* Adding 'count_elements' method.
* Actually reporting errors.
* removing white space.
* Removing white space again.
* Adding an extra example.
* Prettier.
* Making the functionality more error-proof.
* Avoiding exceptions.
* Various fixes including extending count_elements to value types.
* Various fixes.
* Minor fixes.
* Correcting comment.
* Trimming white spaces.
* Add first working version of rapidjson_sax for partial tweets.
* Add cleaner and faster rapidjson_sax
* Add nlohmann_json_sax.
* Replace array of bool by bitsets.
* Replace strdup to copy string in rapidjson_sax.
* Change std::string_view assignment in rapidjson_sax.
* Add rapidjson_sax.h and fix typo in rapidjson.h
* Add nlohmann_json_sax.h and add user key check for screen_name in rapidjson_sax
* Change std::string_view assignement for text and screen_name.
* Add rapidjson_sax.h .
* Add nlohmann_json_sax.h . Fix typos distinct_user_id/nlohmann_json_sax.h, find_tweet/rapidjson.h and find_tweet/rapidjson_sax.h .
* Add extra check for id key when looking for find_id.
When find_field_unordered is used on an empty object, it calls
json_iterator::reenter_child. reenter_child asserts that it doesn't
rewind too far back by consulting parser->start_positions.
When the On Demand parser sees an empty object, it fails to update
parser->start_positions. This means that the assertion in
json_iterator::reenter_child reads stale data, or potentially
uninitialized memory. Reading uninitialized memory can cause spurious
assertion failures and Valgrind memcheck reports:
Running missing_keys_for_empty_top_level_object ...
==170679== Conditional jump or move depends on uninitialised value(s)
==170679== at 0x4943D7: reenter_child (json_iterator-inl.h:208)
==170679== by 0x4943D7: find_field_unordered_raw (value_iterator-inl.h:197)
==170679== by 0x4943D7: find_field_unordered (object-inl.h:13)
==170679== by 0x4943D7: find_field_unordered (object-inl.h:96)
==170679== by 0x4943D7: find_field_unordered (value-inl.h:110)
==170679== by 0x4943D7: find_field_unordered (document-inl.h:105)
==170679== by 0x4943D7: object_tests::missing_keys_for_empty_top_level_object() (ondemand_object_tests.cpp:117)
==170679== by 0x4CA761: object_tests::run() (ondemand_object_tests.cpp:1085)
==170679== by 0x8BA314: int test_main<bool ()>(int, char**, bool ( const&)()) (test_ondemand.h:81)
==170679== by 0x4CA9C8: main (ondemand_object_tests.cpp:1119)
==170679==
Fix the read of uninitialized or stale memory by updating
parser->start_positions regardless of whether we see an empty object or
an object with some keys.
This commit only affects builds where development checks
(SIMDJSON_DEVELOPMENT_CHECKS) are enabled. Builds where development
checks are disabled are unaffected by this bug.
* Verifies bug with missing keys.
* Allowing search from any key.
* Workaround for buggy msys
* Restricting how we can end key searches.
* Adding a few tests.
* Under ARM, it is slightly better to reverse the word once and then extract the bits.
* Guarding the zero_leading_bit call to avoid sanitizer warnings.
* This moves all DOM (benchmark + test) files to a subdir
* Missing file.
* CMake + DLL is not pretty.
* Capitalizing AND
* Fixing mismatch endif
* Flipping the order.
* onedemand => ondemand
* Remove CMP0025 policy
This policy is already set to NEW by the minimum required version.
* Use HOMEPAGE_URL in the project call
* Use VERSION in the project call
* Detect if this is the top project
* Port simdjson-user-cmakecache to a CMake script
* Create a developer mode
The SIMDJSON_DEVELOPER_MODE option set to ON will enable targets that
are only useful for developers of simdjson.
* Consolidate root CML commands into logical sections
* Warn about intended use of developer mode
* Prettify the just_ascii test
* Remove redundant CMake variables
* Inline CML contents from include and src
* Raise minimum CMake requirement to 3.14
* Define proper install rules
* Restore thread support variable
* Add BUILD_SHARED_LIBS as a top level only option
* Force developer mode to be on in CI
* Include flags earlier in developer mode
* Set CMAKE_BUILD_TYPE conditionally
CMAKE_BUILD_TYPE is used only by single configuration generators and is
otherwise completely ignored.
* Remove useless static/shared options
simdjson now uses the CMake builtin BUILD_SHARED_LIBS to switch the
built artifact's type.
* Remove unused CMAKE_MODULE_PATH variable
* Refactor implementation switching into a module
* Factor exception option out into a module
* Reformat simdjson-flags.cmake
* Rename simdjson-flags to developer-options
* Accumulate properties into an include module
This is done this way to avoid using utility targets that must be
exported and installed, which could potentially be misused by users of
the library.
* Port impl definitions to props
* Port exception options to props
* Lift normal options to the top
* Port developer options to props
* Remove simdjson-flags from benchmark
* Document the developer mode in HACKING
* Fix include path in installed config file
* Fix formatting of prop commands
* Fix tests that include .cpp files
* Change GCC AVX fixes back to compile options
* Deprecate SIMDJSON_BUILD_STATIC
* Always link fuzz targets to simdjson
* Install CMake from simdjson's debian repo
* Add gnupg for apt-key
* Make sure ASan link flags come first
* Pass CI env variable to cmake invocation
* Install package for apt-add-repository
* Remove return() from flush macro
* Use directory level commands instead of props
* Restore the github repository variable
* Set developer mode unconditionally for checkperf
The CI env variable is only set in the CI and this target is always run
in developer mode.
* Attempt to fix ODR violation in parsing checks
These tests were compiling the simdjson.cpp file again and linking to
the simdjson library target causes ODR violations.
Instead of linking to the target, just inherit its props.
* Move variables before the source dir
* Mark props to be flushed after adding more
* Use props for every command for the library
* Use keyword form for linking libs
* Handle deprecation of SIMDJSON_JUST_LIBRARY
* Handle deprecations in a separate module
Co-authored-by: friendlyanon <friendlyanon@users.noreply.github.com>
* Truncate final unclosed string.
* Adding more precise remarks.
* Better documentation and more robust code.
* ARM + PPC corrections.
* Patching ARM implementation with new stage1_mode parameter.
* Fixed most problems.
* Correcting white spaces and adding a remark.
* This adds the truncated_bytes() method to the stream instances.
Daniel Lemire
John Keiser
Nicolas Boyer
strager
Pavel Novikov
Ivan Volnov
Amos Bird
Luigi Pinca
PavelP
Dirk Stolle
bobergj
friendlyanon
Daniel Mangum
Paul Dreik