Speed up json_view's parser with SIMD scanning and a hash table
for large objects.
Long runs of string bytes are scanned 16 bytes at a time with NEON
(AArch64, GCC and Clang) and SSE2 (x86-64), both baseline
instruction sets. Keys keep 16 table checks before the vector
loop, because their lengths repeat from record to record; string
values get 8, because their lengths vary more. Non-ASCII text is
validated 16 bytes at a time with simdjson's "lookup4" check
(Keiser and Lemire, 2021), with NEON on AArch64 and, on x86-64,
with SSSE3. SSSE3 is not part of baseline x86-64, so the check is
compiled for SSSE3 with a function attribute and used only where
CPUID reports it, which all x86-64 CPUs since about 2011 do; the
answer is cached in a statically initialized atomic, so there is
no guard of a local static and no global constructor. The same
input is accepted either way. JSON_VIEW_NO_SIMD selects the
portable code.
On x86-64, string runs are now checked vector-first: one SSE2
compare from the first byte finds the end of most keys and short
values, instead of a branch per byte for the first 8-16 bytes.
AArch64 keeps the byte-wise steps, where a NEON mask costs more and
the branches predict well. Entering an object or array no longer
stalls: open() stores the parent's frame field by field instead of
building it on the stack and reading it back with wider loads,
which waited for the narrower stores to retire.
Objects with 128 members or more get an open-addressing hash table
built when the object closes, so operator[], at(), find(),
contains(), count(), value(), and JSON pointers take constant time
on average in such objects; of duplicate keys, the first is kept,
as for the linear search. The idea comes from Boost.JSON.
simdjson is credited in simd.hpp's SPDX block, the README, and
license.md.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
The check rejects inputs of 0xFFFFFFF0 bytes or more, but the exception
message and the documentation said 4 GiB. Name the limit once
(max_input_size), and state 4 GiB minus 16 bytes in the message and the
documentation. Test the limit with a container that only claims the size.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Add json_document and json_view, a read-only, zero-copy index of a
JSON text, as the first public slice of the zero-copy view (#5295).
A parse produces a flat array of 16-byte nodes in document order,
one per value and one per object key. Strings stay in the source
text; escaped strings are decoded into an arena. Integers are
converted while their digits are in the cache; floats keep only
their digit layout and are converted on read. Containers store the
size of their subtree, so a reader can step over one in constant
time. A document makes a handful of allocations, however many
values it has.
The parser accepts exactly what json::parse accepts, with every
combination of ignore_comments and ignore_trailing_commas, with and
without a trailing NUL, and under JSON_STRICT_NUL_HANDLING. It is
portable C++11 and does not depend on byte order.
basic_json_document adds parse, parse_copy, accept, read (reuses a
document's memory), root, is_discarded, source, owns_source,
node_count, memory_usage, and shrink_to_fit. basic_json_view adds
type, the is_* queries, operator bool, size, empty, materialize,
and source_offset. A parse error throws the same exception
basic_json::parse would throw for the same input, message and
position included.
detail::abi_config keeps JSON_STRICT_NUL_HANDLING and
JSON_USE_LEGACY_DISCARDED_VALUE_COMPARISON readable after json.hpp
undefines them, in the ABI namespace so they always match the
basic_json in use.
A NUL byte that ends a // comment is the end of the input, as in
parse() since #5696.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Pretty-printed JSON indents every line with a run of spaces, which
skip_whitespace() read one character at a time. For contiguous input
(bulk_scan) the lexer now skips the spaces after a newline 8 bytes at a
time directly in the input buffer, including the leading spaces of the
last, partial word. This bypasses get(), which is safe because
bulk_scan implies the lazy position from the previous commit: nothing
is counted per character.
Compared to lazy positions alone, pretty-printed input (twitter,
citm_catalog with indent 2 and 4) parses up to 13% faster and accept()
gets up to 30% faster (GCC 13 and clang 18 on x86-64, Apple clang 21 on
arm64). Minified input is unchanged within noise.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Derive the lexer position lazily for seekable input
The lexer counted the read position on every character: two increments
and a newline check per get(), all on members kept in memory. For
seekable input (pointers, strings, other random-access byte ranges) the
position can be derived instead: chars_read_total is the adapter's read
count (including reads at EOF) minus a pending unget, and line and
column are recomputed from the consumed input only when get_position()
is called, i.e. on parse errors. JSON_DIAGNOSTIC_POSITIONS uses the new
O(1) get_chars_read_total(). Stream, file, wide-string and user-defined
adapters keep the eager counters.
