These classes are now constructed with an interface adapter. This moves
complexity from various places into the interface adapter class, or to
some factories which now implement the different flavors of input.
Furthermore, input adapters are kept in std::shared_ptr to avoid the
need of manual deletion.
We totally forgot to implement the comparison operators other than ==
and != for scalar types. Consequently, comparing a JSON value with a
scalar type led to compile errors.
- removed uncached input stream adapter; it was too slow anyway
- implemented a class binary_read which parses CBOR based on input
adapters
- in the CBOR parser, numbers are created via memcpy to avoid undefined
behavior
The solution with a std::runtime_error member is more elegant. It
allows to have std::exception as base class again. However, I still
have no idea why GCC thinks the copy constructor may throw...
To have nothrow-copy-constructible exceptions, we inherit from
std::runtime_error which can cope with arbitrary-length error messages.
Intermediate strings are built with static functions and then passed to
the actual constructor.
This commit removed the re2c lexer and replaced it by a manual version.
Its integration is not yet complete: number parsing does not respect
locales or overflows. Furthermore, parsing does not need to end with
EOF. Therefore, a lot of test cases fail. The idea is to push this
branch forward so we can conduct performance comparisons. So far, a
nice side effect are better diagnosis messages in case of parse errors.
- Removed unused headers.
- Added override where needed.
- Added description for parse_error.113 exception.
- Fixed some conversion warnings.
- Integrated cbor_expect_string function for CBOR maps.
- Added documentation on the supported CBOR/MessagePack features.
- Added test to check all initial bytes for CBOR input.
When <Windows.h> is included with MSVC, a macro NOMINMAX is defined
that yields compilation errors when max/min calls are encountered. This
can be fixed by un-defining NOMINMAX, or by placing parentheses around
all min/max calls. We chose the latter.
Since #329, NaN and inf numbers do not yield an exception, but are
stored internally and are dumped as “null”. This commit adjusts the
fuzz testers to deal with this special case.
- If an overflow occurs during parsing a number from a JSON text, an
exception (std::out_of_range for the moment, to be replaced by a
user-defined exception #244) is thrown so that the overflow is detected
early and roundtripping is guaranteed.
- NaN and INF floating-point values can be stored in a JSON value and
are not replaced by null. That is, the basic_json class behaves like
double in this regard (no exception occurs). However, NaN and INF are
serialized to “null”.
- Adjusted test cases appropriately.
To avoid the error described in #497, I added a function
msgpack_expect_string that is executed every time a string is expected
during the parsing of a map. In case the current byte does not belong
to a MsgPack string, an exception is thrown.
Added a test to check if the input stream is good() before executing
getline on it. Also added two test cases that set the failbit and
badbit before calling file_line_buffer.
I used __EXCEPTIONS to detect whether exceptions are supported.
Apparently, this is a macro that is only used by libstdc++
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64276). It’s much cleaner
to use __cpp_exceptions as it is in the standard since C++98.
Note that compiling the unit-tests with “-fno-exceptions” still does
not work, because Catch uses throw internally. However, the library’s
exceptions can be switched off by defining JSON_NOEXCEPTION.
Added class hierarchy for user-defined exceptions (#244). Integrated
parse exceptions 101-103. Parse exceptions include the byte count of
the last read character to locate the position of the error (#301).
The constructor basic_json(std::istream&, const parser_callback_t) has
been deprecated since version 2.0.0. This commit removes it together
with its code example, deprecation macro, and test cases. The code now
also compiles with -W-deprecated-declarations.
- Added comments for the serializer class.
- Added test case for resizing of the indentation string.
- Using std::none_of to check if “.0” needs to be added to
floating-point number.
A lot of small changes to avoid memory allocations:
- The locale is only queried once rather than with every number
serialization.
- The indentation string is recycled between different calls.
- The string escape function avoids a copy if no escaping is necessary.
- The string escape and the space function use a complete switch case
instead of cascaded ifs.
Cachegrind measures some 15% performance improvement.
Treated the size of the range as the number of thousand separators.
This logical error yielded a negative value for written_bytes and
eventually an infinite loop, as written_bytes was converted to an
unsigned value.
The class is currently just a wrapper for an std::ostream and collects
all functions related to serialization. The next step should be
recycling of variables to avoid repetitive initialization for each
recursive dump call.
numtostr now directly writes to a stream. Return value of snprintf is
reused to avoid finding end of string. Cachegrind suggests a 1%
performance increase.
All ‘<<‘ calls have been replaced by write()/put() calls. The
indentation strings needs not to be resized. Cachegrind measures 1%
performance improvement.
Indentation string is recycled to avoid allocations. Comma-separation
in objects does not need an if any more. Cachegrind measures 1%
performance improvement.
Added separate code paths for normal output and pritty-printed output.
This allowed to remove most of the ifs along the way. Benchmarks and
cachegrind suggest a 10% performance improvement.
