Otherwise figuring out what's wrong with your JSON can be tricky,
especially if you're using a single fmt string to validate a large,
complicated schema.
The comma delimiting will make separating keys that contain commas
difficult. For example:
{"foo, bar": true, "baz": false}
will generate errors like:
2 object item(s) left unpacked: foo, bar, baz
but that seems like a small enough corner case to not be worth much
worrying.
I wanted to find a way to handle this without have_unrecognized_keys,
but the strbuffer tooling makes it look like I shouldn't be reaching
in to do things like:
strbuffer_t unrecognized_keys;
unrecognized_keys.value = NULL;
and then using 'unrecognized_keys.value == NULL' in place of
have_unrecognized_keys.
This is particularly useful in modular situations where the allocation
functions are either unknown or private. For instance, in such cases,
the caller of json_dumps() has no way to free the returned buffer.
This is both good practice and nice for OpenBSD users, who will no
longer get the nag message to not use sprintf/strcpy every time they
link against jansson. It's worth noting that the existing code seems
safe to me - bounds checks were already happening before the actual
calls - and that this is for extra security.
This has the consequence that numbers are never converted to integers
when JSON_DECODE_INT_AS_REAL is set, and thus it works correctly all
integers that are representable as double.
Fixes#212.
The previous commit introduced a loop on all input keys to check the
strict mode. We can avoid this if we don't expect an optional key. In
this case, we fallback to the previous method to compare the length of
the set of expected keys and the length of the parsed keys.
On unpack, one may want to mix `JSON_STRICT` and optional keys by using
a format like `{s:i,s?o!}`. Unfortunately, this fails the stric test
with `-1 object item(s) left unpacked` error when the second key is not
specified.
To fix that, we iter on each key and we check if we have successfully
unpacked them. This is less efficient than the previous method but it
brings correctness.
First, wrap "advapi32.dll" into the TEXT() macro. If UNICODE is defined,
GetModuleHandle() redirects to GetModuleHandleW(), which excepts a wchar_t*
UTF-16 parameter, thus causing a compile error. TEXT() prefixes the string
literal with L in this case, and does nothing otherwise.
Second, make sure that CryptGenRandom() is actually called through the function
pointer retrieved by the call to GetProcAddress() above.
And third, replace _getpid() with the equivalent and more ubiquitous Win32 API
function GetCurrentProcessId(). Since _getpid() is not exported by all C
runtimes on Windows (most notably the Driver Development Kit), using it might
introduce previously unneeded runtime dependencies. GetCurrentProcessId(), on
the other hand, has been available in kernel32.dll since at least Windows 95,
just like the other API functions used in this code (GetModuleHandle() and
GetProcAddress()).
This is because it's really easy to get a name collission if compiling
Jansson as a subproject in a larger CMake project. If one project includes
several subprojects each having their own config.h, this will cause the
wrong file to be loaded.