Parsing pretty-printed input gets 8-12% faster and accept() 12-27%
(canada, citm_catalog, twitter; clang 21 arm64 and GCC 13 x86-64).
Error messages, byte offsets, SAX positions and diagnostic positions are
unchanged (differential comparison against develop on 548k inputs).
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Move the std::vector<std::byte> binary type tests out of unit-regression3
The #4804/#4209 tests instantiated a second basic_json specialization
(with_binary_t<std::vector<std::byte>>), including its parser and CBOR
reader. With clang 18, the C++20 object of unit-regression3.cpp had
32656 COFF sections on develop; the lazy position helpers pushed it to
32975, past what the MinGW linker of the Windows clang jobs handles
("relocation truncated to fit: IMAGE_REL_AMD64_REL32 against .rdata").
unit-custom-binary-type.cpp already instantiates the same type, so the
tests move there; unit-regression3.cpp drops to 27616 sections.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Report the parse errors of the position test via SAX
The test caught json::parse_error, so it aborted in the
ci_test_noexceptions build, where JSON_NOEXCEPTION turns the throw into
std::abort(). sax_parse() hands the error position and the exception
object to parse_error() without throwing, so the test now records them
there and runs in every configuration.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
---------
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Clean up README, contribution guide, and repository metadata
- REUSE.toml: fix the Hedley path and SPDX id (CC0-1.0), mark the docset
icons as the public-domain JSON logo
- CITATION.cff: v3.12.0 was released on 2025-04-11
- FILES.md: list all workflows, describe the Meson option defaults
- README: ctest -LE, table of contents, typos, stale Android/MinGW advice,
moved links, merge duplicate Thanks entries, list the analysis tools
used in CI
- CONTRIBUTING: iterative parser, json_literals.hpp is generated, links
- .gitignore: backups and release outputs; .gitattributes: mark
generated files
- labeler: label other build systems and .github documentation
- MODULE.bazel: add the module version
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Clean up CI, CMake, Makefile, and Bazel files
- CMakeLists.txt: avoid VERSION_GREATER_EQUAL, which CMake < 3.7 lacks
- ci.cmake: test JSON_DisableTupleReferenceConversion in ci_cmake_flags,
remove unused variables and unreachable per-compiler targets, look up
Clang tools consistently, forward CMAKE_CXX_FLAGS to ci_module_cpp20,
format json_literals.hpp and remove backups in ci_test_amalgamation
- BUILD.bazel: add json_literals.hpp to the single-header target
- workflows: format json_literals.hpp before copying it, drop the obsolete
natvis --version plumbing, name natvis and macro_builder in failure
messages, drop the duplicate amalgamation job, install Valgrind only
where needed, republish docs on version bumps, fix stale names
- Makefile: complete .PHONY and help, check-amalgamation always restores
the checked-in files, natvis uses its own venv, macro_builder_check
installs astyle, clean removes the fuzzer binaries
- remove tools/amalgamate/config_json_view.json (json_view.hpp is not on
develop yet)
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Fix inaccuracies in the documentation
- version history: json_base_class_t (3.11.3), JSON_HAS_CPP_11 (3.10.0),
JSON_HAS_RANGES exclusions, define-type macros, ABI tags
- releases: 3.12.0 raised the minimum CMake version
- from_*: the (ptr, len) overloads are deleted, not removed, in 4.0.0
- contains/count/find: document the deleted integral overloads
- add JSON_HAS_RANGE_VIEW_CONVERSION and list the JSON_HAS_* macros in
the macro overview
- mention BON8 and error_handler where binary formats are listed
- broken links, outdated URLs, warning count, Hunter v0.26.12
- copy_markdown_source hook: expand snippets in the Markdown copies
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Pass /EHsc to the Windows C++20 module build
ci_module_cpp20 now forwards CMAKE_CXX_FLAGS to the module build. The
Windows workflow sets CMAKE_CXX_FLAGS, which replaces CMake's MSVC
defaults including /EHsc, so <chrono> failed with C4530 under /WX.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
---------
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Complete the architecture documentation page
Replace the placeholder bullets and TODOs with a description of the
component pipeline (with a diagram), the source layout, the template
parameters, the value storage (now in struct data), the input and
output adapters, and the SAX interface.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Link sources and basic_json, document full input adapter interface
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
* Align the default column of the template parameter table
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
---------
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Niels Lohmann