder interface - now providing a user stack instead.
  • Fix verification of nested flatbuffers.
  • Fix certain header fields that was not endian encoded in builder.
  • MAJOR: Generate json printer and parser.
  • Added high performance integer printinger to portable library and fast floating point priting to runtime library (grisu3) for JSON.
  • Comparison agains default value now prints float to generated source with full precision ("%.17g").
  • MAJOR: read-only generated files no longer attempt to be independent of files in the flatcc include dir. Instead they will use one well-defined source of information for flatbuffer types and endian detection.
  • Always depend portable/pendian.h and portable/pendian_detect.h. (The include/portable dir can be copied under include/flatcc if so desired).
  • Updates to set of include files in include/flatcc.
  • Upgrade to pstdint.h 0.1.15 to fix 64-bit printf issue on OS-X.
  • Support for portable unaligned reads.
  • Hide symbols that leak into namespace via parent include.
  • Suppress unused function and variable warnings for GCC (in addition to clang).

[0.1.1]

  • Rename libflatccbuilder.a to libflatccrt.a (flatcc runtime).
  • Add suffix to all generated files (monster_test.h -> monster_test_reader.h)
  • Add buffer verifiers (`monster_test_verifier.h).
  • Assert on error in flatcc builder by default.
  • Fix -I include path regression in flatcc command.

[0.1.0]

  • Initial public release.